first commit

vllm/model_executor/layers/mamba/abstract.py

@@ -0,0 +1,42 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+from abc import abstractmethod
+from collections.abc import Iterable
+from typing import TYPE_CHECKING
+
+import torch
+
+from vllm.model_executor.layers.attention_layer_base import AttentionLayerBase
+
+if TYPE_CHECKING:
+    from vllm.attention.backends.abstract import AttentionBackend
+
+
+class MambaBase(AttentionLayerBase):
+    """
+    Base class for Mamba-like layers which support the v1 engine.
+    Inherit from this class if you implement a custom layer.
+    """
+
+    # Contains the KV cache (mamba state) for the layer
+    # in the shape specified by `self.get_state_shape`.
+    kv_cache: tuple[torch.Tensor, ...]
+
+    @abstractmethod
+    def get_state_shape(self) -> Iterable[tuple[int, ...]]:
+        """
+        Defines the shape of the state.
+        For mamba layers this is usually a (conv_state, ssm_state) tuple.
+        In this case, returns (conv_state_shape, ssm_state_shape).
+        """
+        pass
+
+    @property
+    @abstractmethod
+    def mamba_type(self) -> str:
+        pass
+
+    @abstractmethod
+    def get_attn_backend(self) -> type["AttentionBackend"]:
+        """Get the attention backend class for this Mamba layer."""
+        pass

vllm/model_executor/layers/mamba/linear_attn.py

@@ -0,0 +1,403 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+import math
+from typing import TYPE_CHECKING, Optional, Union
+
+if TYPE_CHECKING:
+    from vllm.attention.backends.abstract import AttentionBackend
+
+from typing import TYPE_CHECKING
+
+import torch
+import torch.distributed
+import torch.nn.functional as F
+from einops import rearrange
+from torch import nn
+
+from vllm.attention import AttentionMetadata
+from vllm.config import CacheConfig, ModelConfig, get_current_vllm_config
+from vllm.distributed.communication_op import tensor_model_parallel_all_reduce
+from vllm.distributed.parallel_state import (
+    get_tensor_model_parallel_rank, get_tensor_model_parallel_world_size)
+from vllm.forward_context import ForwardContext, get_forward_context
+from vllm.model_executor.custom_op import CustomOp
+from vllm.model_executor.layers.lightning_attn import (
+    lightning_attention, linear_decode_forward_triton)
+from vllm.model_executor.layers.linear import (ColumnParallelLinear,
+                                               RowParallelLinear)
+from vllm.model_executor.layers.mamba.abstract import MambaBase
+from vllm.model_executor.layers.mamba.mamba_utils import (
+    MambaStateDtypeCalculator, MambaStateShapeCalculator)
+from vllm.model_executor.layers.quantization.base_config import (
+    QuantizationConfig)
+from vllm.utils import direct_register_custom_op
+from vllm.v1.attention.backends.linear_attn import LinearAttentionMetadata
+
+if TYPE_CHECKING:
+    from vllm.attention.backends.abstract import AttentionBackend
+
+import torch
+import torch.distributed
+
+
+class MiniMaxText01RMSNormTP(CustomOp):
+    name = "MiniMaxText01RMSNormTP"
+
+    def __init__(self, hidden_size: int, eps: float = 1e-6) -> None:
+        super().__init__()
+        self.tp_world = get_tensor_model_parallel_world_size()
+        self.tp_rank = get_tensor_model_parallel_rank()
+        self.weight = nn.Parameter(torch.ones(int(hidden_size /
+                                                  self.tp_world)))
+
+        self.weight.weight_loader = self.weight_loader
+        self.variance_epsilon = eps
+        return
+
+    @staticmethod
+    def weight_loader(
+        param: nn.Parameter,
+        loaded_weight: torch.Tensor,
+    ) -> None:
+        tp_world = get_tensor_model_parallel_world_size()
+        tp_rank = get_tensor_model_parallel_rank()
+
+        shard_size = loaded_weight.shape[0] // tp_world
+        shard = slice(tp_rank * shard_size, (tp_rank + 1) * shard_size)
+        param.data.copy_(loaded_weight[shard])
+        return
+
+    def _forward(
+        self,
+        x: torch.Tensor,
+    ) -> torch.Tensor:
+        orig_dtype = x.dtype
+        x = x.to(torch.float32)
+        variance = x.pow(2).mean(dim=-1, keepdim=True, dtype=torch.float32)
+        if self.tp_world > 1:
+            variance = tensor_model_parallel_all_reduce(
+                variance) / self.tp_world
+        x = x * torch.rsqrt(variance + self.variance_epsilon)
+        x = x.to(orig_dtype) * self.weight
+        return x
+
+    def forward(
+        self,
+        x: torch.Tensor,
+        residual: Optional[torch.Tensor] = None,
+    ) -> Union[torch.Tensor, tuple[torch.Tensor, torch.Tensor]]:
+        assert residual is None, "RMSNorm does not support residual connection."
+        return self._forward(x)
+
+
+class MiniMaxText01LinearKernel:
+
+    @staticmethod
+    def jit_linear_forward_prefix(q: torch.Tensor,
+                                  k: torch.Tensor,
+                                  v: torch.Tensor,
+                                  kv_caches: torch.Tensor,
+                                  slope_rate: torch.Tensor,
+                                  block_size: int,
+                                  layer_idx: Optional[int] = None,
+                                  **kwargs) -> torch.Tensor:
+
+        slope_rate = slope_rate.to(torch.float32)
+        should_pad_dim = q.dim() == 3
+        if should_pad_dim:
+            q = q.unsqueeze(0)
+            k = k.unsqueeze(0)
+            v = v.unsqueeze(0)
+        b, h, n, d = q.shape
+        e = d
+        kv_history = kv_caches.reshape(1, h, d, e).contiguous()
+        output, kv_history = lightning_attention(q,
+                                                 k,
+                                                 v,
+                                                 slope_rate,
+                                                 block_size=block_size,
+                                                 kv_history=kv_history)
+        kv_caches.copy_(kv_history[:, :, -1, :, :].reshape(h, d, e))
+        assert output.shape[0] == 1, "batch size must be 1"
+        return rearrange(output.squeeze(0), "h n d -> n (h d)")
+
+
+class MiniMaxText01LinearAttention(nn.Module, MambaBase):
+
+    @property
+    def mamba_type(self) -> str:
+        return "linear_attention"
+
+    def get_attn_backend(self) -> type["AttentionBackend"]:
+        from vllm.v1.attention.backends.linear_attn import (
+            LinearAttentionBackend)
+        return LinearAttentionBackend
+
+    def get_state_dtype(self) -> tuple[torch.dtype]:
+        assert self.model_config is not None
+        assert self.cache_config is not None
+        return MambaStateDtypeCalculator.linear_attention_state_dtype(
+            self.model_config.dtype,
+            self.cache_config.mamba_cache_dtype,
+        )
+
+    def get_state_shape(self) -> tuple[tuple[int, int, int], ...]:
+        return MambaStateShapeCalculator.linear_attention_state_shape(
+            num_heads=self.num_heads,
+            tp_size=self.tp_size,
+            head_dim=self.head_dim)
+
+    def __init__(
+        self,
+        hidden_size: int,
+        hidden_inner_size: int,
+        num_heads: int,
+        head_dim: int,
+        max_position: int,
+        block_size: int,
+        num_hidden_layer: int,
+        model_config: Optional[ModelConfig] = None,
+        cache_config: Optional[CacheConfig] = None,
+        quant_config: Optional[QuantizationConfig] = None,
+        layer_idx: int = 0,
+        linear_layer_idx: int = 0,
+        prefix: str = "linear_attn",
+    ) -> None:
+        super().__init__()
+
+        self.layer_idx = layer_idx
+        self.BLOCK = block_size
+        self.hidden_size = hidden_size
+        self.num_heads = num_heads
+        self.head_dim = head_dim
+        self.total_num_heads = num_heads
+        self.hidden_inner_size = hidden_inner_size
+        self.tp_size = get_tensor_model_parallel_world_size()
+        self.tp_rank = get_tensor_model_parallel_rank()
+
+        assert self.total_num_heads % self.tp_size == 0
+        self.tp_heads = self.total_num_heads // self.tp_size
+        self.qkv_size = self.num_heads * self.head_dim
+        self.tp_hidden = self.head_dim * self.tp_heads
+        self.model_config = model_config
+        self.cache_config = cache_config
+        self.prefix = prefix
+
+        self.qkv_proj = ColumnParallelLinear(
+            hidden_size,
+            self.hidden_inner_size * 3,
+            bias=False,
+            quant_config=quant_config,
+            prefix=f"{prefix}.qkv_proj",
+        )
+        self.output_gate = ColumnParallelLinear(
+            hidden_size,
+            self.hidden_inner_size,
+            bias=False,
+            quant_config=quant_config,
+            prefix=f"{prefix}.output_gate",
+        )
+        self.out_proj = RowParallelLinear(
+            self.hidden_inner_size,
+            hidden_size,
+            bias=False,
+            quant_config=quant_config,
+            prefix=f"{prefix}.out_proj",
+        )
+        self.norm = MiniMaxText01RMSNormTP(
+            self.hidden_inner_size,
+            eps=1e-5,
+        )
+
+        slope_rate = MiniMaxText01LinearAttention._build_slope_tensor(
+            self.num_heads)
+        if num_hidden_layer <= 1:
+            self.slope_rate = slope_rate * (1 + 1e-5)
+        else:
+            self.slope_rate = slope_rate * (1 - layer_idx /
+                                            (num_hidden_layer - 1) + 1e-5)
+        self.tp_slope = self.slope_rate[self.tp_rank *
+                                        self.tp_heads:(self.tp_rank + 1) *
+                                        self.tp_heads].contiguous()
+
+        compilation_config = get_current_vllm_config().compilation_config
+        if prefix in compilation_config.static_forward_context:
+            raise ValueError(f"Duplicate layer name: {prefix}")
+        compilation_config.static_forward_context[prefix] = self
+
+    @staticmethod
+    def weight_direct_load(param: torch.Tensor,
+                           loaded_weight: torch.Tensor) -> None:
+        assert param.size() == loaded_weight.size()
+        param.data.copy_(loaded_weight)
+        return
+
+    @staticmethod
+    def _build_slope_tensor(n_attention_heads: int):
+
+        def get_slopes(n):
+
+            def get_slopes_power_of_2(n):
+                start = 2**(-(2**-(math.log2(n) - 3)))
+                ratio = start
+                return [start * ratio**i for i in range(n)]
+
+            if math.log2(n).is_integer():
+                return get_slopes_power_of_2(n)
+            else:
+                closest_power_of_2 = 2**math.floor(math.log2(n))
+                return (get_slopes_power_of_2(closest_power_of_2) + get_slopes(
+                    2 * closest_power_of_2)[0::2][:n - closest_power_of_2])
+
+        slopes = torch.tensor(get_slopes(n_attention_heads),
+                              dtype=torch.float32).reshape(
+                                  n_attention_heads, 1, 1)
+        return slopes
+
+    def _prefill_and_mix_infer(self, q, k, v, kv_cache, state_indices_tensor,
+                               attn_metadata):
+        hidden = []
+        for _prefill_idx in range(getattr(attn_metadata, "num_prefills", 0)):
+            if _prefill_idx >= len(attn_metadata.query_start_loc):
+                break
+            if _prefill_idx >= len(state_indices_tensor):
+                break
+            offset = attn_metadata.num_decode_tokens
+            _start = attn_metadata.query_start_loc[offset + _prefill_idx]
+            _end = attn_metadata.query_start_loc[offset + _prefill_idx + 1]
+            slot_id = state_indices_tensor[offset + _prefill_idx]
+            qs = q[_start:_end].transpose(0, 1).contiguous()
+            ks = k[_start:_end].transpose(0, 1).contiguous()
+            vs = v[_start:_end].transpose(0, 1).contiguous()
+            slice_layer_cache = kv_cache[slot_id, ...]
+
+            out_slice = MiniMaxText01LinearKernel.jit_linear_forward_prefix(
+                qs,
+                ks,
+                vs,
+                slice_layer_cache,
+                self.tp_slope,
+                self.BLOCK,
+                layer_idx=self.layer_idx)
+            hidden.append(out_slice.contiguous())
+        if attn_metadata.num_decode_tokens > 0:
+            hidden_decode = self._decode_infer(q, k, v, kv_cache,
+                                               state_indices_tensor,
+                                               attn_metadata)
+            hidden.insert(0, hidden_decode)
+
+        if not hidden:
+            return torch.empty((0, q.size(-1)), device=q.device, dtype=q.dtype)
+
+        hidden = torch.concat(hidden, dim=0).contiguous()
+        return hidden
+
+    def _decode_infer(self, q, k, v, kv_cache, state_indices_tensor,
+                      attn_metadata):
+        q = q[:attn_metadata.num_decode_tokens].unsqueeze(2).contiguous()
+        k = k[:attn_metadata.num_decode_tokens].unsqueeze(2).contiguous()
+        v = v[:attn_metadata.num_decode_tokens].unsqueeze(2).contiguous()
+        slot_id = state_indices_tensor[:attn_metadata.num_decodes]
+        hidden = linear_decode_forward_triton(q, k, v, kv_cache, self.tp_slope,
+                                              slot_id, 32)
+        return hidden
+
+    def forward(self, hidden_states: torch.Tensor, output: torch.Tensor,
+                positions: torch.Tensor) -> None:
+        torch.ops.vllm.linear_attention(
+            hidden_states,
+            output,
+            positions,
+            self.prefix,
+        )
+
+    def _forward(self, hidden_states: torch.Tensor, output: torch.Tensor,
+                 positions: torch.Tensor) -> None:
+        forward_context = get_forward_context()
+        attn_metadata: AttentionMetadata = forward_context.attn_metadata
+        if attn_metadata is not None:
+            assert isinstance(attn_metadata, dict)
+            attn_metadata = attn_metadata[self.prefix]
+            assert isinstance(attn_metadata, LinearAttentionMetadata)
+            num_actual_tokens = attn_metadata.num_prefill_tokens + \
+                attn_metadata.num_decode_tokens
+        else:
+            num_actual_tokens = hidden_states.shape[0]
+
+        qkv, _ = self.qkv_proj(hidden_states[:num_actual_tokens])
+        qkv32 = qkv.to(torch.float32)
+        qkvact = torch.nn.functional.silu(qkv32)
+        qkvact = qkvact.view((qkv.shape[0], self.tp_heads, -1))
+        q, k, v = torch.split(qkvact, [self.head_dim] * 3, dim=-1)
+        if attn_metadata is not None:
+            kv_cache = self.kv_cache[forward_context.virtual_engine][0]
+            state_indices_tensor = attn_metadata.state_indices_tensor
+
+            num_prefills = getattr(attn_metadata, "num_prefills", 0)
+            if num_prefills > 0:
+                num_decode_tokens = getattr(attn_metadata, "num_decode_tokens",
+                                            0)
+                for prefill_idx in range(num_prefills):
+                    q_start = attn_metadata.query_start_loc[num_decode_tokens +
+                                                            prefill_idx]
+                    q_end = attn_metadata.query_start_loc[num_decode_tokens +
+                                                          prefill_idx + 1]
+                    query_len = q_end - q_start
+                    context_len = attn_metadata.seq_lens[
+                        num_decode_tokens + prefill_idx] - query_len
+                    if context_len == 0:
+                        block_to_clear = state_indices_tensor[num_decode_tokens
+                                                              + prefill_idx]
+                        kv_cache[block_to_clear, ...] = 0
+
+        decode_only = getattr(attn_metadata, "num_prefills", 0) == 0
+        if attn_metadata is None:
+            hidden = torch.empty((q.shape[0], q.shape[1] * q.shape[2]),
+                                 device=q.device,
+                                 dtype=q.dtype)
+        else:
+            if not decode_only:
+                hidden = self._prefill_and_mix_infer(q, k, v, kv_cache,
+                                                     state_indices_tensor,
+                                                     attn_metadata)
+            else:
+                hidden = self._decode_infer(q, k, v, kv_cache,
+                                            state_indices_tensor,
+                                            attn_metadata)
+        hidden = self.norm._forward(hidden)
+        gate, _ = self.output_gate(hidden_states[:num_actual_tokens])
+        hidden = F.sigmoid(gate) * hidden
+        hidden = hidden.to(hidden_states.dtype)
+
+        output[:num_actual_tokens], _ = self.out_proj(hidden)
+
+
+def linear_attention(
+    hidden_states: torch.Tensor,
+    output: torch.Tensor,
+    positions: torch.Tensor,
+    layer_name: str,
+) -> None:
+    forward_context: ForwardContext = get_forward_context()
+    self = forward_context.no_compile_layers[layer_name]
+    self._forward(hidden_states=hidden_states,
+                  output=output,
+                  positions=positions)
+
+
+def linear_attention_fake(
+    hidden_states: torch.Tensor,
+    output: torch.Tensor,
+    positions: torch.Tensor,
+    layer_name: str,
+) -> None:
+    return
+
+
+direct_register_custom_op(
+    op_name="linear_attention",
+    op_func=linear_attention,
+    mutates_args=["output"],
+    fake_impl=linear_attention_fake,
+)

vllm/model_executor/layers/mamba/mamba_mixer.py

@@ -0,0 +1,466 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+from typing import TYPE_CHECKING, NamedTuple, Optional
+
+if TYPE_CHECKING:
+    from vllm.attention.backends.abstract import AttentionBackend
+
+import torch
+from torch import nn
+from torch.nn.parameter import Parameter
+
+from vllm.config import CacheConfig, ModelConfig, get_current_vllm_config
+from vllm.distributed.parallel_state import (
+    get_tensor_model_parallel_rank, get_tensor_model_parallel_world_size)
+from vllm.forward_context import ForwardContext, get_forward_context
+from vllm.model_executor.custom_op import CustomOp
+from vllm.model_executor.layers.layernorm import RMSNorm
+from vllm.model_executor.layers.linear import (ColumnParallelLinear,
+                                               MergedColumnParallelLinear,
+                                               RowParallelLinear)
+from vllm.model_executor.layers.mamba.abstract import MambaBase
+from vllm.model_executor.layers.mamba.mamba_utils import (
+    MambaStateDtypeCalculator, MambaStateShapeCalculator)
+from vllm.model_executor.layers.mamba.ops.causal_conv1d import (
+    causal_conv1d_fn, causal_conv1d_update)
+from vllm.model_executor.layers.mamba.ops.mamba_ssm import (
+    selective_scan_fn, selective_state_update)
+from vllm.model_executor.utils import set_weight_attrs
+from vllm.utils import direct_register_custom_op
+from vllm.v1.attention.backends.mamba1_attn import Mamba1AttentionMetadata
+
+
+# Adapted from transformers.models.mamba.modeling_mamba.MambaMixer
+@CustomOp.register("mamba_mixer")
+class MambaMixer(MambaBase, CustomOp):
+    """
+    Compute ∆, A, B, C, and D the state space parameters and compute
+    the `contextualized_states`. A, D are input independent
+    (see Mamba paper [1] Section 3.5.2 "Interpretation of A"
+    for why A isn't selective) ∆, B, C are input-dependent
+    (this is a key difference between Mamba and the linear time
+    invariant S4, and is why Mamba is called
+    **selective** state spaces)
+    """
+
+    def __init__(self,
+                 hidden_size: int,
+                 ssm_state_size: int,
+                 conv_kernel_size: int,
+                 intermediate_size: int,
+                 time_step_rank: int,
+                 use_conv_bias: bool,
+                 use_bias: bool,
+                 use_rms_norm: bool,
+                 rms_norm_has_weight: bool = True,
+                 rms_norm_eps: float = 1e-5,
+                 activation="silu",
+                 is_lora_enabled: bool = False,
+                 model_config: Optional[ModelConfig] = None,
+                 cache_config: Optional[CacheConfig] = None,
+                 prefix: str = ""):
+        super().__init__()
+        self.time_step_rank = time_step_rank
+        self.ssm_state_size = ssm_state_size
+        self.use_rms_norm = use_rms_norm
+        self.activation = activation
+        self.is_lora_enabled = is_lora_enabled
+        self.conv_kernel_size = conv_kernel_size
+        self.intermediate_size = intermediate_size
+
+        self.conv1d = ColumnParallelLinear(
+            input_size=conv_kernel_size,
+            output_size=intermediate_size,
+            bias=use_conv_bias,
+        )
+        # unsqueeze to fit conv1d weights shape into the linear weights shape.
+        # Can't do this in `weight_loader` since it already exists in
+        # `ColumnParallelLinear` and `set_weight_attrs`
+        # doesn't allow to override it
+        self.conv1d.weight.data = self.conv1d.weight.data.unsqueeze(1)
+
+        self.in_proj = MergedColumnParallelLinear(hidden_size,
+                                                  [intermediate_size] * 2,
+                                                  bias=use_bias)
+
+        # selective projection used to make dt, B and C input dependent
+        self.x_proj = RowParallelLinear(
+            intermediate_size,
+            time_step_rank + ssm_state_size * 2,
+            bias=False,
+        )
+        # time step projection (discretization) -
+        # In the forward we need to apply dt_proj without the bias,
+        # as the bias is added in the selective scan kernel.
+        self.dt_proj = ColumnParallelLinear(time_step_rank,
+                                            intermediate_size,
+                                            bias=True,
+                                            skip_bias_add=True)
+
+        def weight_loader(param: Parameter, loaded_weight: torch.Tensor):
+            tp_rank = get_tensor_model_parallel_rank()
+            tp_size = get_tensor_model_parallel_world_size()
+            param.data.copy_(
+                loaded_weight.data.split(loaded_weight.shape[0] // tp_size,
+                                         dim=0)[tp_rank])
+
+        def A_weight_loader(param: Parameter, loaded_weight: torch.Tensor):
+            weight_loader(param, -torch.exp(loaded_weight.float()))
+
+        tp_size = get_tensor_model_parallel_world_size()
+        self.A = nn.Parameter(
+            torch.empty(
+                intermediate_size // tp_size,
+                ssm_state_size,
+                dtype=torch.float32,
+            ))
+        self.D = nn.Parameter(torch.ones(intermediate_size // tp_size))
+
+        set_weight_attrs(self.D, {"weight_loader": weight_loader})
+        set_weight_attrs(self.A, {"weight_loader": A_weight_loader})
+
+        self.out_proj = RowParallelLinear(
+            intermediate_size,
+            hidden_size,
+            bias=use_bias,
+            input_is_parallel=True,
+        )
+
+        self.dt_layernorm = RMSNorm(
+            time_step_rank,
+            eps=rms_norm_eps,
+            has_weight=rms_norm_has_weight,
+        ) if use_rms_norm else None
+
+        self.b_layernorm = RMSNorm(
+            ssm_state_size,
+            eps=rms_norm_eps,
+            has_weight=rms_norm_has_weight,
+        ) if use_rms_norm else None
+
+        self.c_layernorm = RMSNorm(
+            ssm_state_size,
+            eps=rms_norm_eps,
+            has_weight=rms_norm_has_weight,
+        ) if use_rms_norm else None
+
+        compilation_config = get_current_vllm_config().compilation_config
+        if prefix in compilation_config.static_forward_context:
+            raise ValueError(f"Duplicate layer name: {prefix}")
+        compilation_config.static_forward_context[prefix] = self
+        # The inner tuple is (conv_state, ssm_state)
+        self.kv_cache = (torch.tensor([]), torch.tensor([]))
+
+        self.model_config = model_config
+        self.cache_config = cache_config
+        self.prefix = prefix
+
+    def _ssm_transform(
+            self, x: torch.Tensor
+    ) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+        if self.is_lora_enabled:
+            #  Lora kernel requires contiguous tensor.
+            ssm_params = self.x_proj(x.contiguous())[0]
+        else:
+            ssm_params = self.x_proj(x)[0]
+        time_step, B, C = torch.split(
+            ssm_params,
+            [self.time_step_rank, self.ssm_state_size, self.ssm_state_size],
+            dim=-1)
+        if self.use_rms_norm:
+            assert self.dt_layernorm is not None
+            assert self.b_layernorm is not None
+            assert self.c_layernorm is not None
+            time_step = self.dt_layernorm(time_step.contiguous())
+            B = self.b_layernorm(B.contiguous())
+            C = self.c_layernorm(C.contiguous())
+        discrete_time_step = self.dt_proj(time_step)[0].transpose(-2, -1)
+        return discrete_time_step, B, C
+
+    def forward(self, hidden_states: torch.Tensor, output: torch.Tensor):
+        torch.ops.vllm.mamba_mixer(
+            hidden_states,
+            output,
+            self.prefix,
+        )
+
+    def forward_native(self, hidden_states: torch.Tensor,
+                       output: torch.Tensor):
+        pass
+
+    def forward_cuda(self, hidden_states: torch.Tensor, output: torch.Tensor):
+        """
+        Run the Mamba-1 SSM pipeline.
+
+        Steps
+        -----
+        1. Apply the gated-MLP linear projection to the raw input.
+        2. Pass the projected sequence through the convolutional mixing layer.
+        3. Feed the result into the State-Space Model (SSM) blocks.
+        4. Perform the recurrence y ← SSM(A, B, C, Δ)(x)
+           to produce contextual representations.
+        5. Project the contextualised sequence back
+           to the output embedding dimension.
+
+        Batch handling
+        --------------
+        Prefill and decode tokens are processed by dedicated CUDA
+        kernels for both the convolutional (conv1d) and SSM stages.
+        In the case of a mixed batch (containing both prefill and
+        decode tokens), both sets of kernels are executed independently
+        and their outputs are concatenated before the final output projection.
+        """
+
+        forward_context: ForwardContext = get_forward_context()
+        attn_metadata = forward_context.attn_metadata
+
+        if attn_metadata is not None:
+            assert isinstance(attn_metadata, dict)
+            attn_metadata = attn_metadata[self.prefix]
+            mamba1_metadata = attn_metadata
+            assert isinstance(mamba1_metadata, Mamba1AttentionMetadata)
+            query_start_loc = mamba1_metadata.query_start_loc
+            state_indices_tensor = mamba1_metadata.state_indices_tensor
+            self_kv_cache = self.kv_cache[forward_context.virtual_engine]
+            conv_state = self_kv_cache[0].transpose(-1, -2)
+            ssm_state = self_kv_cache[1]
+            has_initial_states = mamba1_metadata.has_initial_states
+            num_padded_decodes = mamba1_metadata.num_padded_decodes
+
+        # 1. Gated MLP's linear projection
+        projected_states = self.in_proj(hidden_states)[0].transpose(-2, -1)
+        hidden_states_BC, gate = projected_states.chunk(2, dim=-2)
+
+        conv_weights = self.conv1d.weight.view(self.conv1d.weight.size(0),
+                                               self.conv1d.weight.size(2))
+
+        if attn_metadata is None:
+            # V1 profile run
+            hidden_states_BC = hidden_states_BC.contiguous()
+            return self.out_proj(hidden_states_BC.transpose(-2, -1))[0]
+
+        num_prefill_tokens = attn_metadata.num_prefill_tokens  # token count
+        num_decode_tokens = attn_metadata.num_decode_tokens
+        num_prefills = attn_metadata.num_prefills  # request count
+        num_decodes = attn_metadata.num_decode_tokens  # token count (=request)
+        has_prefill = num_prefill_tokens > 0
+        has_decode = num_decode_tokens > 0
+        num_actual_tokens = num_prefill_tokens + num_decode_tokens
+
+        prefill_decode_split = split_batch_to_prefill_and_decode(
+            hidden_states_BC,
+            gate,
+            state_indices_tensor,
+            query_start_loc,
+            has_initial_states,
+            num_prefill_tokens,
+            num_decode_tokens,
+            num_prefills,
+            num_decodes,
+            num_padded_decodes,
+        )
+        hidden_states_BC_p = prefill_decode_split.hidden_states_BC_p
+        hidden_states_BC_d = prefill_decode_split.hidden_states_BC_d
+        gate_p = prefill_decode_split.gate_p
+        gate_d = prefill_decode_split.gate_d
+        state_indices_tensor_p = prefill_decode_split.state_indices_tensor_p
+        state_indices_tensor_d = prefill_decode_split.state_indices_tensor_d
+        query_start_loc_p = prefill_decode_split.query_start_loc_p
+        has_initial_states_p = prefill_decode_split.has_initial_states_p
+
+        ssm_outputs = []
+
+        if has_prefill:
+            # 2. Convolution sequence transformation
+            conv_out_p = causal_conv1d_fn(
+                hidden_states_BC_p,
+                conv_weights,
+                self.conv1d.bias,
+                activation=self.activation,
+                conv_states=conv_state,
+                has_initial_state=has_initial_states_p,
+                cache_indices=state_indices_tensor_p,
+                query_start_loc=query_start_loc_p)
+            # 3. State Space Model sequence transformations.
+            discrete_time_step_p, B_p, C_p = self._ssm_transform(
+                conv_out_p.transpose(-2, -1))
+            time_proj_bias = self._time_proj_bias()
+
+            # 4. Perform the recurrence y ← SSM(A, B, C, Δ)(x)
+            scan_out_p = selective_scan_fn(
+                conv_out_p,
+                ssm_state,
+                discrete_time_step_p,
+                self.A,
+                B_p.transpose(-2, -1),
+                C_p.transpose(-2, -1),
+                self.D.float(),
+                gate_p,
+                time_proj_bias,
+                delta_softplus=True,
+                cache_indices=state_indices_tensor_p,
+                has_initial_state=has_initial_states_p,
+                query_start_loc=query_start_loc_p)
+            ssm_outputs.append(scan_out_p)
+
+        if has_decode:
+            # 2. Convolution sequence transformation
+            conv_out_d = causal_conv1d_update(
+                hidden_states_BC_d.transpose(0, 1),
+                conv_state,
+                conv_weights,
+                self.conv1d.bias,
+                self.activation,
+                conv_state_indices=state_indices_tensor_d).transpose(0, 1)
+
+            # 3. State Space Model sequence transformation.
+            discrete_time_step_d, B_d, C_d = self._ssm_transform(
+                conv_out_d.transpose(-2, -1))
+            time_proj_bias = self._time_proj_bias()
+
+            # 4. Perform the recurrence y ← SSM(A, B, C, Δ)(x)
+            scan_outputs_d = torch.empty_like(
+                hidden_states_BC_d.transpose(0, 1))
+            selective_state_update(ssm_state,
+                                   conv_out_d.transpose(0, 1),
+                                   discrete_time_step_d.transpose(0, 1),
+                                   self.A,
+                                   B_d,
+                                   C_d,
+                                   self.D,
+                                   gate_d.transpose(0, 1),
+                                   time_proj_bias,
+                                   dt_softplus=True,
+                                   state_batch_indices=state_indices_tensor_d,
+                                   out=scan_outputs_d)
+            scan_outputs_d = scan_outputs_d.transpose(0, 1)
+
+            ssm_outputs.insert(0, scan_outputs_d)
+
+        scan_outputs_combined = ssm_outputs[0] if len(
+            ssm_outputs) == 1 else torch.cat(ssm_outputs, dim=-1)
+
+        # 5. Final output projection
+        if self.is_lora_enabled:  # Lora kernel requires contiguous tensor.
+            scan_outputs_combined = scan_outputs_combined.transpose(
+                -2, -1).contiguous()
+            out = self.out_proj(scan_outputs_combined)[0]
+        else:
+            out = self.out_proj(scan_outputs_combined.transpose(-2, -1))[0]
+
+        output[:num_actual_tokens] = out
+
+    def get_state_dtype(self) -> tuple[torch.dtype]:
+        assert self.model_config is not None
+        assert self.cache_config is not None
+        return MambaStateDtypeCalculator.mamba1_state_dtype(
+            self.model_config.dtype,
+            self.cache_config.mamba_cache_dtype,
+            self.cache_config.mamba_ssm_cache_dtype,
+        )
+
+    def get_state_shape(self) -> tuple[tuple[int, ...], tuple[int, ...]]:
+        return MambaStateShapeCalculator.mamba1_state_shape(
+            tp_world_size=get_tensor_model_parallel_world_size(),
+            intermediate_size=self.intermediate_size,
+            state_size=self.ssm_state_size,
+            conv_kernel=self.conv_kernel_size,
+        )
+
+    @property
+    def mamba_type(self) -> str:
+        return "mamba1"
+
+    def get_attn_backend(self) -> type["AttentionBackend"]:
+        from vllm.v1.attention.backends.mamba1_attn import (
+            Mamba1AttentionBackend)
+        return Mamba1AttentionBackend
+
+    def _time_proj_bias(self) -> Optional[torch.Tensor]:
+        if hasattr(self.dt_proj, "bias") and self.dt_proj.bias is not None:
+            return self.dt_proj.bias.float()
+        return None
+
+
+class PrefillDecodeSplit(NamedTuple):
+    hidden_states_BC_p: torch.Tensor
+    hidden_states_BC_d: torch.Tensor
+    gate_p: torch.Tensor
+    gate_d: torch.Tensor
+    state_indices_tensor_p: torch.Tensor
+    state_indices_tensor_d: torch.Tensor
+    query_start_loc_p: Optional[torch.Tensor]
+    has_initial_states_p: Optional[torch.Tensor]
+
+
+def split_batch_to_prefill_and_decode(
+    hidden_states_BC: torch.Tensor,
+    gate: torch.Tensor,
+    state_indices_tensor: torch.Tensor,
+    query_start_loc: torch.Tensor,
+    has_initial_states: Optional[torch.Tensor],
+    num_prefill_tokens: int,
+    num_decode_tokens: int,
+    num_prefills: int,
+    num_decodes: int,
+    num_padded_decodes: int,
+) -> PrefillDecodeSplit:
+
+    num_actual_tokens = num_prefill_tokens + num_padded_decodes
+
+    # In v1, decode tokens come first, then prefill tokens.
+    hidden_states_BC_d, hidden_states_BC_p = torch.split(
+        hidden_states_BC[..., :num_actual_tokens],
+        [num_padded_decodes, num_prefill_tokens],
+        dim=-1)
+    gate_d, gate_p = torch.split(gate[..., :num_actual_tokens],
+                                 [num_padded_decodes, num_prefill_tokens],
+                                 dim=-1)
+
+    # num_padded_decodes accounts for CUDA graph padding when applicable
+    state_indices_tensor_d, state_indices_tensor_p = torch.split(
+        state_indices_tensor[:num_padded_decodes + num_prefills],
+        [num_padded_decodes, num_prefills],
+        dim=0)
+    query_start_loc_p = (query_start_loc[-num_prefills - 1:] -
+                         num_padded_decodes if num_prefills > 0 else None)
+    has_initial_states_p = has_initial_states[-num_prefills:] if (
+        has_initial_states is not None and num_prefills > 0) else None
+
+    return PrefillDecodeSplit(
+        hidden_states_BC_p=hidden_states_BC_p,
+        hidden_states_BC_d=hidden_states_BC_d,
+        gate_p=gate_p,
+        gate_d=gate_d,
+        state_indices_tensor_p=state_indices_tensor_p,
+        state_indices_tensor_d=state_indices_tensor_d,
+        query_start_loc_p=query_start_loc_p,
+        has_initial_states_p=has_initial_states_p,
+    )
+
+
+def mamba_mixer(
+    hidden_states: torch.Tensor,
+    output: torch.Tensor,
+    layer_name: str,
+) -> None:
+    forward_context: ForwardContext = get_forward_context()
+    self = forward_context.no_compile_layers[layer_name]
+    self.forward_cuda(hidden_states=hidden_states, output=output)
+
+
+def mamba_mixer_fake(
+    hidden_states: torch.Tensor,
+    output: torch.Tensor,
+    layer_name: str,
+) -> None:
+    return
+
+
+direct_register_custom_op(
+    op_name="mamba_mixer",
+    op_func=mamba_mixer,
+    mutates_args=["output"],
+    fake_impl=mamba_mixer_fake,
+)

vllm/model_executor/layers/mamba/mamba_mixer2.py

@@ -0,0 +1,764 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+from typing import TYPE_CHECKING, Optional, Union
+
+if TYPE_CHECKING:
+    from vllm.attention.backends.abstract import AttentionBackend
+
+import torch
+from torch import nn
+
+from vllm.attention.backends.abstract import AttentionMetadata
+from vllm.config import CacheConfig, ModelConfig, get_current_vllm_config
+from vllm.distributed import (divide, get_tensor_model_parallel_rank,
+                              get_tensor_model_parallel_world_size,
+                              tensor_model_parallel_all_gather,
+                              tensor_model_parallel_all_reduce)
+from vllm.forward_context import ForwardContext, get_forward_context
+from vllm.model_executor.custom_op import CustomOp
+from vllm.model_executor.layers.linear import (ColumnParallelLinear,
+                                               MergedColumnParallelLinear,
+                                               RowParallelLinear)
+from vllm.model_executor.layers.mamba.abstract import MambaBase
+from vllm.model_executor.layers.mamba.mamba_utils import (
+    MambaStateDtypeCalculator, MambaStateShapeCalculator)
+from vllm.model_executor.layers.mamba.ops.causal_conv1d import (
+    causal_conv1d_fn, causal_conv1d_update)
+from vllm.model_executor.layers.mamba.ops.layernorm_gated import rms_norm_gated
+from vllm.model_executor.layers.mamba.ops.mamba_ssm import (
+    selective_state_update)
+from vllm.model_executor.layers.mamba.ops.ssd_combined import (
+    mamba_chunk_scan_combined_varlen)
+from vllm.model_executor.layers.quantization import QuantizationConfig
+from vllm.model_executor.model_loader.weight_utils import (
+    LoaderFunction, composed_weight_loader, sharded_weight_loader)
+from vllm.model_executor.utils import set_weight_attrs
+from vllm.utils import direct_register_custom_op
+from vllm.v1.attention.backends.mamba2_attn import Mamba2AttentionMetadata
+
+# Added by the IBM Team, 2024
+
+
+# Adapted from transformers.models.mamba2.modeling_mamba2.MambaRMSNormGated
+@CustomOp.register("mixer2_gated_rms_norm")
+class Mixer2RMSNormGated(CustomOp):
+
+    def __init__(self,
+                 full_hidden_size: int,
+                 full_n_groups: int,
+                 use_rms_norm: bool = True,
+                 eps: float = 1e-6):
+        super().__init__()
+        self.tp_size = get_tensor_model_parallel_world_size()
+        self.tp_rank = get_tensor_model_parallel_rank()
+        self.full_hidden_size = full_hidden_size
+        self.group_size = full_hidden_size // full_n_groups
+        self.per_rank_hidden_size = full_hidden_size // self.tp_size
+        self.n_groups = full_hidden_size // self.group_size
+
+        self.variance_epsilon = eps
+        self.use_rms_norm = use_rms_norm
+        if self.use_rms_norm:
+            # Register norm weight only if we're actually applying RMSNorm
+            self.weight = nn.Parameter(torch.ones(self.per_rank_hidden_size))
+            set_weight_attrs(self.weight,
+                             {"weight_loader": sharded_weight_loader(0)})
+        else:
+            # Avoid checkpoint mismatch by skipping unused parameter
+            self.register_parameter("weight", None)
+        assert (self.full_hidden_size % self.tp_size == 0
+                ), "Tensor parallel world size must divide hidden size."
+
+    def forward_native(
+        self,
+        x: torch.Tensor,
+        gate: torch.Tensor,
+    ):
+        # Three tensor-parallel cases:
+        #   1. n_groups is 1
+        #      In this case we parallelize along the reduction dim.
+        #      Each rank computes a local sum of squares followed by AllReduce
+        #   2. tp_size divides n_groups
+        #      Each rank only reduces within its local group(s).
+        #      No collective ops necessary.
+        #   3. The general case can be pretty complicated so we AllGather
+        #      the input and then redundantly compute the RMSNorm.
+        input_dtype = x.dtype
+        x = x * nn.functional.silu(gate.to(torch.float32))
+        if not self.use_rms_norm:
+            return x.to(input_dtype)
+
+        if self.n_groups == 1:
+            if self.tp_size > 1:
+                # Compute local sum and then reduce to obtain global sum
+                local_sums = x.pow(2).sum(dim=-1, keepdim=True)
+                global_sums = tensor_model_parallel_all_reduce(local_sums)
+                # Calculate the variance
+                count = self.tp_size * x.shape[-1]
+                variance = global_sums / count
+
+            else:
+                variance = x.pow(2).mean(-1, keepdim=True)
+            x = x * torch.rsqrt(variance + self.variance_epsilon)
+        else:
+            redundant_tp: bool = self.n_groups % self.tp_size != 0
+            if redundant_tp:
+                # To handle the general case, redundantly apply the variance
+                x = tensor_model_parallel_all_gather(x, -1)
+
+            *prefix_dims, hidden_dim = x.shape
+            group_count = hidden_dim // self.group_size
+            x_grouped = x.view(*prefix_dims, group_count, self.group_size)
+            variance = x_grouped.pow(2).mean(-1, keepdim=True)
+            x_grouped = x_grouped * torch.rsqrt(variance +
+                                                self.variance_epsilon)
+            x = x_grouped.view(*prefix_dims, hidden_dim)
+
+            if redundant_tp:
+                start = self.per_rank_hidden_size * self.tp_rank
+                end = start + self.per_rank_hidden_size
+                x = x[..., start:end]
+
+        return self.weight * x.to(input_dtype)
+
+    def forward_cuda(
+        self,
+        x: torch.Tensor,
+        gate: torch.Tensor,
+    ) -> Union[torch.Tensor, tuple[torch.Tensor, torch.Tensor]]:
+        input_dtype = x.dtype
+        if not self.use_rms_norm:
+            # Keep gate in float32 for numerical stability during silu
+            return x * nn.functional.silu(gate.to(
+                torch.float32)).to(input_dtype)
+
+        if (((self.n_groups % self.tp_size) != 0) or self.n_groups != 1):
+            return self.forward_native(x, gate)
+
+        return rms_norm_gated(x,
+                              self.weight.data,
+                              bias=None,
+                              z=gate,
+                              eps=self.variance_epsilon,
+                              norm_before_gate=False)
+
+
+def mamba_v2_sharded_weight_loader(
+    shard_spec: list[tuple[int, int, float]],
+    tp_size: int,
+    tp_rank: int,
+) -> LoaderFunction:
+    """Create a weight loader for mamba v2. This ensures that the projections
+    are correctly sharded so that they can be split into x, B, C. It also
+    ensures that all the groups corresponding to a head shard is placed
+    together with it.
+    """
+
+    def loader(param: torch.Tensor, loaded_weight: torch.Tensor) -> None:
+
+        # - track boundary of (sharded) param, and loaded_weight, respectively
+        boundary, loaded_boundary = 0, 0
+
+        # - iterate over the shard specs
+        for full_dim, extra, duplicate_groups in shard_spec:
+            # - full dim is the model dim (before TP).
+            # - extra > 0, means there is expected overall increase
+            #   of dimensions. This is so because of replication.
+            # - ratio is used map the tp_rank to the actual shard
+            #   rank. This is useful when there is replication of
+            #   groups to accompany head shards.
+
+            # - size of the loaded shard
+            shard_size = full_dim // tp_size
+
+            # - compute the rank into the loaded shard.
+            # - if there is replication, different TP shards will
+            #   take from the same rank.
+            # NOTE: currently we only support duplication
+            # in the case where num_groups == 1
+            rank = 0 if duplicate_groups else tp_rank
+
+            # - leftmost boundary index into loaded weight.
+            loaded_skip = rank * shard_size
+            loaded_start_idx = loaded_boundary + loaded_skip
+
+            # - take these many dims from the loaded weight.
+            take = min(shard_size, full_dim - extra - loaded_skip)
+
+            # - always shard on dim 0
+            # - the ignore is for a mundane mypy error as it does not
+            #   seem to handle slices well.
+            # https://github.com/python/mypy/issues/2410
+            param.data[
+                boundary:(boundary + take),
+                ...  # type: ignore[misc]
+            ] = loaded_weight[loaded_start_idx:(loaded_start_idx +
+                                                take)  # type: ignore[misc]
+                              ]  # type: ignore[misc]
+
+            # move indexing boundaries
+            boundary += shard_size
+            loaded_boundary += full_dim - extra
+
+    return loader
+
+
+# Adapted from transformers.models.mamba.modeling_mamba.MambaMixer
+@CustomOp.register("mamba_mixer2")
+class MambaMixer2(MambaBase, CustomOp):
+    """
+    Compute ∆, A, B, C, and D the state space parameters and compute
+    the `contextualized_states`. A, D are input independent
+    (see Mamba paper [1] Section 3.5.2 "Interpretation of A"
+    for why A isn't selective) ∆, B, C are input-dependent
+    (this is a key difference between Mamba and the linear time
+    invariant S4, and is why Mamba is called
+    **selective** state spaces)
+    """
+
+    def __init__(self,
+                 hidden_size: int,
+                 ssm_state_size: int,
+                 conv_kernel_size: int,
+                 intermediate_size: int,
+                 use_conv_bias: bool,
+                 use_bias: bool,
+                 n_groups: int = 1,
+                 num_heads: int = 128,
+                 head_dim: int = 64,
+                 rms_norm_eps: float = 1e-5,
+                 activation: str = "silu",
+                 use_rms_norm: bool = True,
+                 model_config: Optional[ModelConfig] = None,
+                 cache_config: Optional[CacheConfig] = None,
+                 quant_config: Optional[QuantizationConfig] = None,
+                 prefix: str = ""):
+        super().__init__()
+
+        # For TP, the sharding plan is as follows:
+        # - for the conv modules, since
+        #   conv_dim = intermediate_size * 2 * n_groups * ssm_state_size,
+        #   we shard intermediate_size and n_groups
+        # - since intermediate_size = n_heads * head_dim, sharding on
+        #   intermediate_size is achieved by sharding on n_heads.
+        # - IF, world_size divides groups, then sharding
+        #   (n_groups / world_size, n_heads / world_size)
+        #   also maintains the invariant n_heads % n_groups == 0
+        # - HOWEVER IF, world_size DOES NOT divide groups, then we need
+        #   to allocate extra space in the shard, such that groups
+        #   may be replicated to follow the head shard.
+        # - NOTE: currently for the world size DOES NOT divide groups
+        #   case, we only support the case when n_groups == 1
+        self.tp_size = get_tensor_model_parallel_world_size()
+        tp_rank = get_tensor_model_parallel_rank()
+
+        assert (num_heads % self.tp_size == 0
+                ), "Tensor parallel world size must divide num heads."
+
+        assert (n_groups % self.tp_size) == 0 or n_groups == 1, (
+            "If tensor parallel world size does not divide num_groups, "
+            "then num_groups must equal 1.")
+
+        assert (n_groups % self.tp_size == 0) or self.tp_size == 1 or \
+            quant_config is None, (
+            "Tensor parallel currently supported for quantized models only "
+            "if tensor parallel world size divides num groups."
+        )
+
+        self.ssm_state_size = ssm_state_size
+        self.conv_kernel_size = conv_kernel_size
+        self.activation = activation
+
+        self.intermediate_size = intermediate_size
+        self.head_dim = head_dim
+        self.num_heads = num_heads
+
+        self.n_groups = n_groups
+        if n_groups % self.tp_size != 0:
+            # - for TP we shard conv_dim by sharding on n_groups,
+            # - but if n_groups cannot divide tp_size, we need to
+            #   extend some extra groups
+            groups = MambaStateShapeCalculator.extra_groups_for_head_shards(
+                n_groups, self.tp_size)
+            self.n_groups = n_groups + groups
+
+        self.groups_ssm_state_size = self.n_groups * self.ssm_state_size
+        self.conv_dim = intermediate_size + 2 * self.groups_ssm_state_size
+
+        if n_groups % self.tp_size == 0:
+            self.conv1d = MergedColumnParallelLinear(
+                input_size=conv_kernel_size,
+                output_sizes=[
+                    intermediate_size,
+                    self.groups_ssm_state_size,
+                    self.groups_ssm_state_size,
+                ],
+                bias=use_conv_bias,
+                quant_config=None,
+                prefix=f"{prefix}.conv1d",
+            )
+
+            self.in_proj = MergedColumnParallelLinear(
+                input_size=hidden_size,
+                output_sizes=[
+                    intermediate_size,
+                    intermediate_size,
+                    self.groups_ssm_state_size,
+                    self.groups_ssm_state_size,
+                    self.num_heads,
+                ],
+                bias=use_bias,
+                quant_config=quant_config,
+                prefix=f"{prefix}.in_proj",
+            )
+        else:
+            # This is the n_groups == 1 case,
+            # where we need to duplicate groups if TP>1.
+
+            self.conv1d = ColumnParallelLinear(
+                input_size=conv_kernel_size,
+                output_size=self.conv_dim,
+                bias=use_conv_bias,
+                quant_config=None,
+                prefix=f"{prefix}.conv1d",
+            )
+
+            self.in_proj = ColumnParallelLinear(
+                input_size=hidden_size,
+                output_size=intermediate_size + self.conv_dim + self.num_heads,
+                bias=use_bias,
+                quant_config=quant_config,
+                prefix=f"{prefix}.in_proj",
+            )
+
+            # - because in_proj is a concatenation of 3 weights, we
+            #   need to interleave them before sharding
+            # - use the custom weight loader mamba_v2_sharded_weight_loader
+            #   for conv1d.bias, covn1d.weight and in_proj.weight
+            # - need to set these settings, to assign the groups
+            #   to the head shards
+            group_shard_settings = (
+                self.groups_ssm_state_size,  # expected model size
+                (self.n_groups - n_groups) *
+                self.ssm_state_size,  # extra dims assigned
+                n_groups == 1,  # if there was only one group
+            )
+            intermediate_settings = (intermediate_size, 0, False)
+            head_settings = (self.num_heads, 0, False)
+
+            # - the weight already has a "weight_loader" attribute
+            #   which set_weight_attrs will raise if we do not
+            #   delete before trying to override it
+            # - ditto for the other two weights below
+            delattr(self.conv1d.bias, "weight_loader")
+            set_weight_attrs(
+                self.conv1d.bias,
+                {
+                    "weight_loader":
+                    mamba_v2_sharded_weight_loader(
+                        [
+                            intermediate_settings,
+                            group_shard_settings,
+                            group_shard_settings,
+                        ],
+                        self.tp_size,
+                        tp_rank,
+                    )
+                },
+            )
+
+            delattr(self.conv1d.weight, "weight_loader")
+            set_weight_attrs(
+                self.conv1d.weight,
+                {
+                    "weight_loader":
+                    mamba_v2_sharded_weight_loader(
+                        [
+                            intermediate_settings,
+                            group_shard_settings,
+                            group_shard_settings,
+                        ],
+                        self.tp_size,
+                        tp_rank,
+                    )
+                },
+            )
+
+            if quant_config is None:
+                # - quant layers do not have a weight loader
+                delattr(self.in_proj.weight, "weight_loader")
+                set_weight_attrs(
+                    self.in_proj.weight,
+                    {
+                        "weight_loader":
+                        mamba_v2_sharded_weight_loader(
+                            [
+                                intermediate_settings,  # for gate
+                                intermediate_settings,
+                                group_shard_settings,
+                                group_shard_settings,
+                                head_settings,  # for dt
+                            ],
+                            self.tp_size,
+                            tp_rank,
+                        )
+                    },
+                )
+
+        # unsqueeze to fit conv1d weights shape into the linear weights shape.
+        # Can't do this in `weight_loader` since it already exists in
+        # `ColumnParallelLinear` and `MergedColumnParallelLinear`,
+        # and `set_weight_attrs` doesn't allow to override it
+        self.conv1d.weight.data = self.conv1d.weight.data.unsqueeze(1)
+
+        # - these are TPed by heads to reduce the size of the
+        #   temporal shape
+        self.A = nn.Parameter(
+            torch.empty(
+                divide(num_heads, self.tp_size),
+                dtype=torch.float32,
+            ))
+        self.D = nn.Parameter(torch.ones(num_heads // self.tp_size))
+        self.dt_bias = nn.Parameter(torch.ones(num_heads // self.tp_size))
+        self.use_rms_norm = use_rms_norm
+
+        set_weight_attrs(self.D, {"weight_loader": sharded_weight_loader(0)})
+        a_weight_loader = composed_weight_loader(
+            sharded_weight_loader(0), lambda x: -torch.exp(x.float()))
+        set_weight_attrs(self.A, {"weight_loader": a_weight_loader})
+        set_weight_attrs(self.dt_bias,
+                         {"weight_loader": sharded_weight_loader(0)})
+
+        self.out_proj = RowParallelLinear(
+            intermediate_size,
+            hidden_size,
+            bias=use_bias,
+            input_is_parallel=True,
+            quant_config=quant_config,
+            prefix=f"{prefix}.out_proj",
+        )
+
+        self.norm = Mixer2RMSNormGated(intermediate_size,
+                                       n_groups,
+                                       self.use_rms_norm,
+                                       eps=rms_norm_eps)
+
+        compilation_config = get_current_vllm_config().compilation_config
+        if prefix in compilation_config.static_forward_context:
+            raise ValueError(f"Duplicate layer name: {prefix}")
+        compilation_config.static_forward_context[prefix] = self
+        # The tuple is (conv_state, ssm_state)
+        self.kv_cache = (torch.tensor([]), torch.tensor([]))
+
+        self.model_config = model_config
+        self.cache_config = cache_config
+        self.prefix = prefix
+
+    def forward_native(
+        self,
+        hidden_states: torch.Tensor,
+        output: torch.Tensor,
+        mup_vector: Optional[torch.Tensor] = None,
+    ):
+        pass
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        output: torch.Tensor,
+        mup_vector: Optional[torch.Tensor] = None,
+    ):
+        torch.ops.vllm.mamba_mixer2(
+            hidden_states,
+            output,
+            self.prefix,
+            mup_vector,
+        )
+
+    def forward_cuda(
+        self,
+        hidden_states: torch.Tensor,
+        output: torch.Tensor,
+        mup_vector: Optional[torch.Tensor] = None,
+    ):
+        forward_context = get_forward_context()
+        # attn_metadata contains metadata necessary for the mamba2 triton
+        # kernels to operate in continuous batching and in chunked prefill
+        # modes; they are computed at top-level model forward since they
+        # stay the same and reused for all mamba layers in the same iteration
+        attn_metadata: AttentionMetadata = forward_context.attn_metadata
+
+        if attn_metadata is not None:
+            assert isinstance(attn_metadata, dict)
+            attn_metadata = attn_metadata[self.prefix]
+            assert isinstance(attn_metadata, Mamba2AttentionMetadata)
+            self_kv_cache = self.kv_cache[forward_context.virtual_engine]
+            # conv_state = (..., dim, width-1) yet contiguous along 'dim'
+            conv_state = self_kv_cache[0].transpose(-1, -2)
+            ssm_state = self_kv_cache[1]
+            state_indices_tensor = attn_metadata.state_indices_tensor
+            has_initial_states_p = attn_metadata.has_initial_states_p
+            prep_initial_states = attn_metadata.prep_initial_states
+            chunk_size = attn_metadata.chunk_size
+            seq_idx_p = attn_metadata.seq_idx_p
+            chunk_indices_p = attn_metadata.chunk_indices_p
+            chunk_offsets_p = attn_metadata.chunk_offsets_p
+            query_start_loc_p = attn_metadata.query_start_loc_p
+
+        # 1. Gated MLP's linear projection
+        projected_states, _ = self.in_proj(hidden_states)
+
+        if mup_vector is not None:
+            projected_states = projected_states * mup_vector
+
+        gate, hidden_states_B_C, dt = torch.split(
+            projected_states,
+            [
+                self.intermediate_size // self.tp_size,
+                self.conv_dim // self.tp_size,
+                self.num_heads // self.tp_size,
+            ],
+            dim=-1,
+        )
+
+        conv_weights = self.conv1d.weight.view(self.conv1d.weight.size(0),
+                                               self.conv1d.weight.size(2))
+
+        # - get hidden_states, B and C after depthwise convolution.
+        split_hidden_states_B_C_fn = lambda hidden_states_B_C: torch.split(
+            hidden_states_B_C,
+            [
+                self.intermediate_size // self.tp_size,
+                self.groups_ssm_state_size // self.tp_size,
+                self.groups_ssm_state_size // self.tp_size,
+            ],
+            dim=-1,
+        )
+
+        if attn_metadata is None:
+            # profile run
+            hidden_states_B_C = (hidden_states_B_C.transpose(
+                0, 1).clone().transpose(0, 1)).contiguous()
+            hidden_states, _B, _C = split_hidden_states_B_C_fn(
+                hidden_states_B_C)
+            hidden_states = self.norm(hidden_states, gate)
+            out, _ = self.out_proj(hidden_states)
+            return out
+
+        # NOTE: V0 put prefill before decode, v1 puts decode before prefill
+        num_prefills = attn_metadata.num_prefills  # request count
+        num_decodes = attn_metadata.num_decode_tokens  # token count (=request)
+        num_prefill_tokens = attn_metadata.num_prefill_tokens  # token count
+        has_prefill = num_prefills > 0
+        has_decode = num_decodes > 0
+        num_actual_tokens = num_prefill_tokens + num_decodes
+
+        # Separate prefill and decode by splitting varlen input
+        # Split along token dimension
+        hidden_states_B_C_d, hidden_states_B_C_p = torch.split(
+            hidden_states_B_C[:num_actual_tokens],
+            [num_decodes, num_prefill_tokens],
+            dim=0,
+        )
+        dt_d, dt_p = torch.split(
+            dt[:num_actual_tokens],
+            [num_decodes, num_prefill_tokens],
+            dim=0,
+        )
+        # Split along batch dimension
+        state_indices_tensor_d, state_indices_tensor_p = torch.split(
+            state_indices_tensor[:num_actual_tokens],
+            [num_decodes, num_prefills],
+            dim=0,
+        )
+
+        # Preallocate output tensor to avoid memcpy cost for merging prefill
+        # and decode outputs
+        preallocated_ssm_out = torch.empty(
+            [
+                num_prefill_tokens + num_decodes,
+                (self.num_heads // self.tp_size) * self.head_dim
+            ],
+            dtype=hidden_states.dtype,
+            device=hidden_states.device,
+        )
+        preallocated_ssm_out_d, preallocated_ssm_out_p = torch.split(
+            preallocated_ssm_out,
+            [num_decodes, num_prefill_tokens],
+            dim=0,
+        )
+
+        # Process prefill requests
+        if has_prefill:
+            # 2. Convolution sequence transformation
+            # - "cache_indices" updates the conv_state cache in positions
+            #   pointed to by "state_indices_tensor"
+            x = hidden_states_B_C_p.transpose(
+                0, 1)  # this is the form that causal-conv see
+            hidden_states_B_C_p = causal_conv1d_fn(
+                x,
+                conv_weights,
+                self.conv1d.bias,
+                activation=self.activation,
+                conv_states=conv_state,
+                has_initial_state=has_initial_states_p,
+                cache_indices=state_indices_tensor_p,
+                metadata=attn_metadata,
+                query_start_loc=query_start_loc_p).transpose(
+                    0, 1)[:num_prefill_tokens]
+
+            hidden_states_p, B_p, C_p = split_hidden_states_B_C_fn(
+                hidden_states_B_C_p)
+
+            # 3. State Space Model sequence transformation
+            initial_states = None
+            if (has_initial_states_p is not None and prep_initial_states):
+                initial_states = torch.where(
+                    has_initial_states_p[:, None, None, None],
+                    ssm_state[state_indices_tensor_p], 0)
+
+            # NOTE: final output is an in-place update of out tensor
+            varlen_states = mamba_chunk_scan_combined_varlen(
+                hidden_states_p.view(num_prefill_tokens,
+                                     self.num_heads // self.tp_size,
+                                     self.head_dim),
+                dt_p,
+                self.A,
+                B_p.view(num_prefill_tokens, self.n_groups // self.tp_size,
+                         -1),
+                C_p.view(num_prefill_tokens, self.n_groups // self.tp_size,
+                         -1),
+                chunk_size=chunk_size,
+                D=self.D,
+                z=None,
+                dt_bias=self.dt_bias,
+                seq_idx=seq_idx_p,
+                chunk_indices=chunk_indices_p,
+                chunk_offsets=chunk_offsets_p,
+                cu_seqlens=query_start_loc_p,
+                initial_states=initial_states,
+                dt_softplus=True,
+                dt_limit=(0.0, float("inf")),
+                out=preallocated_ssm_out_p.view(num_prefill_tokens, -1,
+                                                self.head_dim),
+                state_dtype=ssm_state.dtype)
+
+            # update ssm states
+            # - varlen state is a (num_prefills, nheads, headdim, dstate) tensor
+            ssm_state[state_indices_tensor_p] = varlen_states
+
+        # Process decode requests
+        if has_decode:
+            # 2. Convolution sequence transformation
+            hidden_states_B_C_d = causal_conv1d_update(
+                hidden_states_B_C_d,
+                conv_state,
+                conv_weights,
+                self.conv1d.bias,
+                self.activation,
+                conv_state_indices=state_indices_tensor_d)
+
+            hidden_states_d, B_d, C_d = split_hidden_states_B_C_fn(
+                hidden_states_B_C_d)
+
+            # 3. State Space Model sequence transformation
+            n_groups = self.n_groups // self.tp_size
+            A_d = self.A[:, None, ...][:, :, None].expand(
+                -1, self.head_dim, self.ssm_state_size).to(dtype=torch.float32)
+            dt_d = dt_d[:, :, None].expand(-1, -1, self.head_dim)
+            dt_bias = self.dt_bias[:, None, ...].expand(-1, self.head_dim)
+            D_d = self.D[:, None, ...].expand(-1, self.head_dim)
+            B_d = B_d.view(-1, n_groups, B_d.shape[1] // n_groups)
+            C_d = C_d.view(-1, n_groups, C_d.shape[1] // n_groups)
+            hidden_states_d = hidden_states_d.view(
+                -1, self.num_heads // self.tp_size, self.head_dim)
+
+            # - the hidden is reshaped into (bs, num_heads, head_dim)
+            # - mamba_cache_params.ssm_state's slots will be selected
+            #   using state_indices_tensor_d
+            # NOTE: final output is an in-place update of out tensor
+            selective_state_update(
+                ssm_state,
+                hidden_states_d,
+                dt_d,
+                A_d,
+                B_d,
+                C_d,
+                D_d,
+                z=None,
+                dt_bias=dt_bias,
+                dt_softplus=True,
+                state_batch_indices=state_indices_tensor_d,
+                out=preallocated_ssm_out_d.view(num_decodes, -1,
+                                                self.head_dim),
+            )
+
+        # 4. gated MLP
+        # GatedRMSNorm internally applying SiLU to the gate
+        # SiLU is applied internally before normalization, unlike standard
+        # norm usage
+        hidden_states = self.norm(preallocated_ssm_out,
+                                  gate[:num_actual_tokens])
+
+        # 5. Final linear projection
+        output[:num_actual_tokens], _ = self.out_proj(hidden_states)
+
+    def get_state_dtype(self) -> tuple[torch.dtype, torch.dtype]:
+        assert self.model_config is not None
+        assert self.cache_config is not None
+        return MambaStateDtypeCalculator.mamba2_state_dtype(
+            self.model_config.dtype,
+            self.cache_config.mamba_cache_dtype,
+            self.cache_config.mamba_ssm_cache_dtype,
+        )
+
+    def get_state_shape(self) -> tuple[tuple[int, ...], tuple[int, ...]]:
+        return MambaStateShapeCalculator.mamba2_state_shape(
+            intermediate_size=self.intermediate_size,
+            tp_world_size=get_tensor_model_parallel_world_size(),
+            n_groups=self.n_groups,
+            num_heads=self.num_heads,
+            head_dim=self.head_dim,
+            state_size=self.ssm_state_size,
+            conv_kernel=self.conv_kernel_size,
+        )
+
+    @property
+    def mamba_type(self) -> str:
+        return "mamba2"
+
+    def get_attn_backend(self) -> type["AttentionBackend"]:
+        from vllm.v1.attention.backends.mamba2_attn import (
+            Mamba2AttentionBackend)
+        return Mamba2AttentionBackend
+
+
+def mamba_mixer2(
+    hidden_states: torch.Tensor,
+    output: torch.Tensor,
+    layer_name: str,
+    mup_vector: Optional[torch.Tensor] = None,
+) -> None:
+    forward_context: ForwardContext = get_forward_context()
+    self = forward_context.no_compile_layers[layer_name]
+    self.forward_cuda(hidden_states=hidden_states,
+                      output=output,
+                      mup_vector=mup_vector)
+
+
+def mamba_mixer2_fake(
+    hidden_states: torch.Tensor,
+    output: torch.Tensor,
+    layer_name: str,
+    mup_vector: Optional[torch.Tensor] = None,
+) -> None:
+    return
+
+
+direct_register_custom_op(
+    op_name="mamba_mixer2",
+    op_func=mamba_mixer2,
+    mutates_args=["output"],
+    fake_impl=mamba_mixer2_fake,
+)

vllm/model_executor/layers/mamba/mamba_utils.py

@@ -0,0 +1,186 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+from typing import Union
+
+import torch
+
+from vllm.config import MambaDType, ModelDType
+from vllm.distributed import divide
+from vllm.utils import STR_DTYPE_TO_TORCH_DTYPE, get_kv_cache_torch_dtype
+
+
+class MambaStateDtypeCalculator:
+
+    @classmethod
+    def linear_attention_state_dtype(
+        cls,
+        model_dtype: Union[ModelDType, torch.dtype],
+        mamba_cache_dtype: MambaDType,
+    ) -> tuple[torch.dtype, ...]:
+        # TODO (tdoublep) requires testing
+        if mamba_cache_dtype == "float32":
+            raise ValueError("fp32 state for minimax is not yet supported")
+        state_dtype = get_kv_cache_torch_dtype(mamba_cache_dtype, model_dtype)
+        return (state_dtype, )
+
+    @classmethod
+    def mamba1_state_dtype(
+        cls,
+        model_dtype: Union[ModelDType, torch.dtype],
+        mamba_cache_dtype: MambaDType,
+        mamba_ssm_cache_dtype: MambaDType,
+    ) -> tuple[torch.dtype, ...]:
+        return cls._mamba_state_dtype(model_dtype, mamba_cache_dtype,
+                                      mamba_ssm_cache_dtype)
+
+    @classmethod
+    def mamba2_state_dtype(
+        cls,
+        model_dtype: Union[ModelDType, torch.dtype],
+        mamba_cache_dtype: MambaDType,
+        mamba_ssm_cache_dtype: MambaDType,
+    ) -> tuple[torch.dtype, ...]:
+        return cls._mamba_state_dtype(model_dtype, mamba_cache_dtype,
+                                      mamba_ssm_cache_dtype)
+
+    @classmethod
+    def _mamba_state_dtype(
+        cls,
+        model_dtype: Union[ModelDType, torch.dtype],
+        mamba_cache_dtype: MambaDType,
+        mamba_ssm_cache_dtype: MambaDType,
+    ) -> tuple[torch.dtype, ...]:
+        conv_state_dtype = get_kv_cache_torch_dtype(mamba_cache_dtype,
+                                                    model_dtype)
+        if mamba_ssm_cache_dtype == "auto":
+            temporal_state_dtype = conv_state_dtype
+        else:
+            temporal_state_dtype = (
+                STR_DTYPE_TO_TORCH_DTYPE[mamba_ssm_cache_dtype])
+
+        return (conv_state_dtype, temporal_state_dtype)
+
+    @classmethod
+    def short_conv_state_dtype(
+        cls,
+        model_dtype: Union[ModelDType, torch.dtype],
+        mamba_cache_dtype: MambaDType,
+    ) -> tuple[torch.dtype, ...]:
+        conv_state_dtype = get_kv_cache_torch_dtype(mamba_cache_dtype,
+                                                    model_dtype)
+        return (conv_state_dtype, )
+
+    @classmethod
+    def gated_delta_net_state_dtype(
+        cls,
+        model_dtype: Union[ModelDType, torch.dtype],
+        mamba_cache_dtype: MambaDType,
+    ) -> tuple[torch.dtype, torch.dtype]:
+        state_dtype = get_kv_cache_torch_dtype(mamba_cache_dtype, model_dtype)
+        return (state_dtype, state_dtype)
+
+
+class MambaStateShapeCalculator:
+
+    @classmethod
+    def linear_attention_state_shape(
+        cls,
+        num_heads: int,
+        tp_size: int,
+        head_dim: int,
+    ) -> tuple[tuple[int, int, int], ...]:
+
+        state_shape = (num_heads // tp_size, head_dim, head_dim)
+        return (state_shape, )
+
+    @classmethod
+    def mamba1_state_shape(
+        cls,
+        tp_world_size: int,
+        intermediate_size: int,
+        state_size: int,
+        conv_kernel: int,
+    ) -> tuple[tuple[int, int], tuple[int, int]]:
+        conv_state_shape = (divide(intermediate_size,
+                                   tp_world_size), conv_kernel - 1)
+
+        temporal_state_shape = (divide(intermediate_size,
+                                       tp_world_size), state_size)
+
+        conv_state_shape = conv_state_shape[1], conv_state_shape[0]
+
+        return conv_state_shape, temporal_state_shape
+
+    @classmethod
+    def mamba2_state_shape(
+        cls,
+        tp_world_size: int,
+        intermediate_size: int,
+        n_groups: int,
+        num_heads: int,
+        head_dim: int,
+        state_size: int,
+        conv_kernel: int,
+    ) -> tuple[tuple[int, int], tuple[int, int, int]]:
+        # if n_groups is not divisible by world_size, need to extend the shards
+        # to ensure all groups needed by a head is sharded along with it
+        n_groups = n_groups + cls.extra_groups_for_head_shards(
+            n_groups, tp_world_size)
+        # heads and n_groups are TP-ed
+        conv_dim = intermediate_size + 2 * n_groups * state_size
+
+        # contiguous along 'dim' axis
+        conv_state_shape = (conv_kernel - 1, divide(conv_dim, tp_world_size))
+
+        # These are not TP-ed as they depend on A, dt_bias, D
+        # - they are typically small
+        #   e.g., (h_heads, head_dim, state_size) = (128, 64, 128)
+        temporal_state_shape = (divide(num_heads,
+                                       tp_world_size), head_dim, state_size)
+        return conv_state_shape, temporal_state_shape
+
+    @classmethod
+    def short_conv_state_shape(
+        cls,
+        tp_world_size: int,
+        intermediate_size: int,
+        conv_kernel: int,
+    ) -> tuple[tuple[int, int]]:
+        conv_dim = divide(intermediate_size, tp_world_size)
+        conv_state_shape = (conv_kernel - 1, conv_dim)
+        return (conv_state_shape, )
+
+    @classmethod
+    def extra_groups_for_head_shards(cls, ngroups: int, tp_size: int):
+        """Compute the increase in group numbers to account for
+        replication in order to accompany the head shards."""
+
+        # in the case ngoups % tp_size == 0, this will be zero
+        if ngroups % tp_size == 0:
+            return 0
+
+        # for n_groups == 1, this is exactly tp_size - n_groups
+        return tp_size - ngroups
+
+    @classmethod
+    def gated_delta_net_state_shape(
+        cls,
+        tp_world_size: int,
+        num_k_heads: int,
+        num_v_heads: int,
+        head_k_dim: int,
+        head_v_dim: int,
+        conv_kernel_size: int,
+        num_spec: int = 0,
+    ):
+        conv_dim = (head_k_dim * num_k_heads * 2 + head_v_dim * num_v_heads)
+        conv_state_shape = (
+            divide(conv_dim, tp_world_size),
+            conv_kernel_size - 1 + num_spec,
+        )
+
+        conv_state_shape = conv_state_shape[1], conv_state_shape[0]
+
+        temporal_state_shape = (divide(num_v_heads,
+                                       tp_world_size), head_k_dim, head_v_dim)
+        return conv_state_shape, temporal_state_shape

vllm/model_executor/layers/mamba/ops/layernorm_gated.py

@@ -0,0 +1,168 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+# Copyright (c) 2024, Tri Dao.
+# Adapted from https://github.com/state-spaces/mamba/blob/60dadf2e0ee730ac337035d5533de10bc26e4847/mamba_ssm/ops/triton/layernorm_gated.py
+
+import torch
+
+from vllm.triton_utils import tl, triton
+
+
+@triton.heuristics({"HAS_BIAS": lambda args: args["B"] is not None})
+@triton.heuristics({"HAS_Z": lambda args: args["Z"] is not None})
+@triton.jit
+def _layer_norm_fwd_1pass_kernel(
+    X,  # pointer to the input
+    Y,  # pointer to the output
+    W,  # pointer to the weights
+    B,  # pointer to the biases
+    Z,  # pointer to the other branch
+    Mean,  # pointer to the mean
+    Rstd,  # pointer to the 1/std
+    stride_x_row: tl.int64,
+    stride_y_row: tl.int64,
+    stride_z_row: tl.int64,
+    M: tl.int64,  # number of rows in X
+    N: tl.int64,  # number of columns in X
+    eps,  # epsilon to avoid division by zero
+    BLOCK_N: tl.constexpr,
+    HAS_BIAS: tl.constexpr,
+    HAS_Z: tl.constexpr,
+    NORM_BEFORE_GATE: tl.constexpr,
+    IS_RMS_NORM: tl.constexpr,
+):
+    # Map the program id to the row of X and Y it should compute.
+    row = tl.program_id(0)
+    group = tl.program_id(1)
+    X += row * stride_x_row + group * N
+    Y += row * stride_y_row + group * N
+    if HAS_Z:
+        Z += row * stride_z_row + group * N
+    if not IS_RMS_NORM:
+        Mean += group * M
+    Rstd += group * M
+    W += group * N
+    if HAS_BIAS:
+        B += group * N
+    # Compute mean and variance
+    cols = tl.arange(0, BLOCK_N)
+    x = tl.load(X + cols, mask=cols < N, other=0.).to(tl.float32)
+    if HAS_Z and not NORM_BEFORE_GATE:
+        z = tl.load(Z + cols, mask=cols < N).to(tl.float32)
+        x *= z * tl.sigmoid(z)
+    if not IS_RMS_NORM:
+        mean = tl.sum(x, axis=0) / N
+        tl.store(Mean + row, mean)
+        xbar = tl.where(cols < N, x - mean, 0.)
+        var = tl.sum(xbar * xbar, axis=0) / N
+    else:
+        xbar = tl.where(cols < N, x, 0.)
+        var = tl.sum(xbar * xbar, axis=0) / N
+    rstd = 1 / tl.sqrt(var + eps)
+    tl.store(Rstd + row, rstd)
+    # Normalize and apply linear transformation
+    mask = cols < N
+    w = tl.load(W + cols, mask=mask).to(tl.float32)
+    if HAS_BIAS:
+        b = tl.load(B + cols, mask=mask).to(tl.float32)
+    x_hat = (x - mean) * rstd if not IS_RMS_NORM else x * rstd
+    y = x_hat * w + b if HAS_BIAS else x_hat * w
+    if HAS_Z and NORM_BEFORE_GATE:
+        z = tl.load(Z + cols, mask=mask).to(tl.float32)
+        y *= z * tl.sigmoid(z)
+    # Write output
+    tl.store(Y + cols, y, mask=mask)
+
+
+def _layer_norm_fwd(x,
+                    weight,
+                    bias,
+                    eps,
+                    z=None,
+                    out=None,
+                    group_size=None,
+                    norm_before_gate=True,
+                    is_rms_norm=False):
+    M, N = x.shape
+    if group_size is None:
+        group_size = N
+    assert N % group_size == 0
+    ngroups = N // group_size
+    assert x.stride(-1) == 1
+    if z is not None:
+        assert z.stride(-1) == 1
+        assert z.shape == (M, N)
+    assert weight.shape == (N, )
+    assert weight.stride(-1) == 1
+    if bias is not None:
+        assert bias.stride(-1) == 1
+        assert bias.shape == (N, )
+    # allocate output
+    if out is not None:
+        assert out.shape == x.shape
+    else:
+        out = torch.empty_like(x)
+    assert out.stride(-1) == 1
+    mean = torch.empty((ngroups * M, ), dtype=torch.float32,
+                       device=x.device) if not is_rms_norm else None
+    rstd = torch.empty((ngroups * M, ), dtype=torch.float32, device=x.device)
+    # Less than 64KB per feature: enqueue fused kernel
+    MAX_FUSED_SIZE = 65536 // x.element_size()
+    BLOCK_N = min(MAX_FUSED_SIZE, triton.next_power_of_2(group_size))
+    if group_size > BLOCK_N:
+        raise RuntimeError(
+            "This layer norm doesn't support feature dim >= 64KB.")
+    # heuristics for number of warps
+    num_warps = min(max(BLOCK_N // 256, 1), 8)
+    grid = (M, ngroups)
+    with torch.cuda.device(x.device.index):
+        _layer_norm_fwd_1pass_kernel[grid](x,
+                                           out,
+                                           weight,
+                                           bias,
+                                           z,
+                                           mean,
+                                           rstd,
+                                           x.stride(0),
+                                           out.stride(0),
+                                           z.stride(0) if z is not None else 0,
+                                           M,
+                                           group_size,
+                                           eps,
+                                           BLOCK_N=BLOCK_N,
+                                           NORM_BEFORE_GATE=norm_before_gate,
+                                           IS_RMS_NORM=is_rms_norm,
+                                           num_warps=num_warps)
+    return out, mean, rstd
+
+
+def rms_norm_gated(x,
+                   weight,
+                   bias,
+                   z=None,
+                   eps=1e-6,
+                   group_size=None,
+                   norm_before_gate=True):
+    x_shape_og = x.shape
+    # reshape input data into 2D tensor
+    x = x.reshape(-1, x.shape[-1])
+    if x.stride(-1) != 1:
+        x = x.contiguous()
+    if z is not None:
+        assert z.shape == x_shape_og
+        z = z.reshape(-1, z.shape[-1])
+        if z.stride(-1) != 1:
+            z = z.contiguous()
+    weight = weight.contiguous()
+    if bias is not None:
+        bias = bias.contiguous()
+    y, _, _ = _layer_norm_fwd(x,
+                              weight,
+                              bias,
+                              eps,
+                              z=z,
+                              group_size=group_size,
+                              norm_before_gate=norm_before_gate,
+                              is_rms_norm=True)
+
+    return y.reshape(x_shape_og)

vllm/model_executor/layers/mamba/ops/mamba_ssm.py

@@ -0,0 +1,414 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+# Copyright (c) 2024, Tri Dao, Albert Gu.
+# Adapted from https://github.com/state-spaces/mamba/blob/v2.2.4/mamba_ssm/ops/triton/selective_state_update.py
+
+import torch
+from packaging import version
+
+from vllm import _custom_ops as ops
+from vllm.attention.backends.utils import PAD_SLOT_ID
+from vllm.triton_utils import HAS_TRITON, tl, triton
+
+TRITON3 = HAS_TRITON and (version.parse(triton.__version__)
+                          >= version.parse("3.0.0"))
+
+if TRITON3:
+
+    @triton.jit
+    def softplus(dt):
+        dt = tl.where(dt <= 20.0, tl.math.log(tl.math.exp(dt) + 1), dt)
+        return dt
+else:
+
+    @triton.jit
+    def softplus(dt):
+        dt = tl.where(dt <= 20.0, tl.math.log1p(tl.exp(dt)), dt)
+        return dt
+
+
+@triton.heuristics(
+    {"HAS_DT_BIAS": lambda args: args["dt_bias_ptr"] is not None})
+@triton.heuristics({"HAS_D": lambda args: args["D_ptr"] is not None})
+@triton.heuristics({"HAS_Z": lambda args: args["z_ptr"] is not None})
+@triton.heuristics({
+    "HAS_STATE_BATCH_INDICES":
+    lambda args: args["state_batch_indices_ptr"] is not None
+})
+@triton.heuristics(
+    {"BLOCK_SIZE_DSTATE": lambda args: triton.next_power_of_2(args["dstate"])})
+@triton.jit
+def _selective_scan_update_kernel(
+    # Pointers to matrices
+    state_ptr,
+    x_ptr,
+    dt_ptr,
+    dt_bias_ptr,
+    A_ptr,
+    B_ptr,
+    C_ptr,
+    D_ptr,
+    z_ptr,
+    out_ptr,
+    state_batch_indices_ptr,
+    pad_slot_id,
+    # Matrix dimensions
+    batch,
+    nheads,
+    dim,
+    dstate,
+    nheads_ngroups_ratio,
+    # Strides
+    stride_state_batch,
+    stride_state_head,
+    stride_state_dim,
+    stride_state_dstate,
+    stride_x_batch,
+    stride_x_head,
+    stride_x_dim,
+    stride_dt_batch,
+    stride_dt_head,
+    stride_dt_dim,
+    stride_dt_bias_head,
+    stride_dt_bias_dim,
+    stride_A_head,
+    stride_A_dim,
+    stride_A_dstate,
+    stride_B_batch,
+    stride_B_group,
+    stride_B_dstate,
+    stride_C_batch,
+    stride_C_group,
+    stride_C_dstate,
+    stride_D_head,
+    stride_D_dim,
+    stride_z_batch,
+    stride_z_head,
+    stride_z_dim,
+    stride_out_batch,
+    stride_out_head,
+    stride_out_dim,
+    # Meta-parameters
+    DT_SOFTPLUS: tl.constexpr,
+    TIE_HDIM: tl.constexpr,
+    BLOCK_SIZE_M: tl.constexpr,
+    HAS_DT_BIAS: tl.constexpr,
+    HAS_D: tl.constexpr,
+    HAS_Z: tl.constexpr,
+    HAS_STATE_BATCH_INDICES: tl.constexpr,
+    BLOCK_SIZE_DSTATE: tl.constexpr,
+):
+    pid_m = tl.program_id(axis=0)
+    pid_b = tl.program_id(axis=1)
+    pid_h = tl.program_id(axis=2)
+
+    # If HAS_STATE_BATCH_INDICES is true, then the ssm state's batch coordinate
+    # is taken from the state_batch_indices_ptr Otherwise, the state coordinate
+    # is the same as the batch id.
+    if HAS_STATE_BATCH_INDICES:
+        state_batch_indices_ptr += pid_b
+        state_batch_idx = tl.load(state_batch_indices_ptr).to(tl.int64)
+        state_ptr += (state_batch_idx * stride_state_batch +
+                      pid_h * stride_state_head)
+    else:
+        state_ptr += pid_b * stride_state_batch + pid_h * stride_state_head
+
+    x_ptr += pid_b * stride_x_batch + pid_h * stride_x_head
+    dt_ptr += pid_b * stride_dt_batch + pid_h * stride_dt_head
+    if HAS_DT_BIAS:
+        dt_bias_ptr += pid_h * stride_dt_bias_head
+    A_ptr += pid_h * stride_A_head
+    B_ptr += pid_b * stride_B_batch + (pid_h //
+                                       nheads_ngroups_ratio) * stride_B_group
+    C_ptr += pid_b * stride_C_batch + (pid_h //
+                                       nheads_ngroups_ratio) * stride_C_group
+    if HAS_Z:
+        z_ptr += pid_b * stride_z_batch + pid_h * stride_z_head
+    out_ptr += pid_b * stride_out_batch + pid_h * stride_out_head
+
+    offs_m = pid_m * BLOCK_SIZE_M + tl.arange(0, BLOCK_SIZE_M)
+    offs_n = tl.arange(0, BLOCK_SIZE_DSTATE)
+    state_ptrs = state_ptr + (offs_m[:, None] * stride_state_dim +
+                              offs_n[None, :] * stride_state_dstate)
+    x_ptrs = x_ptr + offs_m * stride_x_dim
+    dt_ptrs = dt_ptr + offs_m * stride_dt_dim
+    if HAS_DT_BIAS:
+        dt_bias_ptrs = dt_bias_ptr + offs_m * stride_dt_bias_dim
+    if HAS_D:
+        D_ptr += pid_h * stride_D_head
+    A_ptrs = A_ptr + (offs_m[:, None] * stride_A_dim +
+                      offs_n[None, :] * stride_A_dstate)
+    B_ptrs = B_ptr + offs_n * stride_B_dstate
+    C_ptrs = C_ptr + offs_n * stride_C_dstate
+    if HAS_D:
+        D_ptrs = D_ptr + offs_m * stride_D_dim
+    if HAS_Z:
+        z_ptrs = z_ptr + offs_m * stride_z_dim
+    out_ptrs = out_ptr + offs_m * stride_out_dim
+    mask = (offs_m[:, None] < dim) & (offs_n[None, :] < dstate)
+    if HAS_STATE_BATCH_INDICES:
+        mask &= (state_batch_idx != pad_slot_id)
+    state = tl.load(state_ptrs, mask=mask, other=0.0)
+
+    x = tl.load(x_ptrs, mask=offs_m < dim, other=0.0).to(tl.float32)
+    if not TIE_HDIM:
+        dt = tl.load(dt_ptrs, mask=offs_m < dim, other=0.0).to(tl.float32)
+        if HAS_DT_BIAS:
+            dt += tl.load(dt_bias_ptrs, mask=offs_m < dim,
+                          other=0.0).to(tl.float32)
+        if DT_SOFTPLUS:
+            dt = softplus(dt)
+        A = tl.load(A_ptrs,
+                    mask=(offs_m[:, None] < dim) & (offs_n[None, :] < dstate),
+                    other=0.0).to(tl.float32)
+        dA = tl.exp(A * dt[:, None])
+    else:
+        dt = tl.load(dt_ptr).to(tl.float32)
+        if HAS_DT_BIAS:
+            dt += tl.load(dt_bias_ptr).to(tl.float32)
+        if DT_SOFTPLUS:
+            dt = softplus(dt)
+        A = tl.load(A_ptr).to(tl.float32)
+        dA = tl.exp(A * dt)  # scalar, not a matrix
+
+    B = tl.load(B_ptrs, mask=offs_n < dstate, other=0.0).to(tl.float32)
+    C = tl.load(C_ptrs, mask=offs_n < dstate, other=0.0).to(tl.float32)
+    if HAS_D:
+        D = tl.load(D_ptrs, mask=offs_m < dim, other=0.0).to(tl.float32)
+    if HAS_Z:
+        z = tl.load(z_ptrs, mask=offs_m < dim, other=0.0).to(tl.float32)
+
+    dB = B[None, :] * dt[:, None] if not TIE_HDIM else B * dt
+    state = state * dA + dB * x[:, None]
+
+    mask = (offs_m[:, None] < dim) & (offs_n[None, :] < dstate)
+    if HAS_STATE_BATCH_INDICES:
+        mask &= (state_batch_idx != pad_slot_id)
+    tl.store(state_ptrs, state, mask=mask)
+    out = tl.sum(state * C[None, :], axis=1)
+    if HAS_D:
+        out += x * D
+    if HAS_Z:
+        out *= z * tl.sigmoid(z)
+    tl.store(out_ptrs, out, mask=offs_m < dim)
+
+
+def selective_state_update(state,
+                           x,
+                           dt,
+                           A,
+                           B,
+                           C,
+                           D=None,
+                           z=None,
+                           dt_bias=None,
+                           dt_softplus=False,
+                           state_batch_indices=None,
+                           pad_slot_id=PAD_SLOT_ID,
+                           out=None):
+    """
+    Argument:
+        state: (batch, dim, dstate) or (batch, nheads, dim, dstate)
+        x: (batch, dim) or (batch, nheads, dim)
+        dt: (batch, dim) or (batch, nheads, dim)
+        A: (dim, dstate) or (nheads, dim, dstate)
+        B: (batch, dstate) or (batch, ngroups, dstate)
+        C: (batch, dstate) or (batch, ngroups, dstate)
+        D: (dim,) or (nheads, dim)
+        z: (batch, dim) or (batch, nheads, dim)
+        dt_bias: (dim,) or (nheads, dim)
+        pad_slot_id: int
+            if cache_indices is passed, lets the kernel identify padded 
+            entries that will not be processed, 
+            for example: cache_indices = [pad_slot_id, 1, 20, pad_slot_id] 
+            in this case, the kernel will not process entries at 
+            indices 0 and 3
+        out: Preallocated ssm output tensor. Assume same shape as x. 
+             In-place updated.
+    """
+    if state.dim() == 3:
+        state = state.unsqueeze(1)
+    if x.dim() == 2:
+        x = x.unsqueeze(1)
+    if dt.dim() == 2:
+        dt = dt.unsqueeze(1)
+    if A.dim() == 2:
+        A = A.unsqueeze(0)
+    if B.dim() == 2:
+        B = B.unsqueeze(1)
+    if C.dim() == 2:
+        C = C.unsqueeze(1)
+    if D is not None and D.dim() == 1:
+        D = D.unsqueeze(0)
+    if z is not None and z.dim() == 2:
+        z = z.unsqueeze(1)
+    if dt_bias is not None and dt_bias.dim() == 1:
+        dt_bias = dt_bias.unsqueeze(0)
+    if out.dim() == 2:
+        out = out.unsqueeze(1)
+
+    _, nheads, dim, dstate = state.shape
+    batch = x.shape[0]
+
+    assert x.shape == (batch, nheads, dim)
+    assert dt.shape == x.shape
+    assert A.shape == (nheads, dim, dstate)
+    ngroups = B.shape[1]
+    assert nheads % ngroups == 0, "nheads must be divisible by ngroups"
+    assert B.shape == (batch, ngroups, dstate)
+    assert C.shape == B.shape
+    if D is not None:
+        assert D.shape == (nheads, dim)
+    if z is not None:
+        assert z.shape == x.shape
+    if dt_bias is not None:
+        assert dt_bias.shape == (nheads, dim)
+    if state_batch_indices is not None:
+        assert state_batch_indices.shape == (batch, )
+    assert out.shape == x.shape
+
+    grid = lambda META: (triton.cdiv(dim, META['BLOCK_SIZE_M']), batch, nheads)
+    z_strides = ((z.stride(0), z.stride(1), z.stride(2)) if z is not None else
+                 (0, 0, 0))
+    # We don't want autotune since it will overwrite the state
+    # We instead tune by hand.
+    BLOCK_SIZE_M, num_warps = ((32, 4) if dstate <= 16 else
+                               ((16, 4) if dstate <= 32 else
+                                ((8, 4) if dstate <= 64 else
+                                 ((4, 4) if dstate <= 128 else ((4, 8))))))
+    tie_hdim = A.stride(-1) == 0 and A.stride(-2) == 0 and dt.stride(
+        -1) == 0 and dt_bias.stride(-1) == 0
+    with torch.cuda.device(x.device.index):
+        _selective_scan_update_kernel[grid](
+            state,
+            x,
+            dt,
+            dt_bias,
+            A,
+            B,
+            C,
+            D,
+            z,
+            out,
+            state_batch_indices,
+            pad_slot_id,
+            batch,
+            nheads,
+            dim,
+            dstate,
+            nheads // ngroups,
+            state.stride(0),
+            state.stride(1),
+            state.stride(2),
+            state.stride(3),
+            x.stride(0),
+            x.stride(1),
+            x.stride(2),
+            dt.stride(0),
+            dt.stride(1),
+            dt.stride(2),
+            *(dt_bias.stride(0),
+              dt_bias.stride(1)) if dt_bias is not None else 0,
+            A.stride(0),
+            A.stride(1),
+            A.stride(2),
+            B.stride(0),
+            B.stride(1),
+            B.stride(2),
+            C.stride(0),
+            C.stride(1),
+            C.stride(2),
+            *(D.stride(0), D.stride(1)) if D is not None else 0,
+            z_strides[0],
+            z_strides[1],
+            z_strides[2],
+            out.stride(0),
+            out.stride(1),
+            out.stride(2),
+            dt_softplus,
+            tie_hdim,
+            BLOCK_SIZE_M,
+            num_warps=num_warps,
+        )
+
+
+def selective_scan_fn(u,
+                      ssm_states,
+                      delta,
+                      A,
+                      B,
+                      C,
+                      D=None,
+                      z=None,
+                      delta_bias=None,
+                      delta_softplus=False,
+                      query_start_loc=None,
+                      cache_indices=None,
+                      has_initial_state=None,
+                      pad_slot_id=PAD_SLOT_ID) -> torch.Tensor:
+    """
+    u: (dim, total_length) for varlen or (batch, dim, seqlen) 
+        applies changes in place.
+    ssm_states: (batch, dim, dstate) or (batch, nheads, dim, dstate)
+        applies changes in place.
+    delta: (dim, total_length) for varlen or (batch, dim, seqlen)
+    A: (dim, dstate) 
+    B: (ngroups, dstate, total_length) for varlen or 
+                                        (batch,ngroups,dstate,seqlen)
+    C: (ngroups, dstate, total_length) for varlen or 
+                                        (batch,ngroups,dstate,seqlen)
+    D: (dim,) 
+    z: (dim, total_length) for varlen or (batch, dim, seqlen) 
+    dt_bias: (dim,) or (dim)
+    query_start_loc: (batch + 1) int32
+        The cumulative sequence lengths of the sequences in
+        the batch, used to index into sequence. prepended with 0.
+        for example: query_start_loc = torch.Tensor([0,10,16,17]), 
+        x.shape=(dim,17)
+    cache_indices: (batch) int32
+        A tensor with each cell is a correspondent 
+        input and output ssm_state index
+    has_initial_state: (batch) bool
+        A tensor populated with ones and zeros, 
+        indicate if the ssm_state at the corresponding index should be 
+        used as initial state. Not providing argument assumes 
+        there's no initial state
+    pad_slot_id: int
+        if cache_indices is passed, lets the kernel identify padding entries 
+        that will not be processed, 
+        for example: cache_indices = [pad_slot_id, 1 ,20 ,pad_slot_id] 
+        in this case, the kernel will not process entries at indices 0 and 3
+    returns
+        output: (dim, total_length) for varlen or (batch, dim, seqlen) 
+                supports inplace replacement
+    """
+    if u.stride(-1) != 1:
+        u = u.contiguous()
+    if delta.stride(-1) != 1:
+        delta = delta.contiguous()
+    if D is not None:
+        D = D.contiguous()
+    if B.stride(-1) != 1:
+        B = B.contiguous()
+    if C.stride(-1) != 1:
+        C = C.contiguous()
+    if z is not None and z.stride(-1) != 1:
+        z = z.contiguous()
+    if B.dim() == 3 and query_start_loc is None:
+        B = B.unsqueeze(1)
+    if B.dim() == 2 and query_start_loc is not None:
+        B = B.unsqueeze(0)
+    if C.dim() == 3 and query_start_loc is None:
+        C = C.unsqueeze(1)
+    if C.dim() == 2 and query_start_loc is not None:
+        C = C.unsqueeze(0)
+
+    ops.selective_scan_fwd(u, delta, A, B, C, D, z, delta_bias, delta_softplus,
+                           query_start_loc, cache_indices, has_initial_state,
+                           ssm_states, pad_slot_id)
+
+    if z is None:
+        return delta  # output written inplace to delta
+    else:
+        return z  # output written inplace to z

vllm/model_executor/layers/mamba/ops/ssd_bmm.py

@@ -0,0 +1,242 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+# Copyright (c) 2024, Tri Dao, Albert Gu.
+# Adapted from https://github.com/state-spaces/mamba/blob/v2.2.4/mamba_ssm/ops/triton/ssd_bmm.py
+
+# ruff: noqa: E501,SIM102
+
+import math
+
+import torch
+
+from vllm.triton_utils import tl, triton
+
+
+@triton.autotune(
+    configs=[
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 128,
+                'BLOCK_SIZE_N': 256,
+                'BLOCK_SIZE_K': 64
+            },
+            num_stages=3,
+            num_warps=8),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 64,
+                'BLOCK_SIZE_N': 256,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 128,
+                'BLOCK_SIZE_N': 128,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 128,
+                'BLOCK_SIZE_N': 64,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 64,
+                'BLOCK_SIZE_N': 128,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 128,
+                'BLOCK_SIZE_N': 32,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 64,
+                'BLOCK_SIZE_N': 32,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=5,
+            num_warps=2),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 32,
+                'BLOCK_SIZE_N': 64,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=5,
+            num_warps=2),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 64,
+                'BLOCK_SIZE_N': 64,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=2),
+    ],
+    key=['chunk_size', 'K', 'IS_CAUSAL'],
+)
+@triton.jit
+def _bmm_chunk_fwd_kernel(
+    # Pointers to matrices
+    a_ptr,
+    b_ptr,
+    out_ptr,
+    seq_idx_ptr,
+    # Matrix dimensions
+    seqlen,
+    chunk_size: tl.constexpr,
+    K: tl.constexpr,
+    ngroups: tl.constexpr,
+    stride_a_seqlen: tl.int64,
+    stride_a_head: tl.int64,
+    stride_ak: tl.constexpr,
+    stride_b_seqlen: tl.int64,
+    stride_b_head: tl.int64,
+    stride_bk: tl.constexpr,
+    stride_out_chunk: tl.int64,
+    stride_out_head: tl.int64,
+    stride_outm: tl.int64,
+    stride_outn: tl.constexpr,
+    stride_seq_idx_seqlen: tl.constexpr,
+    # Meta-parameters
+    IS_CAUSAL: tl.constexpr,
+    dot_dtype: tl.constexpr,
+    BLOCK_SIZE_M: tl.constexpr,
+    BLOCK_SIZE_N: tl.constexpr,
+    BLOCK_SIZE_K: tl.constexpr,
+):
+    pid_ch = tl.program_id(axis=1).to(tl.int64)
+    pid_c = pid_ch // ngroups
+    pid_h = pid_ch - pid_c * ngroups
+    num_pid_n = tl.cdiv(chunk_size, BLOCK_SIZE_N)
+    pid_m = tl.program_id(axis=0) // num_pid_n
+    pid_n = tl.program_id(axis=0) % num_pid_n
+    if IS_CAUSAL:
+        if pid_n * BLOCK_SIZE_N >= (pid_m + 1) * BLOCK_SIZE_M:
+            return
+    a_ptr += pid_c * chunk_size * stride_a_seqlen + pid_h * stride_a_head
+    b_ptr += pid_c * chunk_size * stride_b_seqlen + pid_h * stride_b_head
+
+    seq_idx_ptr += pid_c * chunk_size * stride_seq_idx_seqlen
+
+    offs_m = pid_m * BLOCK_SIZE_M + tl.arange(0, BLOCK_SIZE_M)
+    offs_n = pid_n * BLOCK_SIZE_N + tl.arange(0, BLOCK_SIZE_N)
+    offs_k = tl.arange(0, BLOCK_SIZE_K)
+    a_ptrs = a_ptr + (offs_m[:, None] * stride_a_seqlen +
+                      offs_k[None, :] * stride_ak)
+    b_ptrs = b_ptr + (offs_k[:, None] * stride_bk +
+                      offs_n[None, :] * stride_b_seqlen)
+    chunk_size_limit = min(chunk_size, seqlen - pid_c * chunk_size)
+
+    acc = tl.zeros((BLOCK_SIZE_M, BLOCK_SIZE_N), dtype=tl.float32)
+
+    # compute a * b.T
+    for k in range(0, tl.cdiv(K, BLOCK_SIZE_K)):
+        a = tl.load(a_ptrs,
+                    mask=(offs_m[:, None] < chunk_size_limit) &
+                    (offs_k[None, :] < K - k * BLOCK_SIZE_K),
+                    other=0.0).to(dot_dtype)
+        b = tl.load(b_ptrs,
+                    mask=(offs_k[:, None] < K - k * BLOCK_SIZE_K) &
+                    (offs_n[None, :] < chunk_size_limit),
+                    other=0.0).to(dot_dtype)
+        acc += tl.dot(a, b)
+        a_ptrs += BLOCK_SIZE_K * stride_ak
+        b_ptrs += BLOCK_SIZE_K * stride_bk
+
+    offs_m = pid_m * BLOCK_SIZE_M + tl.arange(0, BLOCK_SIZE_M)
+    offs_n = pid_n * BLOCK_SIZE_N + tl.arange(0, BLOCK_SIZE_N)
+
+    # Zero out the results that are not from the same request
+    # in the varlen batch
+    seq_idx_m = tl.load(seq_idx_ptr + offs_m * stride_seq_idx_seqlen,
+                        mask=offs_m < chunk_size_limit,
+                        other=-1)
+    seq_idx_n = tl.load(seq_idx_ptr + offs_n * stride_seq_idx_seqlen,
+                        mask=offs_n < chunk_size_limit,
+                        other=-2)
+    acc = tl.where(seq_idx_m[:, None] == seq_idx_n[None, :], acc, 0.0)
+
+    out = acc.to(out_ptr.dtype.element_ty)
+    out_ptr += pid_c * stride_out_chunk + pid_h * stride_out_head
+    out_ptrs = out_ptr + (stride_outm * offs_m[:, None] +
+                          offs_n[None, :] * stride_outn)
+    tl.store(out_ptrs,
+             out,
+             mask=(offs_m[:, None] < chunk_size) &
+             (offs_n[None, :] < chunk_size))
+
+
+def _bmm_chunk_fwd(a, b, chunk_size, seq_idx, causal=False, output_dtype=None):
+    """
+    Argument:
+        a: (seqlen, ngroups, k)
+        b: (seqlen, ngroups, k)
+        seq_idx: (seqlen,). out[i, j] for seq_idx[i] != seq_idx[j] will be zeroed out.
+        causal: if True, then out[i, j] for i > j will be arbitrary, only out[i, j] for i <= j are
+            guaranteed to be correct.
+    Return:
+        out: (nchunks, ngroups, chunk_size, chunk_size)
+    """
+    seqlen, ngroups, k = a.shape
+    assert b.shape == a.shape
+    assert seq_idx is not None
+    assert seq_idx.shape == (seqlen, )
+    if a.stride(-1) != 1 and a.stride(0) != 1:
+        a = a.contiguous()
+    if b.stride(-1) != 1 and b.stride(0) != 1:
+        b = b.contiguous()
+
+    nchunks = math.ceil(seqlen / chunk_size)
+    # Allocates output.
+    out_dtype = a.dtype if output_dtype is None else output_dtype
+    out = torch.empty((nchunks, ngroups, chunk_size, chunk_size),
+                      device=a.device,
+                      dtype=out_dtype)
+    dot_dtype = (tl.bfloat16
+                 if a.dtype == torch.bfloat16 or b.dtype == torch.bfloat16 else
+                 (tl.float16 if a.dtype == torch.float16
+                  or b.dtype == torch.float16 else tl.float32))
+    grid = lambda META: (triton.cdiv(
+        chunk_size, META['BLOCK_SIZE_M']) * triton.cdiv(
+            chunk_size, META['BLOCK_SIZE_N']), nchunks * ngroups)
+    with torch.cuda.device(a.device.index):
+        _bmm_chunk_fwd_kernel[grid](
+            a_ptr=a,
+            b_ptr=b,
+            out_ptr=out,
+            seq_idx_ptr=seq_idx,
+            seqlen=seqlen,
+            chunk_size=chunk_size,
+            K=k,
+            ngroups=ngroups,
+            stride_a_seqlen=a.stride(0),
+            stride_a_head=a.stride(1),
+            stride_ak=a.stride(2),
+            stride_b_seqlen=b.stride(0),
+            stride_b_head=b.stride(1),
+            stride_bk=b.stride(2),
+            stride_out_chunk=out.stride(0),
+            stride_out_head=out.stride(1),
+            stride_outm=out.stride(-2),
+            stride_outn=out.stride(-1),
+            stride_seq_idx_seqlen=seq_idx.stride(0),
+            IS_CAUSAL=causal,
+            dot_dtype=dot_dtype,
+        )
+    return out

vllm/model_executor/layers/mamba/ops/ssd_chunk_scan.py

@@ -0,0 +1,527 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+# Copyright (c) 2024, Tri Dao, Albert Gu.
+# Adapted from https://github.com/state-spaces/mamba/blob/v2.2.4/mamba_ssm/ops/triton/ssd_chunk_scan.py
+
+# ruff: noqa: E501,SIM102
+
+from packaging import version
+
+from vllm.triton_utils import tl, triton
+
+TRITON_22 = version.parse(triton.__version__) >= version.parse('2.2.0')
+
+
+@triton.autotune(
+    configs=[
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 128,
+                'BLOCK_SIZE_N': 256,
+                'BLOCK_SIZE_K': 64
+            },
+            num_stages=3,
+            num_warps=8),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 64,
+                'BLOCK_SIZE_N': 256,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 128,
+                'BLOCK_SIZE_N': 128,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 128,
+                'BLOCK_SIZE_N': 64,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 64,
+                'BLOCK_SIZE_N': 128,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 128,
+                'BLOCK_SIZE_N': 64,
+                'BLOCK_SIZE_K': 64
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 64,
+                'BLOCK_SIZE_N': 128,
+                'BLOCK_SIZE_K': 64
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 128,
+                'BLOCK_SIZE_N': 32,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 64,
+                'BLOCK_SIZE_N': 32,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=5,
+            num_warps=2),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 32,
+                'BLOCK_SIZE_N': 64,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=5,
+            num_warps=2),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 64,
+                'BLOCK_SIZE_N': 64,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=2),
+    ],
+    key=['chunk_size', 'hdim', 'dstate', 'IS_CAUSAL'],
+)
+@triton.jit
+def _chunk_scan_fwd_kernel(
+    # Pointers to matrices
+    cb_ptr,
+    x_ptr,
+    z_ptr,
+    out_ptr,
+    dt_ptr,
+    dA_cumsum_ptr,
+    seq_idx_ptr,
+    C_ptr,
+    states_ptr,
+    D_ptr,
+    initstates_ptr,
+    chunk_indices_ptr,
+    chunk_offsets_ptr,
+    chunk_meta_num,
+    # Matrix dimensions
+    chunk_size: tl.constexpr,
+    hdim: tl.constexpr,
+    dstate: tl.constexpr,
+    seqlen,
+    nheads_ngroups_ratio: tl.constexpr,
+    # Strides
+    stride_cb_chunk: tl.int64,
+    stride_cb_head: tl.int64,
+    stride_cb_csize_m: tl.int64,
+    stride_cb_csize_k: tl.constexpr,
+    stride_x_seqlen: tl.int64,
+    stride_x_head: tl.int64,
+    stride_x_hdim: tl.constexpr,
+    stride_z_seqlen: tl.int64,
+    stride_z_head: tl.int64,
+    stride_z_hdim: tl.constexpr,
+    stride_out_seqlen: tl.int64,
+    stride_out_head: tl.int64,
+    stride_out_hdim: tl.constexpr,
+    stride_dt_chunk: tl.int64,
+    stride_dt_head: tl.int64,
+    stride_dt_csize: tl.constexpr,
+    stride_dA_cs_chunk: tl.int64,
+    stride_dA_cs_head: tl.int64,
+    stride_dA_cs_csize: tl.constexpr,
+    stride_seq_idx_seqlen: tl.constexpr,
+    stride_C_seqlen: tl.int64,
+    stride_C_head: tl.int64,
+    stride_C_dstate: tl.constexpr,
+    stride_states_chunk: tl.int64,
+    stride_states_head: tl.int64,
+    stride_states_hdim: tl.int64,
+    stride_states_dstate: tl.constexpr,
+    stride_init_states_batch: tl.int64,
+    stride_init_states_head: tl.int64,
+    stride_init_states_hdim: tl.int64,
+    stride_init_states_dstate: tl.constexpr,
+    stride_D_head: tl.constexpr,
+    # Meta-parameters
+    IS_CAUSAL: tl.constexpr,
+    HAS_D: tl.constexpr,
+    D_HAS_HDIM: tl.constexpr,
+    HAS_Z: tl.constexpr,
+    BLOCK_SIZE_M: tl.constexpr,
+    BLOCK_SIZE_N: tl.constexpr,
+    BLOCK_SIZE_K: tl.constexpr,
+    BLOCK_SIZE_DSTATE: tl.constexpr,
+    IS_TRITON_22: tl.constexpr,
+    HAS_INITSTATES: tl.constexpr,
+):
+    pid_c = tl.program_id(axis=1).to(tl.int64)
+    if not HAS_INITSTATES:
+        c_idx = pid_c
+        c_off = 0
+    else:
+        c_idx = tl.load(chunk_indices_ptr + pid_c, mask=pid_c > -1, other=0)
+        c_off = tl.load(chunk_offsets_ptr + pid_c, mask=pid_c > -1, other=0)
+
+    pid_h = tl.program_id(axis=2)
+    num_pid_n = tl.cdiv(hdim, BLOCK_SIZE_N)
+    pid_m = tl.program_id(axis=0) // num_pid_n
+    pid_n = tl.program_id(axis=0) % num_pid_n
+    cb_ptr += c_idx * stride_cb_chunk + (pid_h //
+                                         nheads_ngroups_ratio) * stride_cb_head
+    x_ptr += c_idx * chunk_size * stride_x_seqlen + pid_h * stride_x_head
+    dt_ptr += c_idx * stride_dt_chunk + pid_h * stride_dt_head
+    dA_cumsum_ptr += c_idx * stride_dA_cs_chunk + pid_h * stride_dA_cs_head
+    C_ptr += c_idx * chunk_size * stride_C_seqlen + (
+        pid_h // nheads_ngroups_ratio) * stride_C_head
+
+    # M-block offsets and prev states
+    #  - logic in next block may override these if there is an active offset
+    offs_m = pid_m * BLOCK_SIZE_M + c_off + tl.arange(0, BLOCK_SIZE_M)
+    prev_states_ptr = states_ptr + c_idx * stride_states_chunk + pid_h * stride_states_head
+    prev_states_hdim = stride_states_hdim
+    prev_states_dstate = stride_states_dstate
+
+    chunk_size_limit = min(chunk_size, seqlen - c_idx * chunk_size)
+
+    seq_idx_ptr += c_idx * chunk_size * stride_seq_idx_seqlen
+    # - we only need seq_idx_prev to be aligned to chunk boundary
+    seq_idx_prev = tl.load(seq_idx_ptr - stride_seq_idx_seqlen,
+                           mask=c_idx >= 1,
+                           other=0)
+
+    if HAS_INITSTATES:
+        # if there are init states, we only need seq_idx_m to point
+        # what is the current seq_idx
+
+        # get current seq idx
+        if (pid_m * BLOCK_SIZE_M + c_off) < chunk_size_limit:
+            seq_idx_m = tl.load(
+                seq_idx_ptr +
+                (pid_m * BLOCK_SIZE_M + c_off) * stride_seq_idx_seqlen, )
+
+            # - recall that in ssd_state_passing, for the case c_off == 0
+            # i.e., the very first sequence, we made states_ptr hold its initial state
+            # so this edge case is taken care of
+            if ((c_off == 0) and (seq_idx_prev != seq_idx_m
+                                  )  # if a seq is changed exactly on boundary
+                    or (c_off > 0)  # implies a new example (pseudo chunk)
+                ):
+
+                # - replace prev_states_ptr with init_states
+                prev_states_ptr = initstates_ptr + seq_idx_m * stride_init_states_batch + pid_h * stride_init_states_head
+                prev_states_hdim = stride_init_states_hdim  # override strides
+                prev_states_dstate = stride_init_states_dstate
+
+    offs_n = pid_n * BLOCK_SIZE_N + tl.arange(0, BLOCK_SIZE_N)
+    dA_cs_m = tl.load(dA_cumsum_ptr + offs_m * stride_dA_cs_csize,
+                      mask=offs_m < chunk_size,
+                      other=0.0).to(tl.float32)
+
+    # - handle chunk state limit
+    if HAS_INITSTATES:
+        # have to split this if otherwise compilation will have problems
+        dA_cs_m_boundary = 0.0
+
+        # get the c_idx for the next (logica) chunk
+        c_idx_n = tl.load(
+            chunk_indices_ptr + (pid_c + 1),
+            mask=pid_c > -1 and (pid_c + 1) < chunk_meta_num,
+            other=-1  # to trigger different chunk
+        )
+
+        # - there are things to consider
+        # A. if c_off > 0 then we need to move the dA_cs boundary to ensure correct
+        #    contribution of past states
+        # B. if c_off_n < chunk_size_limit, then we need to adjust this so as not to
+        #    encroach into the next sequence, where c_off_n is the offset of the next
+        #    (logical) chunk.
+        # An equivalent check for B is c_idx == c_idx_n, where there is repetition in
+        # (logical) chunk indices.
+
+        if (c_idx == c_idx_n) or c_off > 0:
+
+            # get the next offset
+            c_off_n = tl.load(chunk_offsets_ptr + (pid_c + 1),
+                              mask=pid_c > -1 and (pid_c + 1) < chunk_meta_num,
+                              other=chunk_size)
+
+            # in this case, adjust down the chunk_size_limit
+            if c_idx == c_idx_n:
+                chunk_size_limit = min(c_off_n, chunk_size_limit)
+
+            # get the cs at the offset boundary
+            # - c_off == 0 is a passthrough
+            # - We need dA_cs at the boundary, defined by c_off - no need
+            #   to increase pointer by pid_m (it is a constant offset,
+            #   i.e. the same for all blocks)
+            dA_cs_m_boundary = tl.load(
+                dA_cumsum_ptr + (c_off - 1) * stride_dA_cs_csize,
+                mask=(((c_off - 1) > -1) and ((c_off) < chunk_size)),
+                other=0.0).to(tl.float32)
+    else:
+        # - handle seq idx when HAS_INITSTATES==False
+        seq_idx_m = tl.load(seq_idx_ptr + offs_m * stride_seq_idx_seqlen,
+                            mask=offs_m < chunk_size_limit,
+                            other=-1)
+
+    acc = tl.zeros((BLOCK_SIZE_M, BLOCK_SIZE_N), dtype=tl.float32)
+
+    # Without the if (pid_c > -1), with Triton 2.1.0, I get
+    # Assertion `!(srcMmaLayout && dstMmaLayout) && "Unexpected mma -> mm a layout conversion"' failed.
+    # With Triton 2.2.0, this works
+    if IS_TRITON_22 or c_idx > -1:
+        # Faster to just do 1 iteration with larger BLOCK_SIZE_K, up to block size 128
+        offs_k_dstate = tl.arange(
+            0, BLOCK_SIZE_DSTATE if BLOCK_SIZE_DSTATE <= 128 else BLOCK_SIZE_K)
+        C_ptrs = C_ptr + (offs_m[:, None] * stride_C_seqlen +
+                          offs_k_dstate[None, :] * stride_C_dstate)
+
+        prev_states_ptrs = prev_states_ptr + (
+            offs_n[None, :] * prev_states_hdim +
+            offs_k_dstate[:, None] * prev_states_dstate)
+
+        if not HAS_INITSTATES:
+            # - this is for continuous batching where there is no init states
+            scale_m = tl.where(seq_idx_m == seq_idx_prev, tl.exp(dA_cs_m), 0.0)
+        else:
+            # - if there is initstates, we will rely on prev_states, no zeroing
+            #   required.
+            scale_m = tl.exp(dA_cs_m - dA_cs_m_boundary)
+
+        if BLOCK_SIZE_DSTATE <= 128:
+            C = tl.load(C_ptrs,
+                        mask=(offs_m[:, None] < chunk_size_limit) &
+                        (offs_k_dstate[None, :] < dstate),
+                        other=0.0)
+
+            prev_states = tl.load(prev_states_ptrs,
+                                  mask=(offs_k_dstate[:, None] < dstate) &
+                                  (offs_n[None, :] < hdim),
+                                  other=0.0)
+            prev_states = prev_states.to(C_ptr.dtype.element_ty)
+            acc = tl.dot(C, prev_states) * scale_m[:, None]
+        else:
+            for k in range(0, dstate, BLOCK_SIZE_K):
+                C = tl.load(C_ptrs,
+                            mask=(offs_m[:, None] < chunk_size_limit) &
+                            (offs_k_dstate[None, :] < dstate - k),
+                            other=0.0)
+                # C = (C * scale_m[:, None]).to(C_ptr.dtype.element_ty)
+                prev_states = tl.load(
+                    prev_states_ptrs,
+                    mask=(offs_k_dstate[:, None] < dstate - k) &
+                    (offs_n[None, :] < hdim),
+                    other=0.0)
+                prev_states = prev_states.to(C_ptr.dtype.element_ty)
+                acc += tl.dot(C, prev_states)
+                C_ptrs += BLOCK_SIZE_K
+                prev_states_ptrs += BLOCK_SIZE_K
+            acc *= scale_m[:, None]
+
+    offs_k = tl.arange(0, BLOCK_SIZE_K) + c_off
+    cb_ptrs = cb_ptr + (offs_m[:, None] * stride_cb_csize_m +
+                        offs_k[None, :] * stride_cb_csize_k)
+    x_ptrs = x_ptr + (offs_k[:, None] * stride_x_seqlen +
+                      offs_n[None, :] * stride_x_hdim)
+    dt_ptrs = dt_ptr + offs_k * stride_dt_csize
+    dA_cumsum_ptrs = dA_cumsum_ptr + offs_k * stride_dA_cs_csize
+    K_MAX = chunk_size_limit if not IS_CAUSAL else min(
+        (pid_m + 1) * BLOCK_SIZE_M, chunk_size_limit)
+    for k in range(0, K_MAX, BLOCK_SIZE_K):
+        cb = tl.load(cb_ptrs,
+                     mask=(offs_m[:, None] < chunk_size) &
+                     (offs_k[None, :] < chunk_size - k),
+                     other=0.0).to(tl.float32)
+        dA_cs_k = tl.load(dA_cumsum_ptrs,
+                          mask=offs_k < chunk_size - k,
+                          other=0.0).to(tl.float32)
+        # If there's seq_idx, we already set cb[i, j] = 0 for seq_idx[i] != seq_idx[j].
+        # So we don't need masking wrt seq_idx here.
+        cb *= tl.exp(dA_cs_m[:, None] - dA_cs_k[None, :])
+        dt_k = tl.load(dt_ptrs, mask=offs_k < chunk_size - k,
+                       other=0.0).to(tl.float32)
+        cb *= dt_k
+        if IS_CAUSAL:
+            mask = offs_m[:, None] >= k + offs_k[None, :]
+            cb = tl.where(mask, cb, 0.0)
+        cb = cb.to(x_ptr.dtype.element_ty)
+        x = tl.load(x_ptrs,
+                    mask=(offs_k[:, None] < chunk_size_limit - k) &
+                    (offs_n[None, :] < hdim),
+                    other=0.0)
+        acc += tl.dot(cb, x)
+        cb_ptrs += BLOCK_SIZE_K * stride_cb_csize_k
+        x_ptrs += BLOCK_SIZE_K * stride_x_seqlen
+        dt_ptrs += BLOCK_SIZE_K * stride_dt_csize
+        dA_cumsum_ptrs += BLOCK_SIZE_K * stride_dA_cs_csize
+
+    offs_out_m = pid_m * BLOCK_SIZE_M + c_off + tl.arange(0, BLOCK_SIZE_M)
+    offs_out_n = pid_n * BLOCK_SIZE_N + tl.arange(0, BLOCK_SIZE_N)
+
+    if HAS_D:
+        if D_HAS_HDIM:
+            D = tl.load(D_ptr + pid_h * stride_D_head + offs_n,
+                        mask=offs_n < hdim,
+                        other=0.0).to(tl.float32)
+        else:
+            D = tl.load(D_ptr + pid_h * stride_D_head).to(tl.float32)
+        x_residual = tl.load(x_ptr + (offs_m[:, None] * stride_x_seqlen +
+                                      offs_n[None, :] * stride_x_hdim),
+                             mask=(offs_m[:, None] < chunk_size_limit) &
+                             (offs_n[None, :] < hdim),
+                             other=0.0).to(tl.float32)
+        acc += x_residual * D
+
+    if HAS_Z:
+        z_ptr += c_idx * chunk_size * stride_z_seqlen + pid_h * stride_z_head
+        z_ptrs = z_ptr + (stride_z_seqlen * offs_out_m[:, None] +
+                          stride_z_hdim * offs_out_n[None, :])
+        z = tl.load(z_ptrs,
+                    mask=(offs_out_m[:, None] < chunk_size_limit) &
+                    (offs_out_n[None, :] < hdim),
+                    other=0.0).to(tl.float32)
+        acc *= z * tl.sigmoid(z)
+
+    out_ptr += c_idx * chunk_size * stride_out_seqlen + pid_h * stride_out_head
+    out_ptrs = out_ptr + (stride_out_seqlen * offs_out_m[:, None] +
+                          offs_out_n[None, :] * stride_out_hdim)
+    tl.store(out_ptrs,
+             acc,
+             mask=(offs_out_m[:, None] < chunk_size_limit) &
+             (offs_out_n[None, :] < hdim))
+
+
+def _chunk_scan_fwd(
+    cb,
+    x,
+    dt,
+    dA_cumsum,
+    C,
+    states,
+    out,
+    seq_idx,
+    D=None,
+    z=None,
+    chunk_indices=None,
+    chunk_offsets=None,
+    initial_states=None,
+):
+    assert seq_idx is not None, "this implementation requires seq_idx"
+
+    seqlen, nheads, headdim = x.shape
+    _, nchunks, chunk_size = dt.shape
+    _, ngroups, dstate = C.shape
+    assert nheads % ngroups == 0
+    assert C.shape == (seqlen, ngroups, dstate)
+    assert cb.shape == (nchunks, ngroups, chunk_size, chunk_size)
+    if D is not None:
+        assert D.shape == (nheads, headdim) or D.shape == (nheads, )
+    if z is not None:
+        assert z.shape == x.shape
+    assert dt.shape == (nheads, nchunks, chunk_size)
+    assert dA_cumsum.shape == (nheads, nchunks, chunk_size)
+    assert states.shape == (nchunks, nheads, headdim, dstate)
+    assert seq_idx.shape == (seqlen, )
+
+    if initial_states is not None:
+        # with initial states, we need to take care of how
+        # seq_idx crosses the boundaries
+        assert chunk_indices is not None and chunk_offsets is not None, \
+            "chunk_indices and chunk_offsets should have been set"
+    else:
+        chunk_indices, chunk_offsets = None, None
+
+    grid = lambda META: (
+        triton.cdiv(chunk_size, META['BLOCK_SIZE_M']) * triton.cdiv(
+            headdim, META['BLOCK_SIZE_N']), nchunks
+        if chunk_offsets is None else len(chunk_offsets), nheads)
+
+    z_strides = ((z.stride(0), z.stride(1), z.stride(2)) if z is not None else
+                 (0, 0, 0))
+    initial_states_strides = ((initial_states.stride(0),
+                               initial_states.stride(1),
+                               initial_states.stride(2),
+                               initial_states.stride(3))
+                              if initial_states is not None else (0, 0, 0, 0))
+
+    _chunk_scan_fwd_kernel[grid](
+        cb_ptr=cb,
+        x_ptr=x,
+        z_ptr=z,
+        out_ptr=out,
+        dt_ptr=dt,
+        dA_cumsum_ptr=dA_cumsum,
+        seq_idx_ptr=seq_idx,
+        C_ptr=C,
+        states_ptr=states,
+        D_ptr=D,
+        initstates_ptr=initial_states,
+        chunk_indices_ptr=chunk_indices,
+        chunk_offsets_ptr=chunk_offsets,
+        chunk_meta_num=len(chunk_indices) if chunk_indices is not None else 0,
+        chunk_size=chunk_size,
+        hdim=headdim,
+        dstate=dstate,
+        seqlen=seqlen,
+        nheads_ngroups_ratio=nheads // ngroups,
+        stride_cb_chunk=cb.stride(0),
+        stride_cb_head=cb.stride(1),
+        stride_cb_csize_m=cb.stride(2),
+        stride_cb_csize_k=cb.stride(3),
+        stride_x_seqlen=x.stride(0),
+        stride_x_head=x.stride(1),
+        stride_x_hdim=x.stride(2),
+        stride_z_seqlen=z_strides[0],
+        stride_z_head=z_strides[1],
+        stride_z_hdim=z_strides[2],
+        stride_out_seqlen=out.stride(0),
+        stride_out_head=out.stride(1),
+        stride_out_hdim=out.stride(2),
+        stride_dt_chunk=dt.stride(1),
+        stride_dt_head=dt.stride(0),
+        stride_dt_csize=dt.stride(2),
+        stride_dA_cs_chunk=dA_cumsum.stride(1),
+        stride_dA_cs_head=dA_cumsum.stride(0),
+        stride_dA_cs_csize=dA_cumsum.stride(2),
+        stride_seq_idx_seqlen=seq_idx.stride(0),
+        stride_C_seqlen=C.stride(0),
+        stride_C_head=C.stride(1),
+        stride_C_dstate=C.stride(2),
+        stride_states_chunk=states.stride(0),
+        stride_states_head=states.stride(1),
+        stride_states_hdim=states.stride(2),
+        stride_states_dstate=states.stride(3),
+        stride_init_states_batch=initial_states_strides[0],
+        stride_init_states_head=initial_states_strides[1],
+        stride_init_states_hdim=initial_states_strides[2],
+        stride_init_states_dstate=initial_states_strides[3],
+        stride_D_head=D.stride(0) if D is not None else 0,
+        IS_CAUSAL=True,
+        HAS_D=D is not None,
+        D_HAS_HDIM=D.dim() == 2 if D is not None else True,
+        HAS_Z=z is not None,
+        BLOCK_SIZE_DSTATE=max(triton.next_power_of_2(dstate), 16),
+        IS_TRITON_22=TRITON_22,
+        HAS_INITSTATES=initial_states is not None,
+    )
+    return

vllm/model_executor/layers/mamba/ops/ssd_chunk_state.py

@@ -0,0 +1,724 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+# Copyright (c) 2024, Tri Dao, Albert Gu.
+# Adapted from https://github.com/state-spaces/mamba/blob/v2.2.4/mamba_ssm/ops/triton/ssd_chunk_state.py
+
+# ruff: noqa: E501
+
+import math
+
+import torch
+
+from vllm.triton_utils import tl, triton
+
+from .mamba_ssm import softplus
+
+
+@triton.autotune(
+    configs=[
+        triton.Config({'BLOCK_SIZE_H': 2}),
+        triton.Config({'BLOCK_SIZE_H': 4}),
+        triton.Config({'BLOCK_SIZE_H': 8}),
+        triton.Config({'BLOCK_SIZE_H': 16}),
+        triton.Config({'BLOCK_SIZE_H': 32}),
+        triton.Config({'BLOCK_SIZE_H': 64}),
+    ],
+    key=['chunk_size', 'nheads'],
+)
+@triton.jit
+def _chunk_cumsum_fwd_kernel(
+    # Pointers to matrices
+    dt_ptr,
+    A_ptr,
+    dt_bias_ptr,
+    dt_out_ptr,
+    dA_cumsum_ptr,
+    # Matrix dimension
+    seqlen,
+    nheads: tl.constexpr,
+    chunk_size: tl.constexpr,
+    dt_min: tl.constexpr,
+    dt_max: tl.constexpr,
+    # Strides
+    stride_dt_seqlen: tl.int64,
+    stride_dt_head: tl.constexpr,
+    stride_A_head: tl.constexpr,
+    stride_dt_bias_head: tl.constexpr,
+    stride_dt_out_head: tl.int64,
+    stride_dt_out_chunk: tl.int64,
+    stride_dt_out_csize: tl.constexpr,
+    stride_dA_cs_head: tl.int64,
+    stride_dA_cs_chunk: tl.int64,
+    stride_dA_cs_csize: tl.constexpr,
+    # Meta-parameters
+    DT_SOFTPLUS: tl.constexpr,
+    HAS_DT_BIAS: tl.constexpr,
+    BLOCK_SIZE_H: tl.constexpr,
+    BLOCK_SIZE_CHUNK: tl.constexpr,
+):
+    # if dt is long, may cause problems, so use 64 bit
+    # https://github.com/triton-lang/triton/issues/1058
+    pid_c = tl.program_id(axis=0).to(tl.int64)
+    pid_h = tl.program_id(axis=1)
+    dt_ptr += pid_c * chunk_size * stride_dt_seqlen
+    dt_out_ptr += pid_c * stride_dt_out_chunk
+    dA_cumsum_ptr += pid_c * stride_dA_cs_chunk
+
+    offs_h = pid_h * BLOCK_SIZE_H + tl.arange(0, BLOCK_SIZE_H)
+    offs_c = tl.arange(0, BLOCK_SIZE_CHUNK)
+    dt_ptrs = dt_ptr + (offs_h[:, None] * stride_dt_head +
+                        offs_c[None, :] * stride_dt_seqlen)
+    A_ptrs = A_ptr + offs_h * stride_A_head
+    dt_out_ptrs = dt_out_ptr + (offs_h[:, None] * stride_dt_out_head +
+                                offs_c[None, :] * stride_dt_out_csize)
+    dA_cs_ptrs = dA_cumsum_ptr + (offs_h[:, None] * stride_dA_cs_head +
+                                  offs_c[None, :] * stride_dA_cs_csize)
+    chunk_size_limit = min(chunk_size, seqlen - pid_c * chunk_size)
+
+    dt = tl.load(dt_ptrs,
+                 mask=(offs_h[:, None] < nheads) &
+                 (offs_c[None, :] < chunk_size_limit),
+                 other=0.0).to(tl.float32)
+    if HAS_DT_BIAS:
+        dt_bias = tl.load(dt_bias_ptr + offs_h * stride_dt_bias_head,
+                          mask=offs_h < nheads,
+                          other=0.0).to(tl.float32)
+        dt += dt_bias[:, None]
+    if DT_SOFTPLUS:
+        dt = tl.where(dt <= 20.0, softplus(dt), dt)
+
+    dt = tl.clamp(dt, dt_min, dt_max)
+    dt = tl.where(
+        (offs_h[:, None] < nheads) & (offs_c[None, :] < chunk_size_limit), dt,
+        0.0)
+    tl.store(dt_out_ptrs,
+             dt,
+             mask=(offs_h[:, None] < nheads) & (offs_c[None, :] < chunk_size))
+    A = tl.load(A_ptrs, mask=offs_h < nheads, other=0.0).to(tl.float32)
+    dA = dt * A[:, None]
+    dA_cs = tl.cumsum(dA, axis=1)
+    tl.store(dA_cs_ptrs,
+             dA_cs,
+             mask=(offs_h[:, None] < nheads) & (offs_c[None, :] < chunk_size))
+
+
+@triton.autotune(
+    configs=[
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 128,
+                'BLOCK_SIZE_N': 256,
+                'BLOCK_SIZE_K': 64
+            },
+            num_stages=3,
+            num_warps=8),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 64,
+                'BLOCK_SIZE_N': 256,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 128,
+                'BLOCK_SIZE_N': 128,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 128,
+                'BLOCK_SIZE_N': 64,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 64,
+                'BLOCK_SIZE_N': 128,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 128,
+                'BLOCK_SIZE_N': 32,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 64,
+                'BLOCK_SIZE_N': 32,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=5,
+            num_warps=2),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 32,
+                'BLOCK_SIZE_N': 64,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=5,
+            num_warps=2),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 64,
+                'BLOCK_SIZE_N': 64,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=2),
+    ],
+    key=['hdim', 'dstate', 'chunk_size'],
+)
+@triton.jit
+def _chunk_state_fwd_kernel(
+    # Pointers to matrices
+    x_ptr,
+    b_ptr,
+    states_ptr,
+    dt_ptr,
+    dA_cumsum_ptr,
+    seq_idx_ptr,
+    # Matrix dimensions
+    hdim: tl.constexpr,
+    dstate: tl.constexpr,
+    chunk_size: tl.constexpr,
+    seqlen,
+    nheads_ngroups_ratio: tl.constexpr,
+    # Strides
+    stride_x_seqlen: tl.int64,
+    stride_x_head: tl.int64,
+    stride_x_hdim: tl.constexpr,
+    stride_b_seqlen: tl.int64,
+    stride_b_head: tl.int64,
+    stride_b_dstate: tl.constexpr,
+    stride_states_chunk: tl.int64,
+    stride_states_head: tl.int64,
+    stride_states_hdim: tl.int64,
+    stride_states_dstate: tl.constexpr,
+    stride_dt_head: tl.int64,
+    stride_dt_chunk: tl.int64,
+    stride_dt_csize: tl.constexpr,
+    stride_dA_cs_head: tl.int64,
+    stride_dA_cs_chunk: tl.int64,
+    stride_dA_cs_csize: tl.constexpr,
+    stride_seq_idx_seqlen: tl.constexpr,
+    # Meta-parameters
+    BLOCK_SIZE_M: tl.constexpr,
+    BLOCK_SIZE_N: tl.constexpr,
+    BLOCK_SIZE_K: tl.constexpr,
+):
+    pid_c = tl.program_id(axis=1).to(tl.int64)
+    pid_h = tl.program_id(axis=2)
+    num_pid_n = tl.cdiv(dstate, BLOCK_SIZE_N)
+    pid_m = tl.program_id(axis=0) // num_pid_n
+    pid_n = tl.program_id(axis=0) % num_pid_n
+    b_ptr += pid_c * chunk_size * stride_b_seqlen + (
+        pid_h // nheads_ngroups_ratio) * stride_b_head
+    x_ptr += pid_c * chunk_size * stride_x_seqlen + pid_h * stride_x_head
+    dt_ptr += pid_c * stride_dt_chunk + pid_h * stride_dt_head
+    dA_cumsum_ptr += pid_c * stride_dA_cs_chunk + pid_h * stride_dA_cs_head
+
+    seq_idx_ptr += pid_c * chunk_size * stride_seq_idx_seqlen
+
+    offs_m = pid_m * BLOCK_SIZE_M + tl.arange(0, BLOCK_SIZE_M)
+    offs_n = pid_n * BLOCK_SIZE_N + tl.arange(0, BLOCK_SIZE_N)
+    offs_k = tl.arange(0, BLOCK_SIZE_K)
+    x_ptrs = x_ptr + (offs_m[:, None] * stride_x_hdim +
+                      offs_k[None, :] * stride_x_seqlen)
+    b_ptrs = b_ptr + (offs_n[None, :] * stride_b_dstate +
+                      offs_k[:, None] * stride_b_seqlen)
+    dt_ptrs = dt_ptr + offs_k * stride_dt_csize
+    dA_cs_last = tl.load(dA_cumsum_ptr +
+                         (chunk_size - 1) * stride_dA_cs_csize).to(tl.float32)
+    dA_cumsum_ptrs = dA_cumsum_ptr + offs_k * stride_dA_cs_csize
+
+    seq_idx_ptrs = seq_idx_ptr + offs_k * stride_seq_idx_seqlen
+    chunk_size_limit = min(chunk_size, seqlen - pid_c * chunk_size)
+    seq_idx_last = tl.load(seq_idx_ptr +
+                           (chunk_size_limit - 1) * stride_seq_idx_seqlen)
+
+    acc = tl.zeros((BLOCK_SIZE_M, BLOCK_SIZE_N), dtype=tl.float32)
+    for k in range(0, chunk_size_limit, BLOCK_SIZE_K):
+        x = tl.load(x_ptrs,
+                    mask=(offs_m[:, None] < hdim) &
+                    (offs_k[None, :] < chunk_size_limit - k),
+                    other=0.0)
+        b = tl.load(b_ptrs,
+                    mask=(offs_k[:, None] < chunk_size_limit - k) &
+                    (offs_n[None, :] < dstate),
+                    other=0.0).to(tl.float32)
+        dA_cs_k = tl.load(dA_cumsum_ptrs,
+                          mask=offs_k < chunk_size_limit - k,
+                          other=0.0).to(tl.float32)
+
+        seq_idx_k = tl.load(seq_idx_ptrs,
+                            mask=offs_k < chunk_size_limit - k,
+                            other=-1)
+        dt_k = tl.load(dt_ptrs, mask=offs_k < chunk_size_limit - k,
+                       other=0.0).to(tl.float32)
+
+        scale = tl.where(seq_idx_k == seq_idx_last,
+                         tl.exp(dA_cs_last - dA_cs_k) * dt_k, 0.0)
+        b *= scale[:, None]
+        b = b.to(x_ptr.dtype.element_ty)
+        acc += tl.dot(x, b)
+
+        x_ptrs += BLOCK_SIZE_K * stride_x_seqlen
+        b_ptrs += BLOCK_SIZE_K * stride_b_seqlen
+        dt_ptrs += BLOCK_SIZE_K * stride_dt_csize
+        dA_cumsum_ptrs += BLOCK_SIZE_K * stride_dA_cs_csize
+        seq_idx_ptrs += BLOCK_SIZE_K * stride_seq_idx_seqlen
+
+    states = acc.to(states_ptr.dtype.element_ty)
+
+    states_ptr += pid_c * stride_states_chunk + pid_h * stride_states_head
+    offs_m = pid_m * BLOCK_SIZE_M + tl.arange(0, BLOCK_SIZE_M)
+    offs_n = pid_n * BLOCK_SIZE_N + tl.arange(0, BLOCK_SIZE_N)
+    states_ptrs = states_ptr + (offs_m[:, None] * stride_states_hdim +
+                                offs_n[None, :] * stride_states_dstate)
+    c_mask = (offs_m[:, None] < hdim) & (offs_n[None, :] < dstate)
+    tl.store(states_ptrs, states, mask=c_mask)
+
+
+@triton.autotune(
+    configs=[
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 128,
+                'BLOCK_SIZE_N': 256,
+                'BLOCK_SIZE_K': 64
+            },
+            num_stages=3,
+            num_warps=8),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 64,
+                'BLOCK_SIZE_N': 256,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 128,
+                'BLOCK_SIZE_N': 128,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 128,
+                'BLOCK_SIZE_N': 64,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 64,
+                'BLOCK_SIZE_N': 128,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 128,
+                'BLOCK_SIZE_N': 32,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=4),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 64,
+                'BLOCK_SIZE_N': 32,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=5,
+            num_warps=2),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 32,
+                'BLOCK_SIZE_N': 64,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=5,
+            num_warps=2),
+        triton.Config(
+            {
+                'BLOCK_SIZE_M': 64,
+                'BLOCK_SIZE_N': 64,
+                'BLOCK_SIZE_K': 32
+            },
+            num_stages=4,
+            num_warps=2),
+    ],
+    key=['hdim', 'dstate', 'chunk_size'],
+)
+@triton.jit
+def _chunk_state_varlen_kernel(
+    # Pointers to matrices
+    x_ptr,
+    b_ptr,
+    dt_ptr,
+    dA_cumsum_ptr,
+    chunk_states_ptr,
+    cu_seqlens_ptr,
+    states_ptr,
+    initstates_ptr,
+    # Matrix dimensions
+    hdim: tl.constexpr,
+    dstate: tl.constexpr,
+    chunk_size: tl.constexpr,
+    nheads_ngroups_ratio: tl.constexpr,
+    # Strides
+    stride_x_seqlen: tl.int64,
+    stride_x_head: tl.int64,
+    stride_x_hdim: tl.constexpr,
+    stride_b_seqlen: tl.int64,
+    stride_b_head: tl.int64,
+    stride_b_dstate: tl.constexpr,
+    stride_dt_head: tl.int64,
+    stride_dt_chunk: tl.int64,
+    stride_dt_csize: tl.constexpr,
+    stride_dA_cs_head: tl.int64,
+    stride_dA_cs_chunk: tl.int64,
+    stride_dA_cs_csize: tl.constexpr,
+    stride_chunk_states_chunk: tl.int64,
+    stride_chunk_states_head: tl.int64,
+    stride_chunk_states_hdim: tl.int64,
+    stride_chunk_states_dstate: tl.constexpr,
+    stride_states_batch: tl.int64,
+    stride_states_head: tl.int64,
+    stride_states_hdim: tl.int64,
+    stride_states_dstate: tl.constexpr,
+    stride_init_states_batch: tl.int64,
+    stride_init_states_head: tl.int64,
+    stride_init_states_hdim: tl.int64,
+    stride_init_states_dstate: tl.constexpr,
+    # Meta-parameters
+    BLOCK_SIZE_M: tl.constexpr,
+    BLOCK_SIZE_N: tl.constexpr,
+    BLOCK_SIZE_K: tl.constexpr,
+    HAS_INITSTATES: tl.constexpr,
+):
+    pid_b = tl.program_id(axis=1)
+    pid_h = tl.program_id(axis=2)
+    num_pid_n = tl.cdiv(dstate, BLOCK_SIZE_N)
+    pid_m = tl.program_id(axis=0) // num_pid_n
+    pid_n = tl.program_id(axis=0) % num_pid_n
+    end_idx = tl.load(cu_seqlens_ptr + pid_b + 1)
+    pid_c = (end_idx - 1) // chunk_size
+    b_ptr += pid_c * chunk_size * stride_b_seqlen + (
+        pid_h // nheads_ngroups_ratio) * stride_b_head
+    x_ptr += pid_c * chunk_size * stride_x_seqlen + pid_h * stride_x_head
+    dt_ptr += pid_c * stride_dt_chunk + pid_h * stride_dt_head
+    dA_cumsum_ptr += pid_c * stride_dA_cs_chunk + pid_h * stride_dA_cs_head
+    chunk_states_ptr += pid_c * stride_chunk_states_chunk + pid_h * stride_chunk_states_head
+
+    if HAS_INITSTATES:
+        # if there are init states provided, we differentiate between states (which
+        # are boundary conditions at a chunk boundary) and initstates (which are boundary
+        # conditions when a new example in a cont batch starts)
+        initstates_ptr += pid_h * stride_init_states_head
+
+    offs_m = pid_m * BLOCK_SIZE_M + tl.arange(0, BLOCK_SIZE_M)
+    offs_n = pid_n * BLOCK_SIZE_N + tl.arange(0, BLOCK_SIZE_N)
+    offs_k = tl.arange(0, BLOCK_SIZE_K)
+    x_ptrs = x_ptr + (offs_m[:, None] * stride_x_hdim +
+                      offs_k[None, :] * stride_x_seqlen)
+    b_ptrs = b_ptr + (offs_n[None, :] * stride_b_dstate +
+                      offs_k[:, None] * stride_b_seqlen)
+    dt_ptrs = dt_ptr + offs_k * stride_dt_csize
+    dA_cs_last = tl.load(dA_cumsum_ptr + (end_idx - pid_c * chunk_size - 1) *
+                         stride_dA_cs_csize).to(tl.float32)
+    dA_cumsum_ptrs = dA_cumsum_ptr + offs_k * stride_dA_cs_csize
+
+    chunk_size_limit = end_idx - pid_c * chunk_size
+    start_idx = tl.load(cu_seqlens_ptr + pid_b)
+    start_idx_cur = tl.maximum(start_idx - pid_c * chunk_size, 0)
+
+    acc = tl.zeros((BLOCK_SIZE_M, BLOCK_SIZE_N), dtype=tl.float32)
+    for k in range(0, chunk_size_limit, BLOCK_SIZE_K):
+        x = tl.load(x_ptrs,
+                    mask=(offs_m[:, None] < hdim) &
+                    (offs_k[None, :] < chunk_size_limit - k) &
+                    (offs_k[None, :] >= start_idx_cur - k),
+                    other=0.0)
+        b = tl.load(b_ptrs,
+                    mask=(offs_k[:, None] < chunk_size_limit - k) &
+                    (offs_n[None, :] < dstate) &
+                    (offs_k[:, None] >= start_idx_cur - k),
+                    other=0.0).to(tl.float32)
+        dA_cs_k = tl.load(dA_cumsum_ptrs,
+                          mask=offs_k < chunk_size_limit - k,
+                          other=0.0).to(tl.float32)
+        dt_k = tl.load(dt_ptrs, mask=offs_k < chunk_size_limit - k,
+                       other=0.0).to(tl.float32)
+        scale = tl.where(
+            (offs_k >= start_idx_cur - k) & (offs_k < chunk_size_limit - k),
+            tl.exp(dA_cs_last - dA_cs_k) * dt_k, 0.0)
+        b *= scale[:, None]
+        b = b.to(x_ptr.dtype.element_ty)
+        acc += tl.dot(x, b)
+        x_ptrs += BLOCK_SIZE_K * stride_x_seqlen
+        b_ptrs += BLOCK_SIZE_K * stride_b_seqlen
+        dt_ptrs += BLOCK_SIZE_K * stride_dt_csize
+        dA_cumsum_ptrs += BLOCK_SIZE_K * stride_dA_cs_csize
+
+    # If the sequence starts after the last chunk idx, we don't need to add the contribution from the last chunk
+    # If HAS_INITSTATES==True need to consider two possibilities
+    # - if start_idx < pid_c * chunk_size, then we need to take the past_states_ptrs
+    # - if state_idx >= pid * chunk_size, then we need to insert initstates
+    if ((start_idx < pid_c * chunk_size)  # first chunk
+            or (HAS_INITSTATES)):
+
+        dA_cs_boundary = 0.0  # default
+
+        if not HAS_INITSTATES:
+            past_states_ptrs = chunk_states_ptr + (
+                offs_m[:, None] * stride_chunk_states_hdim +
+                offs_n[None, :] * stride_chunk_states_dstate)
+        else:
+
+            # - this seems repetitive, buts its to help the compiler
+            if start_idx < pid_c * chunk_size:
+                past_states_ptrs = chunk_states_ptr + (
+                    offs_m[:, None] * stride_chunk_states_hdim +
+                    offs_n[None, :] * stride_chunk_states_dstate)
+            else:
+                past_states_ptrs = initstates_ptr + (
+                    pid_b * stride_init_states_batch +
+                    offs_m[:, None] * stride_init_states_hdim +
+                    offs_n[None, :] * stride_init_states_dstate)
+
+                # need to adjust the boundary
+                if start_idx > pid_c * chunk_size:
+                    dA_cs_boundary = tl.load(dA_cumsum_ptr +
+                                             (start_idx - pid_c * chunk_size -
+                                              1) * stride_dA_cs_csize).to(
+                                                  tl.float32)
+
+        past_states = tl.load(past_states_ptrs,
+                              mask=(offs_m[:, None] < hdim) &
+                              (offs_n[None, :] < dstate),
+                              other=0.0).to(tl.float32)
+
+        scale = tl.exp(dA_cs_last - dA_cs_boundary)
+        acc += past_states * scale
+
+    states = acc.to(states_ptr.dtype.element_ty)
+
+    states_ptr += pid_b * stride_states_batch + pid_h * stride_states_head
+    offs_m = pid_m * BLOCK_SIZE_M + tl.arange(0, BLOCK_SIZE_M)
+    offs_n = pid_n * BLOCK_SIZE_N + tl.arange(0, BLOCK_SIZE_N)
+    states_ptrs = states_ptr + (offs_m[:, None] * stride_states_hdim +
+                                offs_n[None, :] * stride_states_dstate)
+    c_mask = (offs_m[:, None] < hdim) & (offs_n[None, :] < dstate)
+    tl.store(states_ptrs, states, mask=c_mask)
+
+
+def _chunk_cumsum_fwd(dt,
+                      A,
+                      chunk_size,
+                      dt_bias=None,
+                      dt_softplus=False,
+                      dt_limit=(0.0, float("inf"))):
+    seqlen, nheads = dt.shape
+    assert A.shape == (nheads, )
+    if dt_bias is not None:
+        assert dt_bias.shape == (nheads, )
+    nchunks = math.ceil(seqlen / chunk_size)
+    dt_out = torch.empty(nheads,
+                         nchunks,
+                         chunk_size,
+                         device=dt.device,
+                         dtype=torch.float32)
+    dA_cumsum = torch.empty(nheads,
+                            nchunks,
+                            chunk_size,
+                            device=dt.device,
+                            dtype=torch.float32)
+    grid_chunk_cs = lambda META: (nchunks,
+                                  triton.cdiv(nheads, META['BLOCK_SIZE_H']))
+    with torch.cuda.device(dt.device.index):
+        _chunk_cumsum_fwd_kernel[grid_chunk_cs](
+            dt_ptr=dt,
+            A_ptr=A,
+            dt_bias_ptr=dt_bias,
+            dt_out_ptr=dt_out,
+            dA_cumsum_ptr=dA_cumsum,
+            seqlen=seqlen,
+            nheads=nheads,
+            chunk_size=chunk_size,
+            dt_min=dt_limit[0],
+            dt_max=dt_limit[1],
+            stride_dt_seqlen=dt.stride(0),
+            stride_dt_head=dt.stride(1),
+            stride_A_head=A.stride(0),
+            stride_dt_bias_head=dt_bias.stride(0)
+            if dt_bias is not None else 0,
+            stride_dt_out_head=dt_out.stride(0),
+            stride_dt_out_chunk=dt_out.stride(1),
+            stride_dt_out_csize=dt_out.stride(2),
+            stride_dA_cs_head=dA_cumsum.stride(0),
+            stride_dA_cs_chunk=dA_cumsum.stride(1),
+            stride_dA_cs_csize=dA_cumsum.stride(2),
+            DT_SOFTPLUS=dt_softplus,
+            HAS_DT_BIAS=dt_bias is not None,
+            BLOCK_SIZE_CHUNK=triton.next_power_of_2(chunk_size),
+        )
+    return dA_cumsum, dt_out
+
+
+def _chunk_state_fwd(B,
+                     x,
+                     dt,
+                     dA_cumsum,
+                     seq_idx=None,
+                     states=None,
+                     states_in_fp32=True):
+    seqlen, nheads, headdim = x.shape
+    _, nchunks, chunk_size = dt.shape
+    _, ngroups, dstate = B.shape
+    assert nheads % ngroups == 0
+    assert B.shape == (seqlen, ngroups, dstate)
+    assert dt.shape == (nheads, nchunks, chunk_size)
+    assert dA_cumsum.shape == dt.shape
+
+    assert seq_idx is not None
+    assert seq_idx.shape == (seqlen, )
+
+    if states is not None:
+        assert states.shape == (nchunks, nheads, headdim, dstate)
+    else:
+        states_dtype = torch.float32 if states_in_fp32 else B.dtype
+        states = torch.empty((nchunks, nheads, headdim, dstate),
+                             device=x.device,
+                             dtype=states_dtype)
+
+    grid = lambda META: (triton.cdiv(headdim, META['BLOCK_SIZE_M']) * triton.
+                         cdiv(dstate, META['BLOCK_SIZE_N']), nchunks, nheads)
+    with torch.cuda.device(x.device.index):
+        _chunk_state_fwd_kernel[grid](
+            x_ptr=x,
+            b_ptr=B,
+            states_ptr=states,
+            dt_ptr=dt,
+            dA_cumsum_ptr=dA_cumsum,
+            seq_idx_ptr=seq_idx,
+            hdim=headdim,
+            dstate=dstate,
+            chunk_size=chunk_size,
+            seqlen=seqlen,
+            nheads_ngroups_ratio=nheads // ngroups,
+            stride_x_seqlen=x.stride(0),
+            stride_x_head=x.stride(1),
+            stride_x_hdim=x.stride(2),
+            stride_b_seqlen=B.stride(0),
+            stride_b_head=B.stride(1),
+            stride_b_dstate=B.stride(2),
+            stride_states_chunk=states.stride(0),
+            stride_states_head=states.stride(1),
+            stride_states_hdim=states.stride(2),
+            stride_states_dstate=states.stride(3),
+            stride_dt_head=dt.stride(0),
+            stride_dt_chunk=dt.stride(1),
+            stride_dt_csize=dt.stride(2),
+            stride_dA_cs_head=dA_cumsum.stride(0),
+            stride_dA_cs_chunk=dA_cumsum.stride(1),
+            stride_dA_cs_csize=dA_cumsum.stride(2),
+            stride_seq_idx_seqlen=seq_idx.stride(0),
+        )
+    return states
+
+
+def chunk_state_varlen(B,
+                       x,
+                       dt,
+                       dA_cumsum,
+                       cu_seqlens,
+                       chunk_states,
+                       initial_states=None):
+    total_seqlen, nheads, headdim = x.shape
+    _, nchunks, chunk_size = dt.shape
+    _, ngroups, dstate = B.shape
+    batch = cu_seqlens.shape[0] - 1
+    cu_seqlens = cu_seqlens.contiguous()
+    assert nheads % ngroups == 0
+    assert B.shape == (total_seqlen, ngroups, dstate)
+    assert dt.shape == (nheads, nchunks, chunk_size)
+    assert dA_cumsum.shape == dt.shape
+    assert chunk_states.shape == (nchunks, nheads, headdim, dstate)
+
+    if initial_states is not None:
+        assert initial_states.shape == (batch, nheads, headdim, dstate)
+
+    states = torch.empty(batch,
+                         nheads,
+                         headdim,
+                         dstate,
+                         dtype=chunk_states.dtype,
+                         device=chunk_states.device)
+
+    initial_states_strides = ((initial_states.stride(0),
+                               initial_states.stride(1),
+                               initial_states.stride(2),
+                               initial_states.stride(3))
+                              if initial_states is not None else (0, 0, 0, 0))
+
+    grid = lambda META: (triton.cdiv(headdim, META['BLOCK_SIZE_M']) * triton.
+                         cdiv(dstate, META['BLOCK_SIZE_N']), batch, nheads)
+    with torch.cuda.device(x.device.index):
+        _chunk_state_varlen_kernel[grid](
+            x_ptr=x,
+            b_ptr=B,
+            dt_ptr=dt,
+            dA_cumsum_ptr=dA_cumsum,
+            chunk_states_ptr=chunk_states,
+            cu_seqlens_ptr=cu_seqlens,
+            states_ptr=states,
+            initstates_ptr=initial_states,
+            hdim=headdim,
+            dstate=dstate,
+            chunk_size=chunk_size,
+            nheads_ngroups_ratio=nheads // ngroups,
+            stride_x_seqlen=x.stride(0),
+            stride_x_head=x.stride(1),
+            stride_x_hdim=x.stride(2),
+            stride_b_seqlen=B.stride(0),
+            stride_b_head=B.stride(1),
+            stride_b_dstate=B.stride(2),
+            stride_dt_head=dt.stride(0),
+            stride_dt_chunk=dt.stride(1),
+            stride_dt_csize=dt.stride(2),
+            stride_dA_cs_head=dA_cumsum.stride(0),
+            stride_dA_cs_chunk=dA_cumsum.stride(1),
+            stride_dA_cs_csize=dA_cumsum.stride(2),
+            stride_chunk_states_chunk=chunk_states.stride(0),
+            stride_chunk_states_head=chunk_states.stride(1),
+            stride_chunk_states_hdim=chunk_states.stride(2),
+            stride_chunk_states_dstate=chunk_states.stride(3),
+            stride_states_batch=states.stride(0),
+            stride_states_head=states.stride(1),
+            stride_states_hdim=states.stride(2),
+            stride_states_dstate=states.stride(3),
+            stride_init_states_batch=initial_states_strides[0],
+            stride_init_states_head=initial_states_strides[1],
+            stride_init_states_hdim=initial_states_strides[2],
+            stride_init_states_dstate=initial_states_strides[3],
+            HAS_INITSTATES=initial_states is not None)
+    return states

vllm/model_executor/layers/mamba/ops/ssd_combined.py

@@ -0,0 +1,238 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+# Copyright (c) 2024, Tri Dao, Albert Gu.
+# Adapted from https://github.com/state-spaces/mamba/blob/v2.2.4/mamba_ssm/ops/triton/ssd_combined.py
+
+# ruff: noqa: E501
+
+import torch
+from einops import rearrange
+from packaging import version
+
+from vllm.triton_utils import triton
+
+from .ssd_bmm import _bmm_chunk_fwd
+from .ssd_chunk_scan import _chunk_scan_fwd
+from .ssd_chunk_state import (_chunk_cumsum_fwd, _chunk_state_fwd,
+                              chunk_state_varlen)
+from .ssd_state_passing import _state_passing_fwd
+
+TRITON_22 = version.parse(triton.__version__) >= version.parse('2.2.0')
+
+
+def is_int_pow_2(n):
+    return isinstance(n, int) and n > 0 and (n & (n - 1)) == 0
+
+
+def _mamba_chunk_scan_combined_fwd(x,
+                                   dt,
+                                   A,
+                                   B,
+                                   C,
+                                   chunk_size,
+                                   out,
+                                   D=None,
+                                   z=None,
+                                   dt_bias=None,
+                                   initial_states=None,
+                                   seq_idx=None,
+                                   chunk_indices=None,
+                                   chunk_offsets=None,
+                                   cu_seqlens=None,
+                                   dt_softplus=False,
+                                   dt_limit=(0.0, float("inf")),
+                                   state_dtype=None):
+    assert is_int_pow_2(chunk_size), "chunk_size must be integer power of 2"
+    seqlen, nheads, headdim = x.shape
+    _, ngroups, dstate = B.shape
+    assert nheads % ngroups == 0
+    assert B.shape == (seqlen, ngroups, dstate)
+    assert dt.shape == (seqlen, nheads)
+    assert A.shape == (nheads, )
+    assert C.shape == B.shape
+    if z is not None:
+        assert z.shape == x.shape
+    if D is not None:
+        assert D.shape == (nheads, headdim) or D.shape == (nheads, )
+    if seq_idx is not None:
+        assert seq_idx.shape == (seqlen, )
+    if B.stride(-1) != 1:
+        B = B.contiguous()
+    if C.stride(-1) != 1:
+        C = C.contiguous()
+    if x.stride(-1) != 1 and x.stride(
+            0) != 1:  # Either M or K dimension should be contiguous
+        x = x.contiguous()
+    if z is not None and z.stride(-1) != 1 and z.stride(
+            0) != 1:  # Either M or K dimension should be contiguous
+        z = z.contiguous()
+    if D is not None and D.stride(-1) != 1:
+        D = D.contiguous()
+    assert cu_seqlens is not None, "Assuming varlen input - must supply cu_seqlens"
+
+    if initial_states is not None:
+        assert initial_states.shape == (len(cu_seqlens) - 1, nheads, headdim,
+                                        dstate)
+
+    # This function executes 5 sub-functions for computing mamba
+    # - a good resource is the blog https://goombalab.github.io/blog/2024/mamba2-part3-algorithm/
+    #   which has a minimal implementation to understand the below operations
+    # - as explained by the blog, mamba is a special case of causal attention
+    # - the idea is to chunk the attention matrix and compute each
+    #   submatrix separately using different optimizations.
+    # - see the blog and paper for a visualization of the submatrices
+    #   which we refer to in the comments below
+
+    # 1. Compute chunked cumsum of A * dt
+    # - here dt may go through a softplus activation
+    dA_cumsum, dt = _chunk_cumsum_fwd(dt,
+                                      A,
+                                      chunk_size,
+                                      dt_bias=dt_bias,
+                                      dt_softplus=dt_softplus,
+                                      dt_limit=dt_limit)
+
+    # 2. Compute the state for each intra-chunk
+    # (right term of low-rank factorization of off-diagonal blocks; B terms)
+    states = _chunk_state_fwd(B,
+                              x,
+                              dt,
+                              dA_cumsum,
+                              seq_idx=seq_idx,
+                              states_in_fp32=True)
+
+    # 3. Compute the inter-chunk SSM recurrence; produces correct SSM states at chunk boundaries
+    # (middle term of factorization of off-diag blocks; A terms)
+    # - for handling chunked prefill, this requires i) initial_states
+    #   ii) seq_idx iii) is_cont_batched and (iv) chunk_offsets to be all specified.
+    # - When a new seq_idx is detected, we will stop passing the prev_state
+    #   and switch accordingly to the init_state corresponding to the new seq_idx.
+    # - We will also make sure that the dA_cumsum is taken only from the start of the
+    #   sequence (hence we need the full dA_cumsum tensor and not just the values at chunk boundaries)
+    # - this will ensure that states will be updated with the rightmost flushed seq_idx
+    #   of the previous chunk. This implies that the first chunk of states is either 0
+    #   or equal to init_states of the first example.
+    states = _state_passing_fwd(
+        rearrange(states, "... p n -> ... (p n)"),
+        dA_cumsum,  # (nheads, nchunks, chunk_size)
+        initial_states=rearrange(initial_states, "... p n -> ... (p n)")
+        if initial_states is not None else
+        None,  # (batch, nheads, headdim*dstate)
+        seq_idx=seq_idx,
+        out_dtype=state_dtype if state_dtype is not None else C.dtype,
+        chunk_offsets=chunk_offsets)
+    states = rearrange(states, "... (p n) -> ... p n", n=dstate)
+
+    # 4. Compute batched matrix multiply for C_j^T B_i terms
+    CB = _bmm_chunk_fwd(C,
+                        B,
+                        chunk_size,
+                        seq_idx=seq_idx,
+                        output_dtype=torch.float32)
+
+    # 5. Scan and compute the diagonal blocks, taking into
+    #    account past causal states.
+    # - if initial states are provided, then states information will be
+    #   augmented with initial_states.
+    # - to do this properly, we need to account for example changes in
+    #   the continuous batch, therefore we introduce pseudo chunks, which is
+    #   a chunk that is split up each time an example changes.
+    # - in each (pseudo) chunk, we detect if the previous (pseudo) chunk had
+    #   a seq_idx change, in which case we take states information from
+    #   init_states.
+    _chunk_scan_fwd(
+        CB,
+        x,
+        dt,
+        dA_cumsum,
+        C,
+        states,
+        out,  # in-place update
+        seq_idx,
+        D=D,
+        z=z,
+        chunk_indices=chunk_indices,
+        chunk_offsets=chunk_offsets,
+        initial_states=initial_states,
+    )
+
+    varlen_states = chunk_state_varlen(
+        B,
+        x,
+        dt,
+        dA_cumsum,
+        cu_seqlens,
+        states,
+        initial_states=initial_states,
+    )
+
+    return varlen_states
+
+
+def mamba_chunk_scan_combined_varlen(
+        x,
+        dt,
+        A,
+        B,
+        C,
+        chunk_size,
+        cu_seqlens,
+        seq_idx,
+        out,
+        D=None,
+        z=None,
+        dt_bias=None,
+        initial_states=None,
+        chunk_indices=None,
+        chunk_offsets=None,
+        dt_softplus=False,
+        dt_limit=(0.0, float("inf")),
+        state_dtype=None,
+):
+    """
+    Argument:
+        x: (seqlen, nheads, headdim)
+        dt: (seqlen, nheads)
+        A: (nheads)
+        B: (seqlen, ngroups, dstate)
+        C: (seqlen, ngroups, dstate)
+        chunk_size: int
+        seq_idx: (seqlen)
+        cu_seqlens: (batch + 1)
+        out: (seqlen, nheads, headdim) preallocated output tensor
+        D: (nheads, headdim) or (nheads,)
+        z: (seqlen, nheads, headdim)
+        dt_bias: (nheads,)
+        initial_states: (batch, nheads, headdim, dstate)
+        dt_softplus: Whether to apply softplus to dt
+        out: (seqlen, nheads, headdim) preallocated output tensor
+        state_dtype: The data type of the ssm state
+    Return:
+        varlen_states: (batch, nheads, headdim, dstate)
+    """
+
+    assert cu_seqlens is not None, "cu_seqlens must be provided assuming varlen input"
+    assert seq_idx is not None
+
+    varlen_states = _mamba_chunk_scan_combined_fwd(
+        x,
+        dt,
+        A,
+        B,
+        C,
+        chunk_size,
+        out,
+        D=D,
+        z=z,
+        dt_bias=dt_bias,
+        initial_states=initial_states,
+        seq_idx=seq_idx,
+        chunk_indices=chunk_indices,
+        chunk_offsets=chunk_offsets,
+        cu_seqlens=cu_seqlens,
+        dt_softplus=dt_softplus,
+        dt_limit=dt_limit,
+        state_dtype=state_dtype)
+
+    return varlen_states

vllm/model_executor/layers/mamba/ops/ssd_state_passing.py

@@ -0,0 +1,200 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+# Copyright (c) 2024, Tri Dao, Albert Gu.
+# Adapted from https://github.com/state-spaces/mamba/blob/v2.2.4/mamba_ssm/ops/triton/ssd_state_passing.py
+
+# ruff: noqa: E501
+
+import torch
+
+from vllm.triton_utils import tl, triton
+
+
+@triton.autotune(
+    configs=[
+        triton.Config({'BLOCK_SIZE': 64}),
+        triton.Config({'BLOCK_SIZE': 128}),
+        triton.Config({'BLOCK_SIZE': 256}),
+        triton.Config({'BLOCK_SIZE': 512}),
+        triton.Config({'BLOCK_SIZE': 1024}),
+        triton.Config({'BLOCK_SIZE': 2048}),
+    ],
+    key=['dim'],
+)
+@triton.jit
+def _state_passing_fwd_kernel(
+    # Pointers to matrices
+    states_ptr,
+    out_ptr,
+    dA_cs_ptr,
+    initstates_ptr,
+    seq_idx_ptr,
+    chunk_offsets_ptr,
+    chunk_meta_num,
+    # Matrix dimensions
+    dim: tl.constexpr,
+    nchunks,
+    seqlen,
+    chunk_size: tl.constexpr,
+    # Strides
+    stride_states_chunk: tl.int64,
+    stride_states_head: tl.int64,
+    stride_states_dim: tl.constexpr,
+    stride_out_chunk: tl.int64,
+    stride_out_head: tl.int64,
+    stride_out_dim: tl.constexpr,
+    stride_dA_cs_head: tl.int64,
+    stride_dA_cs_chunk: tl.int64,
+    stride_dA_cs_csize: tl.constexpr,
+    stride_initstates_batch: tl.int64,
+    stride_initstates_head: tl.int64,
+    stride_initstates_dim: tl.constexpr,
+    stride_seq_idx_seqlen: tl.constexpr,
+    # Meta-parameters
+    HAS_INITSTATES: tl.constexpr,
+    BLOCK_SIZE: tl.constexpr,
+):
+    pid_h = tl.program_id(axis=1)
+    pid_m = tl.program_id(axis=0)
+    states_ptr += pid_h * stride_states_head
+    dA_cs_ptr += pid_h * stride_dA_cs_head + (chunk_size -
+                                              1) * stride_dA_cs_csize
+    out_ptr += pid_h * stride_out_head
+    if HAS_INITSTATES:
+        initstates_ptr += pid_h * stride_initstates_head
+
+    offs_m = pid_m * BLOCK_SIZE + tl.arange(0, BLOCK_SIZE)
+    states_ptrs = states_ptr + offs_m * stride_states_dim
+    out_ptrs = out_ptr + offs_m * stride_out_dim
+
+    # - states will be the past state of the sequence that continues on the current check
+    if not HAS_INITSTATES:
+        states = tl.zeros((BLOCK_SIZE, ), dtype=tl.float32)
+    else:
+        initstates_ptr += offs_m * stride_initstates_dim
+        initstates_ptrs = initstates_ptr
+        # - for cont batches, for the first chunk mean it will be the first batch's
+        #   init state
+        states = tl.load(initstates_ptrs, mask=offs_m < dim,
+                         other=0.0).to(tl.float32)
+
+    tl.store(out_ptrs, states, mask=offs_m < dim)
+    out_ptrs += stride_out_chunk
+    prev_seq_idx_chunk_end = 0
+    logical_chunk_idx = 0
+    for c in range(nchunks - 1):
+        new_states = tl.load(states_ptrs, mask=offs_m < dim,
+                             other=0.0).to(tl.float32)
+        dA_cs = tl.load(dA_cs_ptr).to(tl.float32)
+        scale_mask = True
+        # - the seq to pass forward is the one that is flushed to the right
+        #   boundary.
+        # - that is given by seq_idx_chunk_end below: the sequence index at the end of the chunk.
+        seq_idx_chunk_end = tl.load(seq_idx_ptr +
+                                    (min((c + 1) * chunk_size, seqlen) - 1) *
+                                    stride_seq_idx_seqlen)
+
+        if HAS_INITSTATES:
+            if prev_seq_idx_chunk_end != seq_idx_chunk_end:
+                # this means in the current chunk the rightmost flushed seq
+                # has changed.
+                # - so we do not propagate the state from previous chunk
+                # - but rather we load that sequence's init state
+                initstates_ptrs = initstates_ptr + seq_idx_chunk_end * stride_initstates_batch
+
+                # - update state with seq_idx_new's init state
+                states = tl.load(initstates_ptrs, mask=offs_m < dim,
+                                 other=0.0).to(tl.float32)
+
+                # - we need to consider the cumsum only of the last sequence in the chunk
+                # - find its starting position (given by c_off of the logical chunk index)
+                # - and subtract the cumsum just before that position from the total cumsum
+                # - first, update the logical chunk index (add the number of sequences in the current physical chunk):
+                # sequence index at the start of the current chunk
+                seq_idx_chunk_start = tl.load(seq_idx_ptr +
+                                              min(c * chunk_size, seqlen) *
+                                              stride_seq_idx_seqlen)
+                logical_chunk_idx += seq_idx_chunk_end - seq_idx_chunk_start
+                # - load the chunk offset:
+                c_off = tl.load(chunk_offsets_ptr + logical_chunk_idx,
+                                mask=logical_chunk_idx < chunk_meta_num,
+                                other=0)
+                # - if offset is 0, then the sequence starts at the beginning of the chunk, and we don't need to subtract anything
+                if c_off > 0:
+                    # - dA_cs_ptr currently points to the cumsum at the end of the chunk - subtract the chunk size and add the offset
+                    dA_cs_boundary = tl.load(
+                        dA_cs_ptr - (chunk_size - 1) * stride_dA_cs_csize +
+                        (c_off - 1) * stride_dA_cs_csize,
+                        mask=(c_off - 1) > -1 and c_off < chunk_size,
+                        other=0.0)
+                    dA_cs -= dA_cs_boundary
+
+            # - increment logical chunk index for every physical chunk
+            logical_chunk_idx += 1
+        else:
+            scale_mask = seq_idx_chunk_end == prev_seq_idx_chunk_end
+        prev_seq_idx_chunk_end = seq_idx_chunk_end
+
+        scale = tl.where(scale_mask, tl.exp(dA_cs), 0.0)
+        states = scale * states + new_states
+        tl.store(out_ptrs, states, mask=offs_m < dim)
+
+        states_ptrs += stride_states_chunk
+        dA_cs_ptr += stride_dA_cs_chunk
+        out_ptrs += stride_out_chunk
+
+
+def _state_passing_fwd(
+    states,
+    dA_cumsum,
+    seq_idx,
+    chunk_offsets,
+    initial_states=None,
+    out_dtype=None,
+):
+    nchunks, nheads, dim = states.shape
+    chunk_size = dA_cumsum.shape[-1]
+    assert dA_cumsum.shape == (nheads, nchunks, chunk_size)
+    seqlen = seq_idx.shape[-1]
+    out_dtype = states.dtype if out_dtype is None else out_dtype
+    out = torch.empty((nchunks, nheads, dim),
+                      device=states.device,
+                      dtype=out_dtype)
+
+    initial_states_strides = ((initial_states.stride(0),
+                               initial_states.stride(1),
+                               initial_states.stride(2))
+                              if initial_states is not None else (0, 0, 0))
+
+    grid = lambda META: (triton.cdiv(dim, META['BLOCK_SIZE']), nheads)
+    with torch.cuda.device(states.device.index):
+        _state_passing_fwd_kernel[grid](
+            states_ptr=states,
+            out_ptr=out,
+            dA_cs_ptr=dA_cumsum,
+            initstates_ptr=initial_states,
+            seq_idx_ptr=seq_idx,
+            chunk_offsets_ptr=chunk_offsets,
+            chunk_meta_num=len(chunk_offsets)
+            if chunk_offsets is not None else 0,
+            dim=dim,
+            nchunks=nchunks,
+            seqlen=seqlen if seq_idx is not None else 0,
+            chunk_size=chunk_size if seq_idx is not None else 0,
+            stride_states_chunk=states.stride(0),
+            stride_states_head=states.stride(1),
+            stride_states_dim=states.stride(2),
+            stride_out_chunk=out.stride(0),
+            stride_out_head=out.stride(1),
+            stride_out_dim=out.stride(2),
+            stride_dA_cs_head=dA_cumsum.stride(0),
+            stride_dA_cs_chunk=dA_cumsum.stride(1),
+            stride_dA_cs_csize=dA_cumsum.stride(2),
+            stride_initstates_batch=initial_states_strides[0],
+            stride_initstates_head=initial_states_strides[1],
+            stride_initstates_dim=initial_states_strides[2],
+            stride_seq_idx_seqlen=seq_idx.stride(0),
+            HAS_INITSTATES=initial_states is not None,
+        )
+    return out

vllm/model_executor/layers/mamba/short_conv.py

@@ -0,0 +1,253 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+from typing import TYPE_CHECKING, Optional
+
+if TYPE_CHECKING:
+    from vllm.attention.backends.abstract import AttentionBackend
+
+import torch
+
+from vllm.attention.backends.abstract import AttentionMetadata
+from vllm.config import CacheConfig, ModelConfig, get_current_vllm_config
+from vllm.distributed import get_tensor_model_parallel_world_size
+from vllm.forward_context import ForwardContext, get_forward_context
+from vllm.model_executor.custom_op import CustomOp
+from vllm.model_executor.layers.linear import (ColumnParallelLinear,
+                                               MergedColumnParallelLinear,
+                                               RowParallelLinear)
+from vllm.model_executor.layers.mamba.abstract import MambaBase
+from vllm.model_executor.layers.mamba.mamba_utils import (
+    MambaStateDtypeCalculator, MambaStateShapeCalculator)
+from vllm.model_executor.layers.mamba.ops.causal_conv1d import (
+    causal_conv1d_fn, causal_conv1d_update)
+from vllm.utils import direct_register_custom_op
+from vllm.v1.attention.backends.short_conv_attn import (
+    ShortConvAttentionMetadata)
+
+
+@CustomOp.register("short_conv")
+class ShortConv(MambaBase, CustomOp):
+
+    def __init__(self,
+                 config,
+                 dim: int,
+                 layer_idx: int,
+                 model_config: Optional[ModelConfig] = None,
+                 cache_config: Optional[CacheConfig] = None,
+                 prefix: str = ""):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.conv_dim = dim
+        self.L_cache = config.conv_L_cache
+        self.bias = config.conv_bias
+
+        self.conv = ColumnParallelLinear(
+            input_size=self.L_cache,
+            output_size=dim,
+            bias=self.bias,
+            prefix=f"{prefix}.conv1d",
+        )
+        # unsqueeze to fit conv1d weights shape into the linear weights shape.
+        # Can't do this in `weight_loader` since it already exists in
+        # `ColumnParallelLinear` and `set_weight_attrs`
+        # doesn't allow to override it
+        self.conv.weight.data = self.conv.weight.data.unsqueeze(1)
+
+        self.in_proj = MergedColumnParallelLinear(
+            input_size=dim,
+            output_sizes=[dim] * 3,
+            bias=self.bias,
+            prefix=f"{prefix}.in_proj",
+        )
+        self.out_proj = RowParallelLinear(
+            input_size=dim,
+            output_size=dim,
+            bias=self.bias,
+            prefix=f"{prefix}.out_proj",
+        )
+
+        compilation_config = get_current_vllm_config().compilation_config
+        if prefix in compilation_config.static_forward_context:
+            raise ValueError(f"Duplicate layer name: {prefix}")
+        compilation_config.static_forward_context[prefix] = self
+        self.kv_cache = (torch.tensor([]), )
+
+        self.model_config = model_config
+        self.cache_config = cache_config
+        self.prefix = prefix
+
+    def forward_native(
+        self,
+        hidden_states: torch.Tensor,
+        output: torch.Tensor,
+    ):
+        return
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        output: torch.Tensor,
+    ):
+        torch.ops.vllm.short_conv(
+            hidden_states,
+            output,
+            self.prefix,
+        )
+
+    def forward_cuda(
+        self,
+        hidden_states: torch.Tensor,
+        output: torch.Tensor,
+    ):
+        forward_context = get_forward_context()
+        # ShortConvAttentionMetadata contains metadata necessary for the
+        # short_conv triton kernels to operate in continuous batching and in
+        # chunked prefill modes; they are computed at top-level model forward
+        # since they stay the same and reused for all mamba layers in the same
+        # iteration.
+        attn_metadata: AttentionMetadata = forward_context.attn_metadata
+        if attn_metadata is not None:
+            assert isinstance(attn_metadata, dict)
+            attn_metadata = attn_metadata[self.prefix]
+            assert isinstance(attn_metadata, ShortConvAttentionMetadata)
+            self_kv_cache = self.kv_cache[forward_context.virtual_engine]
+            conv_state = self_kv_cache[0].transpose(-1, -2)
+            state_indices_tensor = attn_metadata.state_indices_tensor
+            has_initial_states_p = attn_metadata.has_initial_states
+
+        BCx, _ = self.in_proj(hidden_states)
+
+        B, C, x = BCx.chunk(3, dim=-1)
+
+        conv_weights = self.conv.weight.view(self.conv.weight.size(0),
+                                             self.conv.weight.size(2))
+
+        if attn_metadata is None:
+            # V1 profile run
+            Bx = (B * x).contiguous()
+            hidden_states = C * Bx
+            contextualized_states, _ = self.out_proj(hidden_states)
+            return contextualized_states
+
+        num_prefills = attn_metadata.num_prefills  # request count
+        num_decodes = attn_metadata.num_decode_tokens  # token count (=request)
+        num_prefill_tokens = attn_metadata.num_prefill_tokens  # token count
+        has_prefill = num_prefills > 0
+        has_decode = num_decodes > 0
+        num_actual_tokens = num_decodes + num_prefill_tokens
+
+        # NOTE: V1 puts decode before prefill
+        # Separate prefill and decode by splitting varlen input
+        # Split along token dimension
+        B_d, B_p = torch.split(
+            B[:num_actual_tokens],
+            [num_decodes, num_prefill_tokens],
+            dim=0,
+        )
+        C_d, C_p = torch.split(
+            C[:num_actual_tokens],
+            [num_decodes, num_prefill_tokens],
+            dim=0,
+        )
+        x_d, x_p = torch.split(
+            x[:num_actual_tokens],
+            [num_decodes, num_prefill_tokens],
+            dim=0,
+        )
+        # Split along batch dimension
+        state_indices_tensor_d, state_indices_tensor_p = torch.split(
+            state_indices_tensor,
+            [num_decodes, num_prefills],
+            dim=0,
+        )
+        query_start_loc_p = (
+            attn_metadata.query_start_loc[-num_prefills - 1:] -
+            num_decodes if has_prefill else None)
+
+        conv_output_list = []
+
+        if has_prefill:
+            Bx_p = (B_p * x_p).transpose(0, 1)
+            Bx = causal_conv1d_fn(Bx_p,
+                                  conv_weights,
+                                  self.conv.bias,
+                                  activation=None,
+                                  conv_states=conv_state,
+                                  has_initial_state=has_initial_states_p,
+                                  cache_indices=state_indices_tensor_p,
+                                  metadata=attn_metadata,
+                                  query_start_loc=query_start_loc_p).transpose(
+                                      0, 1)[:num_prefill_tokens]
+
+            y = C_p * Bx
+            conv_output_list.append(y)
+
+        if has_decode:
+            Bx_d = (B_d * x_d).contiguous()
+            Bx = causal_conv1d_update(
+                Bx_d,
+                conv_state,
+                conv_weights,
+                self.conv.bias,
+                activation=None,
+                conv_state_indices=state_indices_tensor_d)
+            y = C_d * Bx
+            conv_output_list.insert(0, y)
+
+        # Merge prefill and decode outputs before passing to gated MLP
+        hidden_states = torch.vstack(conv_output_list)
+
+        # Final linear projection
+        output[:num_actual_tokens], _ = self.out_proj(hidden_states)
+
+    def get_state_dtype(self) -> tuple[torch.dtype, ...]:
+        assert self.model_config is not None
+        assert self.cache_config is not None
+        return MambaStateDtypeCalculator.short_conv_state_dtype(
+            self.model_config.dtype,
+            self.cache_config.mamba_cache_dtype,
+        )
+
+    def get_state_shape(self) -> tuple[tuple[int, ...]]:
+        return MambaStateShapeCalculator.short_conv_state_shape(
+            tp_world_size=get_tensor_model_parallel_world_size(),
+            intermediate_size=self.conv_dim,
+            conv_kernel=self.L_cache,
+        )
+
+    @property
+    def mamba_type(self) -> str:
+        return "short_conv"
+
+    def get_attn_backend(self) -> type["AttentionBackend"]:
+        from vllm.v1.attention.backends.short_conv_attn import (
+            ShortConvAttentionBackend)
+        return ShortConvAttentionBackend
+
+
+def short_conv(
+    hidden_states: torch.Tensor,
+    output: torch.Tensor,
+    layer_name: str,
+) -> None:
+    forward_context: ForwardContext = get_forward_context()
+    self = forward_context.no_compile_layers[layer_name]
+    self.forward_cuda(hidden_states=hidden_states, output=output)
+
+
+def short_conv_fake(
+    hidden_states: torch.Tensor,
+    output: torch.Tensor,
+    layer_name: str,
+) -> None:
+    return
+
+
+direct_register_custom_op(
+    op_name="short_conv",
+    op_func=short_conv,
+    mutates_args=["output"],
+    fake_impl=short_conv_fake,
+)