# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project from typing import Optional, Union import torch from torch import nn from vllm.attention.backends.abstract import AttentionMetadata 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 get_forward_context from vllm.model_executor.custom_op import CustomOp from vllm.model_executor.layers.linear import (ColumnParallelLinear, RowParallelLinear) from vllm.model_executor.layers.mamba.mamba2_metadata import Mamba2Metadata 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_state_update) from vllm.model_executor.layers.mamba.ops.ssd_combined import ( mamba_chunk_scan_combined) 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.models.mamba_cache import MambaCacheParams from vllm.model_executor.utils import set_weight_attrs # 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.tp_size > 1 or self.n_groups != 1: return self.forward_native(x, gate) from vllm import _custom_ops as ops # cast x and gate to float32 before silu out = torch.empty_like(x) y = x * nn.functional.silu(gate.to(torch.float32)) ops.rms_norm( out, y.to(x.dtype), self.weight.data, self.variance_epsilon, ) return out def extra_groups_for_head_shards(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 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(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, quant_config: Optional[QuantizationConfig] = None, ): 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_heads, " "then num_groups must equal 1.") assert ( self.tp_size == 1 or quant_config is None ), "Tensor parallel currently not supported for quantized models." self.ssm_state_size = ssm_state_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 self.n_groups = n_groups + extra_groups_for_head_shards( n_groups, self.tp_size) self.conv_dim = intermediate_size + 2 * self.n_groups * ssm_state_size self.conv1d = ColumnParallelLinear( input_size=conv_kernel_size, output_size=self.conv_dim, bias=use_conv_bias, quant_config=None, ) # 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 = ColumnParallelLinear( input_size=hidden_size, output_size=intermediate_size + self.conv_dim + self.num_heads, bias=use_bias, quant_config=quant_config, ) # - 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.n_groups * self.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_setings = (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 otther 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_setings, # for dt ], self.tp_size, tp_rank, ) }, ) # - 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, ) self.norm = Mixer2RMSNormGated(intermediate_size, n_groups, self.use_rms_norm, eps=rms_norm_eps) def forward_native( self, hidden_states: torch.Tensor, conv_state: torch.Tensor, ssm_state: torch.Tensor, ): pass def forward_cuda( self, hidden_states: torch.Tensor, mamba_cache_params: MambaCacheParams, mamba2_metadata: Mamba2Metadata, mup_vector: Optional[torch.Tensor] = None, ): # mamba2_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 = get_forward_context().attn_metadata 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 groups_time_state_size = self.n_groups * self.ssm_state_size # 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)) # Separate prefill and decode by splitting varlen input # Split along token dimension hidden_states_B_C_p, hidden_states_B_C_d = torch.split( hidden_states_B_C, [num_prefill_tokens, num_decodes], dim=0, ) dt_p, dt_d = torch.split( dt, [num_prefill_tokens, num_decodes], dim=0, ) # Split along batch dimension state_indices_tensor_p, state_indices_tensor_d = torch.split( mamba_cache_params.state_indices_tensor, [num_prefills, num_decodes], dim=0, ) query_start_loc_p = (attn_metadata.query_start_loc[:num_prefills + 1] if has_prefill else None) # - 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, groups_time_state_size // self.tp_size, groups_time_state_size // self.tp_size, ], dim=-1, ) ssd_output_list = [] # Process prefill requests if has_prefill: # 2. Convolution sequence transformation # - "cache_indices" updates the conv_state cache in positions # pointed to by "mamba_cache_params.state_indices_tensor" hidden_states_B_C_p = causal_conv1d_fn( hidden_states_B_C_p.transpose(0, 1), conv_weights, self.conv1d.bias, activation=self.activation, conv_states=mamba_cache_params.conv_state, has_initial_state=mamba2_metadata.has_initial_states, cache_indices=state_indices_tensor_p, query_start_loc=query_start_loc_p).transpose( 0, 1)[:num_prefill_tokens] # TODO: Why is this needed? hidden_states_B_C_p = hidden_states_B_C_p.contiguous() 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 (mamba2_metadata.has_initial_states is not None and mamba2_metadata.prep_initial_states): # making a copy of the states initial_states = torch.where( mamba2_metadata.has_initial_states[:, None, None, None], mamba_cache_params.ssm_state[state_indices_tensor_p], 0) scan_output, varlen_state = mamba_chunk_scan_combined( hidden_states_p.view(1, num_prefill_tokens, self.num_heads // self.tp_size, self.head_dim), dt_p.unsqueeze(0), self.A, B_p.view(1, num_prefill_tokens, self.n_groups // self.tp_size, -1), C_p.view(1, num_prefill_tokens, self.n_groups // self.tp_size, -1), chunk_size=mamba2_metadata.chunk_size, D=self.D, z=None, dt_bias=self.dt_bias, seq_idx=mamba2_metadata.seq_idx, chunk_indices=mamba2_metadata.chunk_indices, chunk_offsets=mamba2_metadata.chunk_offsets, cu_seqlens=attn_metadata.query_start_loc[:num_prefills + 1], initial_states=initial_states, return_varlen_states=True, return_final_states=False, dt_softplus=True, dt_limit=(0.0, float("inf")), ) # update ssm states # - varlen state is a (num_prefills, nheads, headdim, dstate) tensor mamba_cache_params.ssm_state[state_indices_tensor_p] = varlen_state # - reshape ssd_output_list.append(scan_output.view(num_prefill_tokens, -1)) # Process decode requests if has_decode: # 2. Convolution sequence transformation hidden_states_B_C_d = causal_conv1d_update( hidden_states_B_C_d, mamba_cache_params.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 hidden_states_d = selective_state_update( mamba_cache_params.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, ) ssd_output_list.append( hidden_states_d.view(-1, (self.num_heads // self.tp_size) * self.head_dim)) # Merge prefill and decode outputs before passing to gated MLP hidden_states = torch.vstack(ssd_output_list) # 4. gated MLP # GatedRMSNorm internally applying SiLU to the gate # SiLU is applied internally before normalization, unlike standard # norm usage hidden_states = self.norm(hidden_states, gate) # 5. Final linear projection out, _ = self.out_proj(hidden_states) return out