Files
enginex-ascend-910-vllm/vllm_ascend/models/deepseek_v4_mtp.py
Sun Ruoxi 7f8a1b1f7a init v0.23.0
Signed-off-by: Sun Ruoxi <sunruoxi@4paradigm.com>
2026-08-27 15:11:51 +08:00

538 lines
22 KiB
Python

# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import typing
from collections.abc import Callable, Iterable
import torch
import torch.nn as nn
from transformers import PretrainedConfig
from vllm._aiter_ops import rocm_aiter_ops
from vllm.compilation.decorators import support_torch_compile
from vllm.config import VllmConfig
from vllm.distributed import get_tensor_model_parallel_rank, get_tensor_model_parallel_world_size
from vllm.model_executor.layers.fused_moe import FusedMoE
from vllm.model_executor.layers.layernorm import RMSNorm
from vllm.model_executor.layers.linear import ReplicatedLinear
from vllm.model_executor.layers.logits_processor import LogitsProcessor
from vllm.model_executor.layers.quantization import QuantizationConfig
from vllm.model_executor.layers.vocab_parallel_embedding import ParallelLMHead, VocabParallelEmbedding
from vllm.model_executor.model_loader.weight_utils import default_weight_loader, maybe_remap_kv_scale_name
from vllm.model_executor.models.interfaces import SupportsPP
from vllm.model_executor.models.utils import PPMissingLayer, maybe_prefix
from vllm.platforms import current_platform
from vllm.sequence import IntermediateTensors
from vllm_ascend.ascend_config import get_ascend_config
from vllm_ascend.utils import enable_dsa_cp, vllm_version_is
if not vllm_version_is("0.23.0"):
from vllm.model_executor.layers.fused_moe import fused_moe_make_expert_params_mapping
from .deepseek_v4 import (
DeepseekV2DecoderLayer,
DeepseekV2MixtureOfExperts,
DeepseekV4MoE,
get_spec_layer_idx_from_weight_name,
)
class SharedHead(nn.Module):
def __init__(
self,
config: PretrainedConfig,
prefix: str,
quant_config: QuantizationConfig | None = None,
) -> None:
super().__init__()
self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.head = ParallelLMHead(
config.vocab_size,
config.hidden_size,
quant_config=quant_config,
prefix=maybe_prefix(prefix, "head"),
)
def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
return self.norm(hidden_states)
class DeepSeekMultiTokenPredictorLayer(nn.Module):
def __init__(self, vllm_config: VllmConfig, prefix: str) -> None:
super().__init__()
config = vllm_config.speculative_config.draft_model_config.hf_config
self.config = config
quant_config = vllm_config.quant_config
self.e_proj = ReplicatedLinear(
config.hidden_size, config.hidden_size, bias=False, quant_config=quant_config, return_bias=False
)
self.h_proj = ReplicatedLinear(
config.hidden_size, config.hidden_size, bias=False, quant_config=quant_config, return_bias=False
)
self.enorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.hnorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.device = current_platform.device_type
self.is_v32 = hasattr(config, "index_topk")
if self.is_v32:
topk_tokens = config.index_topk
topk_indices_buffer = torch.empty(
vllm_config.scheduler_config.max_num_batched_tokens,
topk_tokens,
dtype=torch.int32,
device=self.device,
)
else:
topk_indices_buffer = None
self.shared_head = SharedHead(config=config, prefix=prefix, quant_config=quant_config)
self.mtp_block = DeepseekV2DecoderLayer(
vllm_config,
prefix,
config=self.config,
topk_indices_buffer=topk_indices_buffer,
is_draft_layer=True,
)
self.hc_eps = config.hc_eps
self.hc_mult = hc_mult = config.hc_mult
hc_dim = hc_mult * config.hidden_size
self.hc_head_fn = nn.Parameter(torch.empty(hc_mult, hc_dim, dtype=torch.float32))
self.hc_head_base = nn.Parameter(torch.empty(hc_mult, dtype=torch.float32))
self.hc_head_scale = nn.Parameter(torch.empty(1, dtype=torch.float32))
self.norm_eps = config.rms_norm_eps
def forward(
self,
input_ids: torch.Tensor,
positions: torch.Tensor,
previous_hidden_states: torch.Tensor,
inputs_embeds: torch.Tensor | None = None,
spec_step_index: int = 0,
) -> torch.Tensor:
assert inputs_embeds is not None
# masking inputs at position 0, as not needed by MTP
inputs_embeds = torch.where(positions.unsqueeze(-1) == 0, 0, inputs_embeds)
inputs_embeds = self.enorm(inputs_embeds)
previous_hidden_states = previous_hidden_states.view(-1, self.hc_mult, self.config.hidden_size)
previous_hidden_states = self.hnorm(previous_hidden_states)
hidden_states = self.e_proj(inputs_embeds).unsqueeze(-2) + self.h_proj(previous_hidden_states)
hidden_states, residual = self.mtp_block(positions=positions, hidden_states=hidden_states, residual=None)
# hidden_states = self.hc_head(hidden_states, self.hc_head_fn,
# self.hc_head_scale, self.hc_head_base)
return hidden_states
def hc_head(self, x: torch.Tensor, hc_fn: torch.Tensor, hc_scale: torch.Tensor, hc_base: torch.Tensor):
shape, dtype = x.size(), x.dtype
x = x.flatten(1).float()
rsqrt = torch.rsqrt(x.square().mean(-1, keepdim=True) + self.norm_eps)
mixes = torch.nn.functional.linear(x, hc_fn) * rsqrt
pre = torch.sigmoid(mixes * hc_scale + hc_base) + self.hc_eps
y = torch.sum(pre.unsqueeze(-1) * x.view(shape), dim=1)
return y.to(dtype)
class DeepSeekMultiTokenPredictor(nn.Module):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__()
config = vllm_config.model_config.hf_config
self.mtp_start_layer_idx = config.num_hidden_layers
self.num_mtp_layers = getattr(config, "num_nextn_predict_layers", 1)
# to map the exact layer index from weights
self.layers = torch.nn.ModuleDict(
{
str(idx): DeepSeekMultiTokenPredictorLayer(vllm_config, f"{prefix}.{idx}")
for idx in range(
0,
self.num_mtp_layers,
)
}
)
self.embed_tokens = VocabParallelEmbedding(
config.vocab_size,
config.hidden_size,
)
self.logits_processor = LogitsProcessor(config.vocab_size)
def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
return self.embed_tokens(input_ids)
def forward(
self,
input_ids: torch.Tensor,
positions: torch.Tensor,
previous_hidden_states: torch.Tensor,
inputs_embeds: torch.Tensor | None = None,
spec_step_idx: int = 0,
) -> torch.Tensor:
if inputs_embeds is None:
inputs_embeds = self.embed_tokens(input_ids)
current_step_idx = spec_step_idx % self.num_mtp_layers
return self.layers[str(current_step_idx)](
input_ids,
positions,
previous_hidden_states,
inputs_embeds,
current_step_idx,
)
def compute_logits(
self,
hidden_states: torch.Tensor,
spec_step_idx: int = 0,
) -> torch.Tensor:
current_step_idx = spec_step_idx % self.num_mtp_layers
mtp_layer = self.layers[str(current_step_idx)]
hidden_states = hidden_states.view(-1, mtp_layer.hc_mult, mtp_layer.config.hidden_size)
hidden_states = mtp_layer.hc_head(
hidden_states, mtp_layer.hc_head_fn, mtp_layer.hc_head_scale, mtp_layer.hc_head_base
)
logits = self.logits_processor(mtp_layer.shared_head.head, mtp_layer.shared_head(hidden_states))
return logits
@support_torch_compile
class DeepSeekV4MTP(nn.Module, SupportsPP, DeepseekV2MixtureOfExperts):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__()
self.config = vllm_config.model_config.hf_config
self.quant_config = vllm_config.quant_config
self.model = DeepSeekMultiTokenPredictor(vllm_config=vllm_config, prefix=maybe_prefix(prefix, "mtp"))
# Set MoE hyperparameters
self.set_moe_parameters()
def set_moe_parameters(self):
self.expert_weights = []
self.num_expert_groups = getattr(self.config, "n_group", 1)
self.moe_layers = []
self.moe_mlp_layers = []
example_moe = None
for layer in self.model.layers.values():
if isinstance(layer, PPMissingLayer):
continue
assert isinstance(layer, DeepSeekMultiTokenPredictorLayer)
layer = layer.mtp_block
assert isinstance(layer, DeepseekV2DecoderLayer)
if isinstance(layer.mlp, DeepseekV4MoE):
# Pick last one layer since the first ones may be dense layers.
example_moe = layer.mlp
self.moe_mlp_layers.append(layer.mlp)
self.moe_layers.append(layer.mlp.experts)
self.extract_moe_parameters(example_moe)
def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
return self.model.embed_input_ids(input_ids)
def forward(
self,
input_ids: torch.Tensor,
positions: torch.Tensor,
hidden_states: torch.Tensor,
intermediate_tensors: IntermediateTensors | None = None,
inputs_embeds: torch.Tensor | None = None,
spec_step_idx: int = 0,
) -> torch.Tensor:
hidden_states = self.model(input_ids, positions, hidden_states, inputs_embeds, spec_step_idx)
return hidden_states
def compute_logits(
self,
hidden_states: torch.Tensor,
spec_step_idx: int = 0,
) -> torch.Tensor | None:
return self.model.compute_logits(hidden_states, spec_step_idx)
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
rocm_aiter_moe_shared_expert_enabled = rocm_aiter_ops.is_fusion_moe_shared_experts_enabled()
rocm_aiter_moe_shared_expert_enabled = getattr(get_ascend_config(), "mix_placement", False)
stacked_params_mapping = [
("gate_up_proj", "gate_proj", 0),
("gate_up_proj", "up_proj", 1),
]
if vllm_version_is("0.23.0"):
expert_params_mapping = FusedMoE.make_expert_params_mapping(
model=self.model,
ckpt_gate_proj_name="gate_proj",
ckpt_down_proj_name="down_proj",
ckpt_up_proj_name="up_proj",
num_experts=self.config.n_routed_experts
+ (self.config.n_shared_experts if rocm_aiter_moe_shared_expert_enabled else 0),
num_redundant_experts=self.num_redundant_experts,
)
else:
expert_params_mapping = fused_moe_make_expert_params_mapping(
model=self.model,
ckpt_gate_proj_name="gate_proj",
ckpt_down_proj_name="down_proj",
ckpt_up_proj_name="up_proj",
num_experts=self.config.n_routed_experts
+ (self.config.n_shared_experts if rocm_aiter_moe_shared_expert_enabled else 0),
num_redundant_experts=self.num_redundant_experts,
)
tp_rank = get_tensor_model_parallel_rank()
tp_size = get_tensor_model_parallel_world_size()
# Attention heads per rank
heads_per_rank = self.config.num_attention_heads // tp_size
head_start = tp_rank * heads_per_rank
params_dict = dict(self.named_parameters())
loaded_params: set[str] = set()
for name, loaded_weight in weights:
if "rotary_emb.inv_freq" in name:
continue
if self.quant_config is not None and self.quant_config.get_name() == "fp8":
if name == "embed.weight":
name = "mtp.0.emb.tok_emb.weight"
if name == "head.weight":
name = "mtp.0.head.weight"
spec_layer = get_spec_layer_idx_from_weight_name(self.config, name)
if spec_layer is None:
continue
assert "mtp.0." in name
if ".emb.tok_emb." in name:
name = name.replace("mtp.0.", "model.")
elif self.no_mtp_block_in_name(name):
name = name.replace("mtp.0.", "model.layers.0.")
else:
name = name.replace("mtp.0.", "model.layers.0.mtp_block.")
if ".w1." in name:
name = name.replace(".w1.", ".gate_proj.")
if ".w2." in name:
name = name.replace(".w2.", ".down_proj.")
if ".w3." in name:
name = name.replace(".w3.", ".up_proj.")
if name.endswith(".scale"):
name = name.replace(".scale", ".weight_scale")
if ".head." in name:
name = name.replace(".head.", ".shared_head.head.")
if ".norm." in name:
name = name.replace(".norm.", ".shared_head.norm.")
if ".emb.tok_emb." in name:
name = name.replace(".emb.tok_emb.", ".embed_tokens.")
if "attn" in name and "self_attn" not in name:
name = name.replace(".attn.", ".self_attn.")
if ".ffn." in name:
name = name.replace(".ffn.", ".mlp.")
if ".ffn_norm." in name:
name = name.replace(".ffn_norm.", ".post_attention_layernorm.")
if ".attn_norm." in name:
name = name.replace(".attn_norm.", ".input_layernorm.")
if ".gate.bias" in name:
name = name.replace(".gate.bias", ".gate.e_score_correction_bias")
if "sink" in name:
param = params_dict[name]
if enable_dsa_cp():
param.data.copy_(loaded_weight)
else:
# Handle attention sinks (distributed across ranks)
narrow_weight = loaded_weight.narrow(0, head_start, heads_per_rank)
param.data.copy_(narrow_weight)
loaded_params.add(name)
continue
is_fusion_moe_shared_experts_layer = rocm_aiter_moe_shared_expert_enabled and ("mlp.shared_experts" in name)
for param_name, weight_name, shard_id in stacked_params_mapping:
# Skip non-stacked layers and experts (experts handled below).
if weight_name not in name:
continue
# We have mlp.experts[0].gate_proj in the checkpoint.
# Since we handle the experts below in expert_params_mapping,
# we need to skip here BEFORE we update the name, otherwise
# name will be updated to mlp.experts[0].gate_up_proj, which
# will then be updated below in expert_params_mapping
# for mlp.experts[0].gate_gate_up_proj, which breaks load.
if ("mlp.experts." in name) and name not in params_dict:
continue
if is_fusion_moe_shared_experts_layer:
continue
name_mapped = name.replace(weight_name, param_name)
# QKV fusion is optional, fall back to normal
# weight loading if it's not enabled
if (param_name == "fused_qkv_a_proj") and name_mapped not in params_dict:
continue
else:
name = name_mapped
# Skip loading extra bias for GPTQ models.
if name.endswith(".bias") and name not in params_dict:
continue
param = params_dict[name]
weight_loader = param.weight_loader
weight_loader(param, loaded_weight, shard_id)
break
else:
# Special handling: when AITER fusion_shared_experts is enabled,
# checkpoints may provide a single widened shared_experts tensor
# without explicit expert indices
# (e.g. ...mlp.shared_experts.gate_proj.weight).
# For models with multiple shared experts, split that tensor
# evenly into per-shared-expert slices and load them into
# appended expert slots mlp.experts.{n_routed_experts + j}.*
# accordingly.
num_chunks = 1
if is_fusion_moe_shared_experts_layer:
num_chunks = getattr(self.config, "n_shared_experts", 1) or 1
# Determine split axis based on op type
# gate/up: ColumnParallel → split along dim 0
# down: RowParallel → split along dim 1
split_dim = 1 if "down_proj.weight" in name else 0
total = loaded_weight.shape[split_dim]
assert total % num_chunks == 0, (
f"Shared expert weight dim {total} not divisible by num_chunks {num_chunks}"
)
chunk_size = total // num_chunks
for j in range(num_chunks):
chunk_name = name
weight_to_load = loaded_weight
if is_fusion_moe_shared_experts_layer:
if split_dim == 0:
weight_to_load = loaded_weight[j * chunk_size : (j + 1) * chunk_size, :]
else:
weight_to_load = loaded_weight[:, j * chunk_size : (j + 1) * chunk_size]
# Synthesize an expert-style name so expert mapping
# can route it
chunk_name = name.replace(
"mlp.shared_experts",
f"mlp.experts.{self.config.n_routed_experts + j}",
)
# Use expert_params_mapping to locate the destination
# param and delegate to its expert-aware weight_loader
# with expert_id.
is_expert_weight = False
for mapping in expert_params_mapping:
param_name, weight_name, expert_id, shard_id = mapping
if weight_name not in chunk_name:
continue
# Anyway, this is an expert weight and should not be
# attempted to load as other weights later
is_expert_weight = True
# Do not modify `name` since the loop may continue here
# Instead, create a new variable
name_mapped = chunk_name.replace(weight_name, param_name)
param = params_dict[name_mapped]
# We should ask the weight loader to return success or
# not here since otherwise we may skip experts with
# other available replicas.
weight_loader = typing.cast(Callable[..., bool], param.weight_loader)
success = weight_loader(
param,
weight_to_load,
name_mapped,
shard_id=shard_id,
expert_id=expert_id,
return_success=True,
)
if success:
if not is_fusion_moe_shared_experts_layer:
name = name_mapped
else:
loaded_params.add(name_mapped)
break
else:
if is_expert_weight:
# We've checked that this is an expert weight
# However it's not mapped locally to this rank
# So we simply skip it
continue
# Skip loading extra bias for GPTQ models.
if name.endswith(".bias") and name not in params_dict:
continue
name = maybe_remap_kv_scale_name(name, params_dict)
if name is None:
continue
# # According to DeepSeek-V3 Technical Report, MTP modules
# # shares embedding layer. We only load the first weights.
# if (
# spec_layer != self.model.mtp_start_layer_idx
# and ".layers" not in name
# ):
# continue
param = params_dict[name]
weight_loader = getattr(param, "weight_loader", default_weight_loader)
weight_loader(param, loaded_weight)
if not is_fusion_moe_shared_experts_layer:
loaded_params.add(name)
return loaded_params
def _rewrite_spec_layer_name(self, spec_layer: int, name: str) -> str:
"""
Rewrite the weight name to match the format of the original model.
Add .mtp_block for modules in transformer layer block for spec layer
and rename shared layer weights to be top level.
"""
spec_layer_weight_names = [
"embed_tokens",
"enorm",
"hnorm",
"eh_proj",
"shared_head",
]
shared_weight_names = ["embed_tokens"]
spec_layer_weight = False
shared_weight = False
for weight_name in spec_layer_weight_names:
if weight_name in name:
spec_layer_weight = True
if weight_name in shared_weight_names:
shared_weight = True
break
if not spec_layer_weight:
# treat rest weights as weights for transformer layer block
name = name.replace(f"model.layers.{spec_layer}.", f"model.layers.{spec_layer}.mtp_block.")
elif shared_weight:
# treat shared weights as top level weights
name = name.replace(f"model.layers.{spec_layer}.", "model.")
return name
def no_mtp_block_in_name(self, layer_name: str) -> bool:
names = [
".hc_head_fn",
".hc_head_base",
".hc_head_scale",
".e_proj.",
".h_proj.",
".enorm.",
".hnorm.",
".norm.",
".head.",
".emb.tok_emb.",
]
return any(name in layer_name for name in names)