[v1] Support deepseek with eagle

Signed-off-by: Xin Yang <xyangx@amazon.com>

Author: xyang16

vllm/model_executor/models/deepseek_eagle.py

@@ -0,0 +1,221 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+from collections.abc import Iterable
+
+import torch
+import torch.nn as nn
+
+from vllm.compilation.decorators import support_torch_compile
+from vllm.config import VllmConfig
+from vllm.distributed.parallel_state import get_pp_group
+from vllm.model_executor.layers.fused_moe import FusedMoE
+from vllm.model_executor.layers.layernorm import RMSNorm
+from vllm.model_executor.layers.logits_processor import LogitsProcessor
+from vllm.model_executor.layers.vocab_parallel_embedding import (
+    VocabParallelEmbedding)
+from vllm.model_executor.model_loader.weight_utils import (
+    default_weight_loader, maybe_remap_kv_scale_name)
+from vllm.model_executor.models.deepseek_v2 import (
+    DeepseekV2DecoderLayer, DeepseekV3ForCausalLM,
+    get_spec_layer_idx_from_weight_name)
+
+from .utils import AutoWeightsLoader, maybe_prefix
+
+
+@support_torch_compile
+class DeepseekV2Model(nn.Module):
+
+    def __init__(
+        self,
+        *,
+        vllm_config: VllmConfig,
+        prefix: str = "",
+        start_layer_id: int = 0,
+    ) -> None:
+        super().__init__()
+        self.config = vllm_config. \
+            speculative_config.draft_model_config.hf_config
+        model_config = vllm_config.model_config
+        cache_config = vllm_config.cache_config
+        quant_config = vllm_config.quant_config
+        self.vocab_size = self.config.vocab_size
+
+        self.embed_tokens = VocabParallelEmbedding(
+            self.config.vocab_size,
+            self.config.hidden_size,
+            quant_config=quant_config,
+            prefix=maybe_prefix(prefix, "embed_tokens"),
+        )
+
+        self.layers = nn.ModuleList([
+            DeepseekV2DecoderLayer(
+                self.config,
+                prefix=maybe_prefix(prefix, f"layers.{i + start_layer_id}"),
+                model_config=model_config,
+                cache_config=cache_config,
+                quant_config=quant_config,
+            ) for i in range(self.config.num_hidden_layers)
+        ])
+
+        self.fc = nn.Linear(
+            self.config.model.hidden_size * 2,
+            self.config.model.hidden_size,
+            bias=False,
+        )
+
+        self.enorm = RMSNorm(self.config.hidden_size,
+                             eps=self.config.rms_norm_eps)
+        self.hnorm = RMSNorm(self.config.hidden_size,
+                             eps=self.config.rms_norm_eps)
+        self.norm = RMSNorm(self.config.hidden_size,
+                            eps=self.config.rms_norm_eps)
+
+    def forward(
+        self,
+        input_ids: torch.Tensor,
+        positions: torch.Tensor,
+        hidden_states: torch.Tensor,
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        input_embeds = self.embed_tokens(input_ids)
+
+        inputs = torch.cat(
+            [self.enorm(input_embeds),
+             self.hnorm(hidden_states)], dim=-1)
+        hidden_states = self.fc(inputs)
+
+        # masking inputs at position=0
+        hidden_states[positions == 0] = 0
+        residual = None
+        for layer in self.layers:
+            hidden_states, residual = layer(
+                positions,
+                hidden_states,
+                residual,
+            )
+        hidden_states, _ = self.norm(hidden_states, residual)
+        return hidden_states, hidden_states
+
+    def load_weights(self, weights: Iterable[tuple[str,
+                                                   torch.Tensor]]) -> set[str]:
+        stacked_params_mapping = [
+            # (param_name, shard_name, shard_id)
+            ("gate_up_proj", "gate_proj", 0),
+            ("gate_up_proj", "up_proj", 1),
+        ]
+
+        # Params for weights, fp8 weight scales, fp8 activation scales
+        # (param_name, weight_name, expert_id, shard_id)
+        expert_params_mapping = FusedMoE.make_expert_params_mapping(
+            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)
+
+        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
+
+            spec_layer = get_spec_layer_idx_from_weight_name(self.config, name)
+            if spec_layer is not None:
+                continue  # skip spec decode layers for main model
+
+            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
+                name = name.replace(weight_name, param_name)
+                # 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:
+                for mapping in expert_params_mapping:
+                    param_name, weight_name, expert_id, shard_id = mapping
+                    if weight_name not in name:
+                        continue
+                    name = name.replace(weight_name, param_name)
+
+                    param = params_dict[name]
+                    weight_loader = param.weight_loader
+                    weight_loader(
+                        param,
+                        loaded_weight,
+                        name,
+                        shard_id=shard_id,
+                        expert_id=expert_id,
+                    )
+                    break
+                else:
+                    # if PP disabled then draft will share embed with target
+                    if get_pp_group().world_size == 1 and \
+                            "embed_tokens." in name:
+                        continue
+
+                    # Skip loading extra bias for GPTQ models.
+                    if name.endswith(".bias") and name not in params_dict:
+                        continue
+
+                    # Remapping the name of FP8 kv-scale.
+                    name = maybe_remap_kv_scale_name(name, params_dict)
+                    if name is None:
+                        continue
+
+                    param = params_dict[name]
+                    weight_loader = getattr(param, "weight_loader",
+                                            default_weight_loader)
+                    weight_loader(param, loaded_weight)
+            loaded_params.add(name)
+        return loaded_params
+
+
+class EagleDeepseekForCausalLM(DeepseekV3ForCausalLM):
+
+    def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
+        nn.Module.__init__(self)
+        self.config = vllm_config. \
+            speculative_config.draft_model_config.hf_config
+        target_layer_num = vllm_config.model_config.get_num_layers(
+            vllm_config.parallel_config)
+        self.model = DeepseekV2Model(vllm_config=vllm_config,
+                                     prefix="model",
+                                     start_layer_id=target_layer_num)
+
+        logit_scale = getattr(self.config, "logit_scale", 1.0)
+        self.logits_processor = LogitsProcessor(self.config.vocab_size,
+                                                scale=logit_scale)
+
+    def forward(
+        self,
+        input_ids: torch.Tensor,
+        positions: torch.Tensor,
+        hidden_states: torch.Tensor,
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        return self.model(input_ids, positions, hidden_states)
+
+    def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]):
+        loader = AutoWeightsLoader(
+            self,
+            skip_prefixes=None,
+        )
+
+        model_weights = {}
+        for name, loaded_weight in weights:
+            if "lm_head" not in name:
+                name = "model." + name
+            model_weights[name] = loaded_weight
+        loader.load_weights(model_weights.items())

vllm/model_executor/models/registry.py

@@ -246,6 +246,7 @@
     "EagleLlamaForCausalLM": ("llama_eagle", "EagleLlamaForCausalLM"),
     "EagleMiniCPMForCausalLM": ("minicpm_eagle", "EagleMiniCPMForCausalLM"),
     "Eagle3LlamaForCausalLM": ("llama_eagle3", "Eagle3LlamaForCausalLM"),
+    "EagleDeepSeekMTPModel": ("deepseek_eagle", "EagleDeepseekForCausalLM"),
     "DeepSeekMTPModel": ("deepseek_mtp", "DeepSeekMTP"),
     "MedusaModel": ("medusa", "Medusa"),
     "MLPSpeculatorPreTrainedModel": ("mlp_speculator", "MLPSpeculator"),