Share target model embed and head weights for nextn (#4033)

python/sglang/srt/model_executor/forward_batch_info.py

@@ -280,7 +280,8 @@ class ForwardBatch:
             ).to(device, non_blocking=True)
             if (
                 model_runner.server_args.attention_backend != "torch_native"
-                and model_runner.server_args.speculative_algorithm != "NEXTN"
+                # TODO: Fix triton kernel illegal memory access for EAGLE
+                and model_runner.server_args.speculative_algorithm != "EAGLE"
             ):
                 ret.extend_num_tokens = batch.extend_num_tokens
                 positions, ret.extend_start_loc = compute_position_triton(

python/sglang/srt/models/deepseek_nextn.py

@@ -116,14 +116,14 @@ class DeepseekV3ForCausalLMNextN(DeepseekV3ForCausalLM):
         self.model = DeepseekModelNextN(config, quant_config)
 
         if global_server_args_dict["enable_dp_attention"]:
-            self.model.shared_head.head = ReplicatedLinear(
+            self.lm_head = ReplicatedLinear(
                 config.hidden_size,
                 config.vocab_size,
                 bias=False,
             )
             self.logits_processor = LogitsProcessor(config, skip_all_gather=True)
         else:
-            self.model.shared_head.head = ParallelLMHead(
+            self.lm_head = ParallelLMHead(
                 config.vocab_size,
                 config.hidden_size,
                 quant_config=quant_config,
@@ -139,7 +139,7 @@ class DeepseekV3ForCausalLMNextN(DeepseekV3ForCausalLM):
     ) -> torch.Tensor:
         hidden_states = self.model(input_ids, positions, forward_batch)
         return self.logits_processor(
-            input_ids, hidden_states, self.model.shared_head.head, forward_batch
+            input_ids, hidden_states, self.lm_head, forward_batch
         )
 
     def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
@@ -168,10 +168,8 @@ class DeepseekV3ForCausalLMNextN(DeepseekV3ForCausalLM):
 
         nextn_layer_prefix = "model.layers.0"
         nextn_spec_weight_names = [
-            "shared_head.head",
             "shared_head.norm",
             "eh_proj",
-            "embed_tokens",
             "enorm",
             "hnorm",
         ]
@@ -180,7 +178,11 @@ class DeepseekV3ForCausalLMNextN(DeepseekV3ForCausalLM):
         for name, loaded_weight in weights:
             if not name.startswith(nextn_layer_prefix):
                 continue
-            else:
+
+            # Use shared head and embed weights from target model
+            if "shared_head.head" in name or "embed_tokens" in name:
+                continue
+
             is_decoder = True
             # For nextn specific weights
             for weight_name in nextn_spec_weight_names:

python/sglang/srt/models/deepseek_v2.py

@@ -1179,6 +1179,17 @@ class DeepseekV2ForCausalLM(nn.Module):
                     if is_hip_:
                         self_attn.w_scale *= 2.0
 
+    def get_embed_and_head(self):
+        return self.model.embed_tokens.weight, self.lm_head.weight
+
+    def set_embed_and_head(self, embed, head):
+        del self.model.embed_tokens.weight
+        del self.lm_head.weight
+        self.model.embed_tokens.weight = embed
+        self.lm_head.weight = head
+        torch.cuda.empty_cache()
+        torch.cuda.synchronize()
+
 
 class DeepseekV3ForCausalLM(DeepseekV2ForCausalLM):
     pass

python/sglang/srt/server_args.py

@@ -270,10 +270,11 @@ class ServerArgs:
             )
 
         # Speculative Decoding
-        if (
-            self.speculative_algorithm == "EAGLE"
-            or self.speculative_algorithm == "NEXTN"
-        ):
+        if self.speculative_algorithm == "NEXTN":
+            # NEXTN shares the same implementation of EAGLE
+            self.speculative_algorithm = "EAGLE"
+
+        if self.speculative_algorithm == "EAGLE":
             self.disable_overlap_schedule = True
             self.prefill_only_one_req = True
             self.disable_cuda_graph_padding = True

python/sglang/srt/speculative/eagle_worker.py

@@ -83,7 +83,6 @@ class EAGLEWorker(TpModelWorker):
         self.server_args = server_args
 
         # Share the embedding and lm_head
-        if not self.speculative_algorithm.is_nextn():
         embed, head = self.target_worker.model_runner.model.get_embed_and_head()
         if server_args.speculative_token_map is not None:
             head = head.clone()
@@ -94,12 +93,6 @@ class EAGLEWorker(TpModelWorker):
         else:
             self.hot_token_id = None
         self.model_runner.model.set_embed_and_head(embed, head)
-        else:
-            if server_args.speculative_token_map is not None:
-                raise NotImplementedError(
-                    "NEXTN does not support speculative-token-map now"
-                )
-            self.hot_token_id = None
         self.model_runner.server_args.disable_cuda_graph = backup_disable_cuda_graph
 
         # Create multi-step attn backends and cuda graph runners

python/sglang/srt/speculative/spec_info.py

@@ -5,24 +5,16 @@ class SpeculativeAlgorithm(IntEnum):
     NONE = auto()
     EAGLE = auto()
 
-    # NEXTN spec decoding is for DeepSeek V3/R1
-    # currently it's implemented based on EAGLE
-    NEXTN = auto()
-
     def is_none(self):
         return self == SpeculativeAlgorithm.NONE
 
     def is_eagle(self):
-        return self == SpeculativeAlgorithm.EAGLE or self == SpeculativeAlgorithm.NEXTN
-
-    def is_nextn(self):
-        return self == SpeculativeAlgorithm.NEXTN
+        return self == SpeculativeAlgorithm.EAGLE
 
     @staticmethod
     def from_string(name: str):
         name_map = {
             "EAGLE": SpeculativeAlgorithm.EAGLE,
-            "NEXTN": SpeculativeAlgorithm.NEXTN,
             None: SpeculativeAlgorithm.NONE,
         }
         if name is not None:

scripts/export_deepseek_nextn.py

@@ -62,6 +62,8 @@ def export_nextn_layer_parameters(input_dir, output_dir, nextn_layer_id):
                     continue
 
                 for key in matching_keys:
+                    if "embed_tokens" in key or "shared_head.head" in key:
+                        continue
                     new_key = key.replace(prefix, "model.layers.0")
                     params[new_key] = f.get_tensor(key)