From 8586369f617a964235d0d9d32d6ebb1076a4581d Mon Sep 17 00:00:00 2001 From: Matthew Santiago Date: Sat, 2 May 2026 01:22:14 -0500 Subject: [PATCH] Refactor Step3Text loading to use AutoWeightsLoader (#41492) Signed-off-by: Matthew Santiago --- vllm/model_executor/models/step3_text.py | 103 ++++++++++++----------- 1 file changed, 54 insertions(+), 49 deletions(-) diff --git a/vllm/model_executor/models/step3_text.py b/vllm/model_executor/models/step3_text.py index 8f08f6c607..a0e7e16a9b 100644 --- a/vllm/model_executor/models/step3_text.py +++ b/vllm/model_executor/models/step3_text.py @@ -41,6 +41,7 @@ from vllm.transformers_utils.configs.step3_vl import Step3TextConfig from .interfaces import SupportsPP from .utils import ( + AutoWeightsLoader, PPMissingLayer, is_pp_missing_parameter, make_empty_intermediate_tensors_factory, @@ -382,55 +383,6 @@ class Step3TextModel(nn.Module): hidden_states, _ = self.norm(hidden_states, residual) return hidden_states - -class Step3TextForCausalLM(nn.Module, SupportsPP): - def __init__( - self, - *, - vllm_config: VllmConfig, - prefix: str = "", - ): - super().__init__() - config = vllm_config.model_config.hf_config - - self.config = config - self.vllm_config = vllm_config - - self.model = Step3TextModel(vllm_config=vllm_config, prefix=prefix) - - if get_pp_group().is_last_rank: - self.lm_head = ParallelLMHead( - config.vocab_size, - config.hidden_size, - prefix=maybe_prefix(prefix, "lm_head"), - ) - self.logits_processor = LogitsProcessor(config.vocab_size) - else: - self.lm_head = PPMissingLayer() - - self.make_empty_intermediate_tensors = ( - self.model.make_empty_intermediate_tensors - ) - - 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 | None, - positions: torch.Tensor, - intermediate_tensors: IntermediateTensors | None = None, - inputs_embeds: torch.Tensor | None = None, - ): - hidden_states = self.model( - input_ids, positions, intermediate_tensors, inputs_embeds - ) - return hidden_states - - def compute_logits(self, hidden_states: torch.Tensor) -> torch.Tensor: - logits = self.logits_processor(self.lm_head, hidden_states) - return logits - def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]: qkv_params_mapping = [ # (param_name, shard_name, relative_start_idx, relative_end_idx) @@ -553,3 +505,56 @@ class Step3TextForCausalLM(nn.Module, SupportsPP): weight_loader(param, loaded_weight) loaded_params.add(name) return loaded_params + + +class Step3TextForCausalLM(nn.Module, SupportsPP): + def __init__( + self, + *, + vllm_config: VllmConfig, + prefix: str = "", + ): + super().__init__() + config = vllm_config.model_config.hf_config + + self.config = config + self.vllm_config = vllm_config + + self.model = Step3TextModel(vllm_config=vllm_config, prefix=prefix) + + if get_pp_group().is_last_rank: + self.lm_head = ParallelLMHead( + config.vocab_size, + config.hidden_size, + prefix=maybe_prefix(prefix, "lm_head"), + ) + self.logits_processor = LogitsProcessor(config.vocab_size) + else: + self.lm_head = PPMissingLayer() + + self.make_empty_intermediate_tensors = ( + self.model.make_empty_intermediate_tensors + ) + + 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 | None, + positions: torch.Tensor, + intermediate_tensors: IntermediateTensors | None = None, + inputs_embeds: torch.Tensor | None = None, + ): + hidden_states = self.model( + input_ids, positions, intermediate_tensors, inputs_embeds + ) + return hidden_states + + def compute_logits(self, hidden_states: torch.Tensor) -> torch.Tensor: + logits = self.logits_processor(self.lm_head, hidden_states) + return logits + + def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]: + loader = AutoWeightsLoader(self) + return loader.load_weights(weights)