← Về thư mục
verl/workers/fsdp_workers.py
- ActorRolloutRefWorker · class · L47-L535 — class ActorRolloutRefWorker(Worker)
- init · method · L53-L109 — def init(self, config: DictConfig, role: str)
- _build_model_optimizer · method · L111-L248 — def _build_model_optimizer(self, model_path, fsdp_config, optim_config, override_model_config, use_remove_padding=False, enable_gradient_checkpointing=False, trust_remote_code=False)
- _build_rollout · method · L250-L282 — def _build_rollout(self)
- init_model · method · L285-L353 — def init_model(self)
- update_actor · method · L356-L398 — def update_actor(self, data: DataProto)
- compute_log_prob · method · L401-L429 — def compute_log_prob(self, data: DataProto) -> DataProto
- generate_sequences · method · L432-L477 — def generate_sequences(self, prompts: DataProto)
- compute_ref_log_prob · method · L480-L506 — def compute_ref_log_prob(self, data: DataProto)
- save_checkpoint · method · L509-L535 — def save_checkpoint(self, local_path, hdfs_path=None)
- CriticWorker · class · L538-L792 — class CriticWorker(Worker)
- init · method · L540-L569 — def init(self, config)
- _build_critic_model_optimizer · method · L571-L680 — def _build_critic_model_optimizer(self, config): # the following line is necessary
- init_model · method · L683-L702 — def init_model(self): # This is used to import external_lib into the huggingface systems
- compute_values · method · L705-L727 — def compute_values(self, data: DataProto)
- update_critic · method · L730-L764 — def update_critic(self, data: DataProto)
- save_checkpoint · method · L767-L792 — def save_checkpoint(self, local_path, hdfs_path=None)
- RewardModelWorker · class · L796-L1054 — class RewardModelWorker(Worker)
- init · method · L801-L822 — def init(self, config)
- _build_model · method · L824-L879 — def _build_model(self, config): # the following line is necessary
- init_model · method · L882-L886 — def init_model(self): # This is used to import external_lib into the huggingface systems
- _forward_micro_batch · method · L888-L940 — def _forward_micro_batch(self, micro_batch)
- _expand_to_token_level · method · L942-L955 — def _expand_to_token_level(self, data: DataProto, scores: torch.Tensor)
- _switch_chat_template · method · L957-L1012 — def _switch_chat_template(self, data: DataProto)
- compute_rm_score · method · L1015-L1054 — def compute_rm_score(self, data: DataProto)