mirror of
https://github.com/hpcaitech/ColossalAI.git
synced 2025-08-12 21:25:53 +00:00
9d02590c9a
* refactor: separate log_probs fn from Actor forward fn * refactor: separate generate fn from Actor class * feat: update unwrap_model and get_base_model * unwrap_model returns model not wrapped by Strategy * get_base_model returns HF model for Actor, Critic and RewardModel * feat: simplify Strategy.prepare * style: remove get_base_model method of Actor * perf: tokenize text in batches * refactor: move calc_action_log_probs to utils of model * test: update test with new forward fn * style: rename forward fn args * fix: do not unwrap model in save_model fn of naive strategy * test: add gemini test for train_prompts * fix: fix _set_default_generate_kwargs
37 lines
993 B
Python
37 lines
993 B
Python
from typing import Optional
|
|
|
|
import torch
|
|
import torch.nn as nn
|
|
|
|
from ..lora import LoRAModule
|
|
|
|
|
|
class Actor(LoRAModule):
|
|
"""
|
|
Actor model base class.
|
|
|
|
Args:
|
|
model (nn.Module): Actor Model.
|
|
lora_rank (int): LoRA rank.
|
|
lora_train_bias (str): LoRA bias training mode.
|
|
"""
|
|
|
|
def __init__(self, model: nn.Module, lora_rank: int = 0, lora_train_bias: str = 'none') -> None:
|
|
super().__init__(lora_rank=lora_rank, lora_train_bias=lora_train_bias)
|
|
self.model = model
|
|
self.convert_to_lora()
|
|
|
|
def forward(self,
|
|
input_ids: torch.LongTensor,
|
|
attention_mask: Optional[torch.Tensor] = None,
|
|
**model_kwargs, # HACK: `generate` method may pass more kwargs
|
|
) -> torch.Tensor:
|
|
"""Returns model output.
|
|
"""
|
|
output = self.model(
|
|
input_ids,
|
|
attention_mask=attention_mask,
|
|
**model_kwargs
|
|
)
|
|
return output
|