[ColossalChat] Update RLHF V2 (#5286)

* Add dpo. Fix sft, ppo, lora. Refactor all * fix and tested ppo * 2 nd round refactor * add ci tests * fix ci * fix ci * fix readme, style * fix readme style * fix style, fix benchmark * reproduce benchmark result, remove useless files * rename to ColossalChat * use new image * fix ci workflow * fix ci * use local model/tokenizer for ci tests * fix ci * fix ci * fix ci * fix ci timeout * fix rm progress bar. fix ci timeout * fix ci * fix ci typo * remove 3d plugin from ci temporary * test environment * cannot save optimizer * support chat template * fix readme * fix path * test ci locally * restore build_or_pr * fix ci data path * fix benchmark * fix ci, move ci tests to 3080, disable fast tokenizer * move ci to 85 * support flash attention 2 * add all-in-one data preparation script. Fix colossal-llama2-chat chat template * add hardware requirements * move ci test data * fix save_model, add unwrap * fix missing bos * fix missing bos; support grad accumulation with gemini * fix ci * fix ci * fix ci * fix llama2 chat template config * debug sft * debug sft * fix colossalai version requirement * fix ci * add sanity check to prevent NaN loss * fix requirements * add dummy data generation script * add dummy data generation script * add dummy data generation script * add dummy data generation script * update readme * update readme * update readme and ignore * fix logger bug * support parallel_output * modify data preparation logic * fix tokenization * update lr * fix inference * run pre-commit --------- Co-authored-by: Tong Li <tong.li352711588@gmail.com>
2025-09-05 11:02:05 +00:00 · 2024-03-29 14:12:29 +08:00
parent 36c4bb2893
commit df5e9c53cf
200 changed files with 8848 additions and 8049 deletions
--- a/applications/ColossalChat/coati/models/base.py
+++ b/applications/ColossalChat/coati/models/base.py
@@ -0,0 +1,58 @@
+"""
+Base class for critic and reward model
+"""
+
+from typing import Optional
+
+import torch
+import torch.nn as nn
+from transformers import AutoModel, PretrainedConfig
+
+
+class BaseModel(nn.Module):
+    """
+    Actor model base class.
+
+    Args:
+        pretrained (str): path to pretrained model.
+        config (PretrainedConfig): PretrainedConfig used to initiate the base model.
+        **kwargs: all other kwargs as in AutoModel.from_pretrained
+    """
+
+    def __init__(self, pretrained: str = None, config: Optional[PretrainedConfig] = None, **kwargs) -> None:
+        super().__init__()
+        if pretrained is not None:
+            if config is not None:
+                # initialize with config and load weights from pretrained
+                self.model = AutoModel.from_pretrained(pretrained, config=config, **kwargs)
+            else:
+                # initialize with pretrained
+                self.model = AutoModel.from_pretrained(pretrained, **kwargs)
+        elif config is not None:
+            # initialize with config
+            self.model = AutoModel.from_config(config, **kwargs)
+        else:
+            raise ValueError("Either pretrained or config must be provided.")
+
+        self.config = self.model.config
+        # create dummy input to get the size of the last hidden state
+        if "use_flash_attention_2" in kwargs:
+            self.model = self.model.cuda()
+        dummy_input = torch.zeros((1, 1), dtype=torch.long).to(self.model.device)
+        out = self.model(dummy_input)
+        self.last_hidden_state_size = out.last_hidden_state.shape[-1]
+        self.model = self.model.cpu()
+        # print("self.last_hidden_state_size: ",self.last_hidden_state_size)
+
+    def resize_token_embeddings(self, *args, **kwargs):
+        """
+        Resize the token embeddings of the model.
+
+        Args:
+            *args: Variable length argument list.
+            **kwargs: Arbitrary keyword arguments.
+
+        Returns:
+            The resized token embeddings.
+        """
+        return self.model.resize_token_embeddings(*args, **kwargs)