github
/
ColossalAI
mirror of https://github.com/hpcaitech/ColossalAI
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

[chatgpt] fix lora support for gpt (#3113) 

* fix gpt-actor

* fix gpt-critic

* fix opt-critic
pull/3116/head
BlueRum 2023-03-13 10:37:41 +08:00 committed by GitHub
parent 0aa92c0409
commit 0672b5afac
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 12 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
applications/ChatGPT/chatgpt/models/gpt/gpt_actor.py
View File

@ -14,12 +14,16 @@ class GPTActor(Actor):
pretrained (str): Pretrained model name or path.
config (GPT2Config): Model config.
checkpoint (bool): Enable gradient checkpointing.
lora_rank (int): Rank of the LoRa layer.
lora_train_bias (str): Bias training strategy for the LoRa layer.
"""
def __init__(self,
pretrained: Optional[str] = None,
config: Optional[GPT2Config] = None,
checkpoint: bool = False) -> None:
checkpoint: bool = False,
lora_rank: int = 0,
lora_train_bias: str = 'none') -> None:
if pretrained is not None:
model = GPT2LMHeadModel.from_pretrained(pretrained)
elif config is not None:
@ -28,4 +32,4 @@ class GPTActor(Actor):
model = GPT2LMHeadModel(GPT2Config())
if checkpoint:
model.gradient_checkpointing_enable()
super().__init__(model)
super().__init__(model, lora_rank, lora_train_bias)

7
applications/ChatGPT/chatgpt/models/gpt/gpt_critic.py
View File

@ -15,13 +15,16 @@ class GPTCritic(Critic):
pretrained (str): Pretrained model name or path.
config (GPT2Config): Model config.
checkpoint (bool): Enable gradient checkpointing.
lora_rank (int): Rank of the LO-RA decomposition.
lora_train_bias (str): LoRA bias training mode.
"""
def __init__(self,
pretrained: Optional[str] = None,
config: Optional[GPT2Config] = None,
checkpoint: bool = False,
**kwargs) -> None:
lora_rank: int = 0,
lora_train_bias: str = 'none') -> None:
if pretrained is not None:
model = GPT2Model.from_pretrained(pretrained)
elif config is not None:
@ -31,4 +34,4 @@ class GPTCritic(Critic):
if checkpoint:
model.gradient_checkpointing_enable()
value_head = nn.Linear(model.config.n_embd, 1)
super().__init__(model, value_head, **kwargs)
super().__init__(model, value_head, lora_rank, lora_train_bias)

2
applications/ChatGPT/chatgpt/models/opt/opt_critic.py
View File

@ -34,5 +34,5 @@ class OPTCritic(Critic):
model = OPTModel(OPTConfig())
if checkpoint:
model.gradient_checkpointing_enable()
value_head = nn.Linear(model.config.hidden_size, 1)
value_head = nn.Linear(model.config.word_embed_proj_dim, 1)
super().__init__(model, value_head, lora_rank, lora_train_bias, **kwargs)