423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-08-04 20:52:59 +08:00
Code Issues Packages Projects Releases Wiki Activity

tiny fix

Browse Source 
Former-commit-id: 415bca900e5cc3afaddd5b06d35f472d9ead3263
This commit is contained in:
hiyouga 2023-11-09 17:20:49 +08:00
parent 9b98790fb3
commit 6ee32cf71c
2 changed files with 4 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/llmtuner/hparams/finetuning_args.py
View File

@ -45,7 +45,7 @@ class FinetuningArguments:
default=None, default=None,
metadata={"help": "Name(s) of target modules to apply LoRA. Use commas to separate multiple modules. \ metadata={"help": "Name(s) of target modules to apply LoRA. Use commas to separate multiple modules. \
LLaMA choices: [\"q_proj\", \"k_proj\", \"v_proj\", \"o_proj\", \"gate_proj\", \"up_proj\", \"down_proj\"], \ LLaMA choices: [\"q_proj\", \"k_proj\", \"v_proj\", \"o_proj\", \"gate_proj\", \"up_proj\", \"down_proj\"], \
BLOOM & Falcon & ChatGLM choices: [\"query_key_value\", \"self_attention.dense\", \"mlp.dense_h_to_4h\", \"mlp.dense_4h_to_h\"], \ BLOOM & Falcon & ChatGLM choices: [\"query_key_value\", \"dense\", \"dense_h_to_4h\", \"dense_4h_to_h\"], \
Baichuan choices: [\"W_pack\", \"o_proj\", \"gate_proj\", \"up_proj\", \"down_proj\"], \ Baichuan choices: [\"W_pack\", \"o_proj\", \"gate_proj\", \"up_proj\", \"down_proj\"], \
Qwen choices: [\"c_attn\", \"attn.c_proj\", \"w1\", \"w2\", \"mlp.c_proj\"], \ Qwen choices: [\"c_attn\", \"attn.c_proj\", \"w1\", \"w2\", \"mlp.c_proj\"], \
Phi-1.5 choices: [\"Wqkv\", \"out_proj\", \"fc1\", \"fc2\"], \ Phi-1.5 choices: [\"Wqkv\", \"out_proj\", \"fc1\", \"fc2\"], \

5
src/llmtuner/tuner/dpo/trainer.py
View File

@ -50,7 +50,7 @@ class CustomDPOTrainer(DPOTrainer):
self.ref_model = self.accelerator.prepare_model(self.ref_model, evaluation_mode=True) self.ref_model = self.accelerator.prepare_model(self.ref_model, evaluation_mode=True)
def _prepare_deepspeed(self, model: "PreTrainedModelWrapper"): def _prepare_deepspeed(self, model: "PreTrainedModelWrapper"):
# Adapted from accelerate: https://github.com/huggingface/accelerate/blob/739b135f8367becb67ffaada12fe76e3aa60fefd/src/accelerate/accelerator.py#L1473 # adapted from accelerate: https://github.com/huggingface/accelerate/blob/739b135f8367becb67ffaada12fe76e3aa60fefd/src/accelerate/accelerator.py#L1473
deepspeed_plugin = self.accelerator.state.deepspeed_plugin deepspeed_plugin = self.accelerator.state.deepspeed_plugin
config_kwargs = deepcopy(deepspeed_plugin.deepspeed_config) config_kwargs = deepcopy(deepspeed_plugin.deepspeed_config)
if model is not None: if model is not None:
@ -75,7 +75,8 @@ class CustomDPOTrainer(DPOTrainer):
# Otherwise, we assume the reference model fits in memory and is initialized on each device with ZeRO disabled (stage 0) # Otherwise, we assume the reference model fits in memory and is initialized on each device with ZeRO disabled (stage 0)
if config_kwargs["zero_optimization"]["stage"] != 3: if config_kwargs["zero_optimization"]["stage"] != 3:
config_kwargs["zero_optimization"]["stage"] = 0 config_kwargs["zero_optimization"]["stage"] = 0
# lazy load
# Lazy load
import deepspeed # type: ignore import deepspeed # type: ignore
model, *_ = deepspeed.initialize(model=model, config=config_kwargs) model, *_ = deepspeed.initialize(model=model, config=config_kwargs)
model.eval() model.eval()