423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-08-22 13:42:51 +08:00
Code Issues Packages Projects Releases Wiki Activity

10x generate in ppo w/ zero3

Browse Source 
https://github.com/huggingface/trl/pull/1483

Former-commit-id: 65cd8bdbdbe1b19250ecd813aeb72c8e00ef2f9c
This commit is contained in:
hiyouga 2024-05-29 00:23:23 +08:00
parent bfac965f9c
commit 468d0e7ed1
1 changed files with 5 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
src/llamafactory/train/ppo/trainer.py
View File

@ -13,6 +13,7 @@ from transformers.trainer_utils import PREFIX_CHECKPOINT_DIR
from transformers.utils import SAFE_WEIGHTS_NAME, WEIGHTS_NAME
from trl import PPOConfig, PPOTrainer
from trl.core import PPODecorators, logprobs_from_logits
from trl.models.utils import unwrap_model_for_generation
from ...extras.callbacks import FixValueHeadModelCallback, LogCallback
from ...extras.logging import get_logger
@ -322,7 +323,7 @@ class CustomPPOTrainer(PPOTrainer, Trainer):
for k, v in batch.items():
batch[k] = v[:, start_index:]
unwrapped_model: "AutoModelForCausalLMWithValueHead" = self.accelerator.unwrap_model(self.model)
with unwrap_model_for_generation(self.model, self.accelerator) as unwrapped_model:
generate_output: torch.Tensor = unwrapped_model.generate(
generation_config=self.generation_config, logits_processor=get_logits_processor(), **batch
)