ros/LLaMA-Factory
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix ppo runtime error

Browse Source 
Former-commit-id: 14e2f221e3e720075e59065a3dc42aa4d993a8b6
This commit is contained in:
stephen
2024-03-08 11:48:26 +08:00
parent d46c2bbcba
commit 18cfd5f349
2 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/llmtuner/model/patcher.py
View File

@@ -264,6 +264,7 @@ def patch_config(
config: "PretrainedConfig",
tokenizer: "PreTrainedTokenizer",
model_args: "ModelArguments",
finetuning_args: "FinetuningArguments",
init_kwargs: Dict[str, Any],
is_trainable: bool,
) -> None:
@@ -288,7 +289,8 @@ def patch_config(
if not is_deepspeed_zero3_enabled():
init_kwargs["low_cpu_mem_usage"] = model_args.low_cpu_mem_usage
if "device_map" not in init_kwargs: # quant models cannot use auto device map
init_kwargs["device_map"] = model_args.device_map or {"": get_current_device()}
if finetuning_args.stage not in ["ppo"]: #ppo stage should not set device map
init_kwargs["device_map"] = model_args.device_map or {"": get_current_device()}
def patch_model(