tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

seperately include max_completion_len

Browse Source
This commit is contained in:
Wing Lian
2025-02-05 13:01:52 -05:00
parent b8f258817e
commit a1958b09de
2 changed files with 9 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
src/axolotl/core/trainers/grpo/__init__.py
View File

@@ -43,6 +43,9 @@ class GRPOStrategy:
] = cfg.grpo_vllm_gpu_memory_utilization ] = cfg.grpo_vllm_gpu_memory_utilization
if cfg.grpo_num_generations: if cfg.grpo_num_generations:
grpo_args_kwargs["num_generations"] = cfg.grpo_num_generations grpo_args_kwargs["num_generations"] = cfg.grpo_num_generations
grpo_args_kwargs["max_completion_length"] = (
cfg.max_completion_length or cfg.sequence_len
)
return grpo_args_kwargs return grpo_args_kwargs
@classmethod @classmethod

6
src/axolotl/utils/config/models/input/v0_4_1/__init__.py
View File

@@ -757,6 +757,12 @@ class AxolotlInputConfig(
default=512, default=512,
json_schema_extra={"description": "maximum prompt length for RL training"}, json_schema_extra={"description": "maximum prompt length for RL training"},
) )
max_completion_length: Optional[int] = Field(
default=None,
json_schema_extra={
"description": "Maximum length of the completion for RL training"
},
)
sample_packing: Optional[bool] = None sample_packing: Optional[bool] = None
sample_packing_group_size: Optional[int] = 100_000 sample_packing_group_size: Optional[int] = 100_000
sample_packing_bin_size: Optional[int] = 200 sample_packing_bin_size: Optional[int] = 200