Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 1 addition & 2 deletions trl/trainer/grpo_config.py
Original file line number Diff line number Diff line change
Expand Up @@ -328,8 +328,7 @@ class GRPOConfig(_BaseConfig):
Specifies how Importance Sampling is performed when `vllm_importance_sampling_correction=True`. Possible
values are:

- `"token_truncate"`: Token-level truncated IS (default). Per-token ratios are clipped to
[C_min, C_max].
- `"token_truncate"`: Token-level truncated IS. Per-token ratios are clipped to [C_min, C_max].
- `"token_mask"`: Token-level masked IS. Per-token ratios outside [C_min, C_max] are set to zero.
- `"sequence_truncate"`: Sequence-level truncated IS. A single sequence ratio is clipped to
[C_min, C_max] and applied to all tokens in the sequence.
Expand Down
Loading