Skip to content

Make Megatron RL only materialize last token logit #4551

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
tdene merged 1 commit into from
Jun 16, 2026
+5 −1
Merged

Make Megatron RL only materialize last token logit #4551

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
2 changes: 1 addition & 1 deletion megatron/inference/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -367,7 +367,7 @@ def get_inference_config_from_model_and_args(model: MegatronModule, args):
mamba_inference_state_config=mamba_inference_state_config,
pg_collection=pg_collection,
use_flashinfer_fused_rope=args.use_flashinfer_fused_rope,
materialize_only_last_token_logits=(not args.return_log_probs),
materialize_only_last_token_logits=not (args.return_log_probs and not args.skip_prompt_log_probs),
track_generated_token_events=args.inference_dynamic_batching_track_generated_token_events,
track_paused_request_events=args.inference_dynamic_batching_track_paused_request_events,
enable_chunked_prefill=args.enable_chunked_prefill,
Expand Down
4 changes: 4 additions & 0 deletions megatron/rl/inference/megatron.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -97,6 +97,10 @@ async def launch(cls, model: GPTModel, **kwargs):
"WARNING: Tokenizer has no BOS token so prompt will not have BOS token",
)

# RL needs log probs, but not prompt log probs.
args.return_log_probs = True
args.skip_prompt_log_probs = True

inference_engine: DynamicInferenceEngine = get_dynamic_inference_engine(model=model)
dp_addr = await inference_engine.start_listening_to_data_parallel_coordinator(
inference_coordinator_port=41521, launch_inference_coordinator=True,
Expand Down
Loading