THUDM / THUDM/slime

Question about default arg config on /scripts/run-mimo-7B-rl-eagle.sh

Open
#1,017 2 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
8.5k
Forks
1.3k
Avg merge
5h 36m
Merged PRs (30d)
22

Description

I noticed that in this shell, the SGLANG_ARGS were setted as:
SGLANG_ARGS=(
--rollout-num-gpus-per-engine 1
--sglang-mem-fraction-static 0.6
--sglang-disable-cuda-graph

--sglang-speculative-algorithm EAGLE
--sglang-speculative-num-steps 3
--sglang-speculative-eagle-topk 1
--sglang-speculative-num-draft-tokens 4
--sglang-enable-draft-weights-cpu-backup
)
In this case, --sglang-speculative-num-steps is 3 and --sglang-speculative-eagle-topk is 1, so the Draft process generates only 3 Tokens in total. However, --sglang-speculative-num-draft-tokens is 4, so the system is configured to handle 4 tokens for parallel verification, leaving the extra capacity idle.
While this might not actually waste resources on the system, it can still be confusing for beginners like me. I recommend modifying the default configuration, such as:
--sglang-speculative-num-steps 3
--sglang-speculative-eagle-topk 2
--sglang-speculative-num-draft-tokens 5

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Open scripts/run-mimo-7B-rl-eagle.sh and read the SGLANG_ARGS defaults first. Compare the speculative step, top-k, and draft-token settings with the proposed values, then verify that the script reflects the intended configuration and remains runnable.

Written by the indexing model from the issue text.

Assessment

Tech stack
shell
Domain
tooling
Issue type
Feature
Difficulty
1/5
Estimated time
Under an hour
Activity status
Stale
Clarity
Clearly specified
Newbie friendliness
50/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.