Skip to content

[WIP] Longbench v2 for qwem/ds#2406

Open
yiliu30 wants to merge 18 commits intomasterfrom
longbench
Open

[WIP] Longbench v2 for qwem/ds#2406
yiliu30 wants to merge 18 commits intomasterfrom
longbench

Conversation

@yiliu30
Copy link
Contributor

@yiliu30 yiliu30 commented Feb 13, 2026

  • QWen
  • DS

Type of Change

feature or bug fix or documentation or validation or others
API changed or not

Description

detail description

Expected Behavior & Potential Risk

the expected behavior that triggered by this PR

How has this PR been tested?

how to reproduce the test (including hardware information)

Dependency Change?

any library dependency introduced or removed

Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
* Add LongBench v2 support for DeepSeek

- Add long-bench-eval dependency to requirements.txt
- Refactor run_evaluation.sh to support both standard and LongBench v2 tasks
- Add dynamic max_length configuration (40960 for longbench tasks)
- Implement vLLM server-based evaluation for LongBench
- Add helper functions for server lifecycle management
- Support up to 40K context length evaluation with 512 threads

Signed-off-by: yiliu30 <[email protected]>

* update max len

Signed-off-by: yiliu30 <[email protected]>

---------

Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
Signed-off-by: yiliu30 <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant