Skip to content

code sync#1329

Merged
zhuzilin merged 7 commits intoTHUDM:mainfrom
miles-code-angel:auto/20260104222228
Jan 5, 2026
Merged

code sync#1329
zhuzilin merged 7 commits intoTHUDM:mainfrom
miles-code-angel:auto/20260104222228

Conversation

@miles-code-angel
Copy link
Contributor

This PR sync these three commits:

  1. Cherry Pick commits to local fix CI unit tests radixark/miles#393 : several tiny fixs for CI running locally.
  2. [example] Add SWE-agent example radixark/miles#367 SWE bench example, verified by Yueming and Zijie.
  3. add background health check to miles native router radixark/miles#260 slime native router's health check. Verified by Chenyang and Chengxing.

Thanks!

miles-code-angel and others added 4 commits January 4, 2026 22:22
…HUDM#260)

Co-authored-by: zhaochenyang20 <zhaochen20@outlook.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: zijiexia <zijie_xia@icloud.com>
Co-authored-by: zhaochenyang20 <zhaochen20@outlook.com>
…#393)

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
@miles-code-angel miles-code-angel changed the title Auto/20260104222228 code sync Jan 5, 2026
if args.fp16:
kwargs["dtype"] = "float16"
elif args.bf16:
kwargs["dtype"] = "bfloat16"
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think we can use the default value from sglang because when using fp8 in megatron, the args.bf16 will still be True.

sum(response_lengths) / rollout_time / args.rollout_num_gpus
)
log_dict["perf/longest_effective_sample_tokens_per_sec"] = max(response_lengths) / rollout_time

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

this part has been merged into compute_perf_metrics_from_samples.

log_dict = {}
log_dict |= dict_add_prefix(compute_statistics(response_lengths), "response_len/")
log_dict |= _compute_zero_std_metrics(args, samples)
log_dict |= _compute_spec_metrics(args, samples)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

same as above.

)
metrics["rollout/spec_accept_length"] = (
sum(sample.spec_info.spec_accept_length for sample in all_samples) / num_samples
)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

same as above.

# Precision
gradient_checkpointing: bool = False
fp16: bool = False
bf16: bool = False
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

hmm, I don't think bf16 param for fsdp is necessary because in the megatron side, the bf16 is a default setting.

@zhuzilin zhuzilin merged commit 1c81e5d into THUDM:main Jan 5, 2026
18 checks passed
@zhaochenyang20 zhaochenyang20 deleted the auto/20260104222228 branch January 5, 2026 07:57
kafkayu pushed a commit to kafkayu/slime that referenced this pull request Jan 8, 2026
Co-authored-by: zhaochenyang20 <zhaochen20@outlook.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: zijiexia <zijie_xia@icloud.com>
Yangruipis pushed a commit to rednote-ai/slime that referenced this pull request Feb 28, 2026
Co-authored-by: zhaochenyang20 <zhaochen20@outlook.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: zijiexia <zijie_xia@icloud.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants