Skip to content

[case] domux-4057-rtx3090-eval - 4057-sample BF16 evaluation and failure analysis on RTX 3090Her hack astron 4 4057 eval - #23

Open
posuizhiyu-maker wants to merge 2 commits into
iflytek:mainfrom
posuizhiyu-maker:her-hack-astron-4-4057-eval
Open

[case] domux-4057-rtx3090-eval - 4057-sample BF16 evaluation and failure analysis on RTX 3090Her hack astron 4 4057 eval#23
posuizhiyu-maker wants to merge 2 commits into
iflytek:mainfrom
posuizhiyu-maker:her-hack-astron-4-4057-eval

Conversation

@posuizhiyu-maker

Copy link
Copy Markdown

Ref #20

Reproducible scripts, final case README, and sanitized evidence preview. Ref iflytek#20.

Signed-off-by: posuizhiyu-maker <posuizhiyu@gmail.com>
Aggregate metrics, sanitized logs, latency evidence, and 15 representative failures. Ref iflytek#20.

Signed-off-by: posuizhiyu-maker <posuizhiyu@gmail.com>
@CLAassistant

CLAassistant commented Aug 25, 2026

Copy link
Copy Markdown

CLA assistant check
All committers have signed the CLA.

@FenjuFu

FenjuFu commented Aug 26, 2026

Copy link
Copy Markdown
Member

已按 #20 的 10 条验收标准核对本案例:

  • 真实使用 / 真实运行 / 可复现 ✓:revision 已固定,snapshot_download(repo_id, revision=...) 调用明确,完整 artifacts/(environment、eval/latency/smoke 的 console + json、failure_analysis)+ RUNBOOK + 参数化脚本齐全。
  • 唯一渠道 ✓:Discussion #3 已公开可达,标题合规,channels 匹配。
  • 指标诚实 ✓:4057 样本、各类别 F1、E2E 延迟均标注方法与环境;明示 “no model weights are included”。
  • 包含边界 ✓:失败簇分析到位。
  • 隐私与安全 ✓:核对 environment.txt / download_metadata.json,未见 token / 内网 IP / 家目录路径。
  • 账号信誉 ✓:仅本仓一个 PR。

综合:三份里工程最扎实、可复现性最强的一份。最终评奖由 @FenjuFu 统一裁定。

@FenjuFu

FenjuFu commented Aug 29, 2026

Copy link
Copy Markdown
Member

?????? head 22576e3?????????????????? revision?4,057 ??? BF16 ???RTX 3090 ???RUNBOOK?????????/???/??????????????Discussion #3 ???????????token?????????????DCO ? CLA ????

????????????????????????????????????????????

@FenjuFu

FenjuFu commented Aug 29, 2026

Copy link
Copy Markdown
Member

复审确认(评奖前最终核对):自上轮评审后无新提交,此前核过的验收依然成立——固定 revision、RTX 3090 BF16 完整 artifacts + RUNBOOK + 参数化脚本、明示未含权重、HF Discussion #3 公开、失败簇分析。可复现性与工程完整度是本批最扎实的之一。有效案例。

最终评奖将在活动评奖期(8-31 – 9-6)统一公布。

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.

3 participants