[A2-K] 刘子源 - 完成显存优化与 GPU 内核作业 - #156
Open
Zyrrick wants to merge 1 commit into
Open
Conversation
WillQvQ
requested changes
Aug 8, 2026
WillQvQ
left a comment
Contributor
There was a problem hiding this comment.
审核结论:需要修改后再合并。
代码范围、Python 解析、真实 Triton forward、37 项扩展正确性、6 项 CUDA tests、矩阵形状、23 GiB allocator guard、附件与 git diff --check 均通过;但正式性能证据当前无效:
run_metadata.json与 README 明确记录全部结果来自 RTX 3090,而 A2-K 正式 benchmark 的标准环境是单张 RTX 4090 24GB;PR checklist 中该项也未勾选。RTX 3090 的开发结果可保留,但不能替代正式矩阵。free_memory_mib_at_start、power limit 与 P-state 均缺失,其中开跑前空闲显存不少于 22 GiB 是硬验收字段;memory_evidence.json也不能用 preflight requirement 常量代替实际观测值。- metadata 明确说明 attention 只用了 5 次 warm-up、10 次重复以及同步
perf_counter,未按固定do_bench(warmup=100, rep=300)协议执行;因此 p20/p50/p80 与 speedup 不能作为正式结果。
请在 RTX 4090 24GB 上、开跑前记录实际空闲显存,并在首次 CUDA allocation 前设置 23,552 MiB limit;按固定 do_bench 协议串行独立进程重跑 checkpoint/attention/compile/Flash 正式矩阵,更新 CSV、metadata、memory evidence、图和 README。现有 3090 数据可以明确标为开发结果保留,但不能再称“正式实验”。
审核者:GPT-5.6 Sol Ultra
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
PR Checklist
基本信息
修改范围
students/_template、公共题面或仓库配置。[A编号] 姓名 - 简短说明。公开性与安全
git diff --cached,没有提交大型数据、模型权重、缓存或完整日志。双层提交
README.md提供可公开、已脱敏的作业报告。README.md已填写飞书补充文档链接。A2-K 额外检查
若本 PR 不是 A2-K,请将本节标记为不适用。A2-K 已正式发布,按题面要求提交。
README.md、A2-K Python allowlist、全部规定的轻量results/和至少两张被报告引用的压缩图片。
scripts/sync_a2k_submission.py、官方 GPU tests 和scripts/validate_repo.py,且没有把 CUDA skip 写成 pass。前设置 23552 MiB allocator 上限。
memory_evidence.json已记录 allocator limit/fraction、peak allocated/reserved 和24 GiB 判定。
results/与assets/合计不超过 2 MiB,关键数字可回到轻量结果与命令。给助教的说明
复旦集群长时间没有空闲的 RTX 4090 单卡,因此本次 A2-K 使用单张 RTX 3090 24GB 完成实验。