Tavondo
CompetitionsEventsGuidesPlatforms
中文Start matching
CompetitionsEventsGuidesPlatforms
中文Sign in
Tavondo

Find AI competitions and events worth your time.

CompetitionsEventsGuidesPlatforms
Sign inRegisterStart matching

© 2026 Tavondo

中文
  1. Home
  2. Competitions
  3. 指令微调性能优化挑战赛
OpenAtomLLMIntermediateRegistration unverified

指令微调性能优化挑战赛

A intermediate llm competition requiring roughly 12–40 hours for an end-to-end practice run.

Difficulty
Intermediate
Estimated effort
12–40 hours
Registration deadline
To be confirmed
Compute
CPU only
OverviewPreparationSource

Overview

Who this competition fits

适合已经做过大语言模型指令微调、能在华为昇腾环境中使用MindSpeed-LLM的开发者或研究团队。参赛者还需要能准备合规的开源指令数据,并交付可运行代码、训练日志和微调后的模型文件。

Read the original official blurb
通过调整微调训练策略、优化算子性能、开发高效的训练算法,提升模型在推理任务上的对话表现和评估任务上的性能表现,为大模型领域提供新的解决方案和思路。 Official category: 实战竞技赛. Organizers: 华为技术有限公司. Participants: 5.

Preparation

From registration to a first submission

  1. 01

    MindSpeed-LLM

  2. 02

    昇腾芯片训练环境

  3. 03

    Llama2-7B全参数微调

  4. 04

    分布式并行策略

  5. 05

    指令数据集构建

  6. 06

    MMLU评估

Before you commit: 难点在于同时优化全参SFT的训练吞吐和模型效果:既要通过并行策略、算子或算法优化提高样本处理速度,又要兼顾MMLU与多轮对话表现。提交不仅是结果,还需提供数据集链接、完整流程文档、可跑通代码、日志和模型检查点,并保证作品可脱离原开发环境运行。

Source

How this page was assembled

Competition information is structured from the official page. Scores are platform estimates for decision support; official rules take precedence.

Official competition page
OpenAtom
Last checked
Dec 3, 2024

Entry assessment

est.
48

Prepare before joining

Overall fit estimate

Beginner fit48/100
Learning value60/100

Confirm eligibility and time commitment before moving into formal preparation.

Entry eligibility needs official confirmation

Open official pageSign in to saveOpen guide

Data and review state

Separate what is confirmed from what still needs review. Official rules and deadlines win.

View data stateClose
Source page verifiedMetric confirmedReviewed
Source published
Dec 3, 2024
Source created
Dec 3, 2024
Published on this site
Jul 21, 2026

Official metric is not published on the OpenAtom competition listing; verify scoring on the competition page.

reward_value_usd is a rough CNY→USD estimate (×0.14) for ranking only; use reward_summary for the official prize text.