ByteDance-Seed / ByteDance-Seed/decoupleQ

LinearA16 and LinearW2A16 output does not match

Open
#12 6 comments 0 reactions 0 assignees View on GitHub
Dominant language
Cuda
Stars
153
Forks
10
PR merge metrics
No merged PRs in 30d

Description

使用decoupleQ压缩llama-3-8b,使用 fake_quant - linearA16 的计算结果和true_quant - linearW2A16的计算结果不match
截图中,左边是linearW2A16的输出,右边是linearA16的输出
input数据类型是bfloat16

请问这种情况可能的原因是什么呀?

image

Contributor guide

No contributing guide indexed for this repository

Research direction

Start by reproducing the llama-3-8b decoupleQ case described in the issue, comparing fake_quant with linearA16 against true_quant with linearW2A16. Check the behavior with bfloat16 input and use the attached screenshot as the expected mismatch reference. Done means identifying and documenting the cause of the differing outputs.

Written by the indexing model from the issue text.

Assessment

Domain
machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.