Request for additional features from TE Debug modules for low precision training.
维护者通常 2 天内回复
@pggPL 已经在做这个了。
开始于 2026年3月30日。
评估
这个 Issue 还没有评估数据。
描述
Is your feature request related to a problem? Please describe.
- For LogFp4Tensor stats and LogFP8TensorStats we want to obtain the standard deviation of the scale factors.
- LogFp8 and LogFP4 tensor stats can't currently be used at the same time, which is useful for situations when you have some layers as FP8 and some as FP4.
- [most important] When using MXFP8 or NVFP4 micro-scaling methods we are only obtaining global statistics over the entire tensor instead of per each block.
Describe the solution you'd like
- We want to add the standard deviation metric for the scale factors because simply having min and max doesn't tell us enough information. We could have serious clipping going on but won't know if we only have a max value.
- Currently, I've had to hack in support by modifying yaml files online in order to perform FP8 and FP4 logging for networks that have mixed FP4/FP8 layers https://github.com/NVIDIA/bionemo-framework/blob/main/bionemo-recipes/recipes/esm2_native_te/quantization.py#L94 I am hoping this can be supported natively.
- For LogFP8TensorStats and LogFP4TensorStats specifically for micro-block formats such as MXFP8 / NVFP4 we need an optional mechanism to extract block level metadata in addition to per-tensor level metadata including scale factor statistics per block of (scale_value, block_indices, tensor_chunk) etc saved to files. By default this should be able to work on the first block (0:32 for MXFP8, 0:16 for NVFP4) unless otherwise specified etc. It would be super useful in order to identify issues on a more fine grained scale.
Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.
Additional context
Add any other context or screenshots about the feature request here.
- 主要语言
- Python
- 星标
- 3.6k
- 派生
- 872
- 平均合并
- 4 天 11 小时
- 30 天内合并 PR
- 59
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
NVIDIA/TransformerEngine 的其他 Issue
-
[PyTorch] fp8_cs_quantize fake implementation returns a vector inverse scale instead of a scalar可能重新可做 @sanjana658 于 15 天前认领，目前没有进行中的 PR。 未关闭
难度 2/5 1-3 小时 新手友好度 82/100
NVIDIA/TransformerEngine#3636 · 2 条评论 ·
维护者通常 2 天内回复
-
[Bug] Backend selection picks FA3 for training with head_dim_qk=192 / v_head_dim=128, but FA3 backward cannot run it可能重新可做 @modelpath-dev 于 49 天前认领，目前没有进行中的 PR。 未关闭attention
难度 2/5 1-3 小时 新手友好度 85/100
NVIDIA/TransformerEngine#3481 · 4 条评论 ·
维护者通常 2 天内回复
-
bug
难度 2/5 1-3 小时 新手友好度 68/100
NVIDIA/TransformerEngine#2189 · 7 条评论 · 5 个 reaction ·
维护者通常 2 天内回复
-
难度 5/5 一周以上 新手友好度 25/100
NVIDIA/TransformerEngine#3729 ·
维护者通常 2 天内回复
-
[Feature] Fused gated RMSNorm (RMSNorm(x) * SiLU(gate)) for GDN output norm可能已有人在做 关联的 PR 仍在进行中或已合并。 未关闭enhancement
难度 5/5 一周以上 新手友好度 35/100
NVIDIA/TransformerEngine#3724 · 1 个 reaction ·
维护者通常 2 天内回复
查看 NVIDIA/TransformerEngine 的全部 Issue
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 76/100
aws-samples/sample-per-model-bedrock#464 ·
维护者通常 1 天内回复
-
lesson-submission pending-review
难度 2/5 1-3 小时 新手友好度 67/100
Ikalus1988/MisakaNet#3789 ·
维护者通常 1 天内回复
-
Python Volcengine provider returns failures instead of raising them可能已有人在做 @Ricardo-M-L 今天认领。 未关闭
难度 2/5 1-3 小时 新手友好度 79/100
agent-infra/sandbox#248 ·
-
[Bug] cloud JSON directory loading assumes a Key field in fsspec listings可能已有人在做 @Ricardo-M-L 今天认领。 未关闭
难度 2/5 1-3 小时 新手友好度 88/100
hiyouga/LlamaFactory#10910 ·
维护者通常 1 天内回复
-
Bundled uv is never updated and trusts the compromised eMudhra certificates可能已有人在做 @FabienDanieau 今天认领。 未关闭bug
难度 2/5 1-3 小时 新手友好度 76/100