fmo-mt

Results 5 issues of fmo-mt

模型转换失败：error::op convert failed onnx: If

**1. 使用环境（environment）** - OS: Ubuntu - OS Version: 20.04 **2. Github版本** - branch：master - commit(optional): a05102fd7174f31111c1c6f49c77e6b829b2093b **3. 详细描述bug 情况 (Describe the bug)** - issue type: 模型转换失败（converter failed） - original model：...

Accuracy drop for Llama

9 comment

I tried to quantize a Llama model (Llama 13b) by smooth quant, and found that if I only quantize `LlamaDecoderLayer` then the accuracy would not drop even directly quantize weights...

OOP range doesn't match

preflop setting like: solver result: As you can see, we missed the AX which is set in preflop range

Build from source?

9 comment

Hey I found that the released version is kinda old and many commits were not included. Can you provide the ways we can build from source on different platforms?

load datasets error in FP8 training for BERT

this line of code https://github.com/IntelLabs/FP8-Emulation-Toolkit/blob/main/examples/training/bert/run_qa_no_trainer.py#L725 cannot work since squad or squad_v2 didn't provide an evaluate subclass.