Yinlei Sun

[v1] Add support for ShareGPT format.

2

# What does this PR do? This PR handles data conversion: Adds logic to process the "history" in Alpaca samples, implements converting ShareGPT format samples to SFT training format (with...

## Summary This PR is the first step in the adaptation of Ascend NPU to Liger Kernel: adding NPU device support. For details, refer to [[RFC] Native Ascend NPU Support...

[RFC] Native Ascend NPU Support for Liger Kernel

1

## 1. Background & Motivation Ascend NPU is a default PyTorch device backend, natively compatible with ecosystems like Transformers, FlagGems, and Llama Factory. We’re also enabling Triton support (repo: [triton-ascend](https://gitcode.com/Ascend/triton-ascend))....

[Fix] Fix the teacher_model is not automatically placed on the right device

2

## Motivation In the model LAD, self.teacher_model is not placed in nn.ModuleList, which makes the module cannot automatically run on the Ascend NPU. ## Modification Add self.teacher_model to nn.ModuleList so...