Skip to content

Conversation

@WHoutstanding
Copy link

@WHoutstanding WHoutstanding commented Jan 18, 2026

姓名

王豪

实习项目

GraphNet 计算子图构建与推广

本周工作

  1. 修复dtype_generalizer.py的bug:在判断计算图need_rewrite之前, parse_immutable_model_path_into_sole_graph_module()返回的trace_model不包含tensor_meta导致dtype_pass.need_rewrite(traced_model)返回总是false,在need_rewrite之前为trace_model加入tensor_meta,因此调用ShapeProp(traced_model).propagate(*inputs)

  2. 将dtype_gen_test.sh改为标准脚本格式并对samll100_samples.txt的进行dtype generalizaion pass测试,并用 graph_net.torch.validate 验证生成的 samples of dtype generalization pass

  3. 把dtype_generalizer.py做成了SamplePass的子类并放到torch/sample_pass目录下。

下周工作

  1. 调试分析未通过的sample的原因
  2. 开一个关于数据类型泛化的讨论区,并讨论如何集成进加工流程。

详细周报链接

dbdf5b0

@WHoutstanding WHoutstanding marked this pull request as draft January 18, 2026 01:58
@WHoutstanding WHoutstanding marked this pull request as ready for review January 18, 2026 01:59
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant