7 Pull requests merged by 2 people

• Improvements to learned round

  #1107 merged Dec 3, 2024

• Hierarchical scales

  #1038 merged Dec 3, 2024

• Fix (hadamard): remove hadamard loading warning

  #1108 merged Dec 3, 2024

• Feat auto round

  #1064 merged Dec 3, 2024

• Feat (brevitas_examples/llm): remove dependencies from optimum-amd

  #1094 merged Dec 2, 2024

• Fix LLM tests

  #1088 merged Nov 28, 2024

• Feat: Per-Row po2 float ocp

  #1102 merged Nov 27, 2024

3 Pull requests opened by 1 person

• [DRAFT] Handle inplace operations in forward when offloading

  #1103 opened Nov 27, 2024

• [DRAFT] LLM benchmark

  #1104 opened Nov 27, 2024

• [DO NOT MERGE] MSE optimal scale

  #1105 opened Dec 2, 2024

1 Issue closed by 1 person

• Doubt about the __add__ implementation of the IntQuantTensor

  #1106 closed Dec 3, 2024

4 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Feat (nn/sdpa): quantization of scaled dot-product attention

  #1090 commented on Dec 3, 2024 • 2 new comments

• RuntimeError: Module <class 'brevitas.proxy.float_runtime_quant.ActFloatQuantProxyFromInjector'> not supported for export

  #1091 commented on Nov 29, 2024 • 0 new comments

• Feat (export): dequantize during export

  #1083 commented on Nov 28, 2024 • 0 new comments

• Setup: unpin Numpy

  #1093 commented on Nov 28, 2024 • 0 new comments