3 Pull requests merged by 3 people

• add slack channel to readme

  #1340 merged Nov 2, 2024

• Update packaging in AOTI path

  #896 merged Nov 1, 2024

• [distributed] Add Llama3-70B for distributed inference

  #1335 merged Oct 30, 2024

4 Pull requests opened by 4 people

• Deprecating Int8DynActInt4WeightQuantizer

  #1332 opened Oct 28, 2024

• Granite code support

  #1336 opened Oct 31, 2024

• [AOTI] Remove the original model weights in Python deployment

  #1337 opened Nov 1, 2024

• tokenizer was missing an include

  #1339 opened Nov 1, 2024

3 Issues closed by 3 people

• can't build AOTI runner

  #1338 closed Nov 1, 2024

• Llama 3.2 MM Multiturn Browser: Second message errors out

  #1224 closed Oct 31, 2024

• x86 CPU: BF16 should improve decoding performance relative to FP32 on x86, even without hardware BF16

  #1253 closed Oct 30, 2024

2 Issues opened by 2 people

• RFC: Multimodal Eval Enablement (Looking for Developer to Implement Design)

  #1334 opened Oct 29, 2024

• RFC: Code sharing for ET export, C runner and tokenizer, with ExecuTorch

  #1333 opened Oct 28, 2024

11 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.

• Replace WeightOnlyInt8Linear with TorchAO int8_weight_only quantization

  #1328 commented on Oct 31, 2024 • 3 new comments

• RFC: Make quantization a first class feature

  #1032 commented on Oct 30, 2024 • 0 new comments

• RFC: Quantization Evaluation

  #1325 commented on Oct 30, 2024 • 0 new comments

• convert_hf_checkpoint only relies on model_name to resolve TransformerArgs

  #1179 commented on Oct 30, 2024 • 0 new comments

• AOTI Export ignores user --device flag - expected behavior?

  #1278 commented on Nov 1, 2024 • 0 new comments

• Out of memory AOTI using llama 3.1 8b on RTX 4090

  #1302 commented on Nov 1, 2024 • 0 new comments

• Add benchmarking scripts

  #1030 commented on Oct 31, 2024 • 0 new comments

• [aoti] Remove need for -l in cmake call

  #1159 commented on Oct 31, 2024 • 0 new comments

• Tokenizers tokenizer

  #1261 commented on Oct 30, 2024 • 0 new comments

• Install ET nightly and bump up ET version to 20241101

  #1312 commented on Nov 1, 2024 • 0 new comments

• Use training IR in torchchat export

  #1319 commented on Nov 2, 2024 • 0 new comments