main
[generate] Never use `cache_position` anymore in generation (#44816)
Fix AWQ tests for GPTQModel migration (#44654)
Update AutoRound XPU/CPU backend (#47826)
Fix BitNet packed-weight unpacking dtype (`F.linear` dtype mismatch) (#46808)
Bump torchao >=0.15 and fix quantization CI (#44604)
Fix compressed tensors impl (#47652)
[loading] Really initialize on meta device for huge perf gains (#42941)
fix fbgemm fp8 multi-device load failure. (#43581)
Fix deepgemm on multiple devices (#47323)
Update quants tests (#45480)
Quantization for small models (#46449)
Add full GGUF loading support for GPT‑OSS (fixes #43366, supersedes #43757) latest (#45506)
extend tests/quantization/gptq/test_gptq.py::GPTQTestCUDA and tests/q… (#47166)
Use `HF_TOKEN` directly and remove `require_read_token` (#43233)
Fix tests quantization (#42703)
SINQ quantization strategy integration (adapted for Transformers V5) (#43112)
Migrate torchao integration off deleted torchao.dtypes (#47797)