Handle mtp `prefix/files`in `out_of_model_tensors` (#2677)

[CI] use same cuda version for CI & torch (#2676)

Extend OpenVINO's GPTQ patcher to understand GPTQModel new kernels. (#2675)

fix failed test (#2674)

fix marlin import paths probe (#2673)

remove hf kernels dependency for cpu (#2671)

Kernels migrate to jit compile (#2670)

Move more kernels to JIT compile path (#2668)

[CI] remove verbose log flag for build (#2669)

[CI] adjust venv logic (#2667)

Refactor input capture flow into BaseQModel and model-specific QModels (#2666)

Update README.md (#2665)

fix cpu fallback device restore (#2664)

add gemma4 (#2663)

Update README.md (#2662)