Commit History

Autor SHA1 Mensaxe Data
  AlpinDale cda0e93a10 abstract away the platform for device capability hai 7 meses
  AlpinDale 0f4a9ee77b quantized lm_head (#582) hai 7 meses
  AlpinDale 7d79c0e726 chore: use nvml query to avoid accidental cuda initialization hai 7 meses
  AlpinDale 156f577f79 feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569) hai 7 meses
  AlpinDale 5cedee9024 fix gemma with gptq marlin hai 7 meses
  AlpinDale 5b0c11d190 support pipeline parallel pynccl groups hai 8 meses
  AlpinDale c66b1b57b1 Marlin 2:4 sparsity (#555) hai 8 meses
  AlpinDale ad1c6b86a1 gptq_marlin: enable bfloat16 hai 8 meses
  AlpinDale c154578c97 gptq_marlin: 8bit GPTQ support hai 8 meses
  AlpinDale ac5b4b6aa7 broadcast metadata through cpu hai 8 meses
  AlpinDale f22b700ee4 feat: marlin kernels for GPTQ (#547) hai 8 meses