Commit History

Autor SHA1 Mensaxe Data
  AlpinDale 9d81716bfd [v0.5.3] Release Candidate (#388) hai 8 meses
  AlpinDale e702f587cf feat: add batched RoPE kernels (#371) hai 9 meses
  AlpinDale 3d6695cfbb feat: add approximate gelu activation kernels (#370) hai 9 meses
  AlpinDale f8dfac6372 chore: attention refactor and upstream sync apr01 (#365) hai 9 meses
  AlpinDale c41462cfcd feat: exllamav2 quantization (#305) hai 10 meses
  AlpinDale 9810daa699 feat: INT8 KV Cache (#298) hai 10 meses
  AlpinDale e0c35bb353 feat: bitsandbytes and `--load-in{4,8}bit` support (#294) hai 10 meses
  AlpinDale 705821a7fe feat: AQLM quantization support (#293) hai 10 meses
  AlpinDale 72229a94da feat: better marlin kernels (#285) hai 10 meses
  AlpinDale e31c6f0b45 feat: refactor modeling logic and support more models (#274) hai 10 meses
  AlpinDale 224b87b484 feat: add fused mixtral moe support (#238) hai 10 meses
  AlpinDale d9b65e6c5f feat: DeepSeek MoE support (#237) hai 11 meses
  AlpinDale c3a221eb02 feat: GGUF, QuIP#, and Marlin support (#228) hai 11 meses
  AlpinDale 31c95011a6 feat: FP8 E5M2 KV Cache (#226) hai 11 meses
  AlpinDale 641bb0f6e9 feat: add custom allreduce kernels (#224) hai 11 meses
  AlpinDale 5053743c1c feat: speedup AWQ (#223) hai 11 meses
  AlpinDale 8fa608aeb7 feat: replace Ray with NCCL for control plane comms (#221) hai 11 meses
  AlpinDale 7e72ce0a73 feat: mixtral tensor parallelism (#193) hai 1 ano
  AlpinDale 15a0454172 feat: FP8 KV Cache (#185) hai 1 ano
  AlpinDale 801eda0b7a feat: support GPTQ 2, 3, and 8bit quants (#181) hai 1 ano
  AlpinDale 7c6fdea535 fix: GPTQ warnings and exllama states (#171) hai 1 ano
  AlpinDale 02f3ab3501 fix: replace head_mapping with num_kv_heads (#161) hai 1 ano
  AlpinDale 62b2c4119d feat: re-write GPTQ and refactor exllama kernels (#152) hai 1 ano
  AlpinDale 1334a833a4 feat: AMD ROCm support (#95) hai 1 ano
  AlpinDale 2b1ba581f9 feat: re-implement GPTQ (#141) hai 1 ano
  AlpinDale 8223f85c1b feat: SqueezeLLM support (#140) hai 1 ano
  AlpinDale 1aab8a7d6f feat: speedup compilation times by 3x (#130) hai 1 ano