All pull requests

Pull requests

• Fix fp16 precision loss in Qwen2/Qwen2.5-VL vision RoPE table

  #822·tritsystem opened Oct 3, 2026·

  ·

• Learn the last token in the ngram cache

  #821·SashaMIT opened Sep 28, 2026·1 task done·

  ·

• fix: compile error on RDNA4 (gfx1201) ROCm

  #815·Kaihui-AMD opened May 15, 2026·

  ·
  1

• Fix ROCm 7.0 build

  #807·amd-vlarakic opened Sep 25, 2025·

  ·
  1

• DRY: Fix Greedy Clone typo in sampler.

  #801·Ph0rk0z opened Jul 28, 2025·

  ·

• small tensor parallelism optimization

  #796·Alex-GZ opened Jun 14, 2025·

  ·
  2

• Hacky tensor parallel for Vision Models

  #791·Ph0rk0z opened May 25, 2025·

  ·

• Fix ext.py silently ignoring ImportErrors

  #774·ThisIsPIRI opened Apr 24, 2025·

  ·

• Llama-3_1-Nemotron 51B support

  #726·ymcki opened Jan 28, 2025·

  ·
  12

• Add more choices to quantization tool. Post processing after sim_anneal(). (optimizer.py/ext_quant.cpp)

  #712·Originalimoc opened Jan 10, 2025·

  ·
  2

• Add ExLlamaV2Sampler.Settings.logits_processor

  #634·lapp0 opened Sep 23, 2024·

  ·
  4

• [kernel] switch from PYBIND11 to TORCH_LIBRARY

  #617·AlpinDale opened Sep 6, 2024Contributor·

  ·

• added option for tokenized input to dynamic generator

  #613·KT313 opened Sep 2, 2024·

  ·

• Add batch processing to MMLU and Humaneval evaluation scripts to prevent OOM errors

  #597·LlamaEnjoyer opened Aug 23, 2024Contributor·

  ·
  8

• Adding stream to 1 kernel.

  #590·Narsil opened Aug 14, 2024·

  ·
  4

• Simple QuaRot proof of concept.

  #407·sgsdxzy opened Apr 11, 2024·

  ·
  5

• Refactor token healing initialization.

  #330·bjj opened Feb 10, 2024·

  ·
  7

• Update setup.py UX responses and module checks with helpful messaging

  #277·bgorlick opened Jan 15, 2024·

  ·
  4

• Repeat layers to create FrankenModels

  #275·dnhkng opened Jan 12, 2024·

  ·
  35

• add QuiP quant support

  #217·zumi87 opened Dec 7, 2023·

  ·
  3

• Adding return_lowest_perplexity

  #206·ziadloo opened Dec 3, 2023·

  ·

• Add copilot server example

  #23·chenhunghan opened Sep 13, 2023·

  ·
  9