All pull requests

Pull requests

• fix: handle incomplete and non-base64 image data URLs

  #2413·kevin9327 opened Nov 12, 2026·

  ·

• fix(chat): identify remote image requests with a user agent

  #2411·hakarrr opened Nov 10, 2026·

  ·
  1

• perf(cache): index RAM cache longest-prefix lookups

  #2410·hakarrr opened Nov 10, 2026·

  ·

• fix(index): include available SHA256 hashes in wheel links

  #2409·hakarrr opened Nov 10, 2026·

  ·

• docs: fix LlamaIndex and llama.h documentation links

  #2408·cloudgpuapp opened Nov 10, 2026·

  ·

• perf(speculative): speed up prompt lookup -- 71% less time

  #2404·dexhunter opened Nov 6, 2026·

  ·

• fix: limit logits processors to the sampled prefix

  #2401·YaochengTong opened Nov 4, 2026·

  ·
  1

• fix(cache): honor configured disk cache capacity

  #2400·DiadkoShmek opened Nov 2, 2026·

  ·
  1

• fix: hide unverified draft tokens from logits processors

  #2399·Syedmudassir1604 opened Nov 1, 2026·4 of 5 tasks·

  ·
  1

  2

• feat: update llama.cpp to b12032 (dde879bc4)

  #2398·lpalbou opened Nov 1, 2026·

  ·

• fix: preserve native cleanup callbacks during interpreter shutdown

  #2397·emme1t opened Nov 1, 2026·

  ·
  2

• fix: load cached pretrained models in offline mode

  #2396·emme1t opened Nov 1, 2026·

  ·
  1

• fix: preserve sequence offsets in unpooled batched embeddings

  #2395·linjiaye929 opened Oct 31, 2026·

  ·

• docs: explain shared Llama instance concurrency

  #2394·karthiksenv opened Oct 29, 2026·

  ·
  1

• docs(server): explain streaming request interruptions

  #2391·original4422 opened Oct 26, 2026·

  ·

• feat: improve server robustness, context manager support, and fix windows dll handling

  #2390·SiddardhaShayini opened Oct 22, 2026·

  ·

• fix(tokenizer): honor add_bos in HF tokenization

  #2389·dajiaohuang opened Oct 21, 2026·

  ·
  1

• fix: resize detokenization buffer for long token pieces

  #2388·grinder81 opened Oct 21, 2026·

  ·

• fix(embed): respect physical batch limits for non-causal embeddings

  #2387·original4422 opened Oct 17, 2026·

  ·
  1

  1

• fix(server): accept max_completion_tokens in chat requests

  #2386·laulauraluo opened Oct 15, 2026·

  ·
  1

• fix: respect CPU affinity and cgroup CPU quota for default thread counts

  #2385·manunicholasjacob opened Oct 14, 2026·

  ·

• fix: correctly size Llama.scores when draft_model implies logits_all

  #2384·yadukpb opened Oct 13, 2026·

  ·

• fix: retain end-of-generation token logprobs

  #2383·Kumar-Abhimanyu opened Oct 13, 2026·

  ·
  1

• Bump llama.cpp to 4fbc76dec51d0add466f0210855c0596589b60d4

  #2382·jamesbiederbeck opened Oct 12, 2026Contributor·

  ·
  1

• feat(llama): add CPU MoE offloading support

  #2381·Madrawn opened Oct 10, 2026·

  ·