All pull requests

Pull requests

• feat(nvidia): add Mamba2 and RWKV5 adapters

  #591·lgeln10 opened Sep 20, 2026·

  ·

• perf: reduce KV admission over-reservation

  #590·lgeln10 opened Sep 20, 2026·

  ·

• feat(nvidia): 为 v0.2.9c 增加可复用的 GGUF Route B 支持

  #589·xindongliu594 opened Sep 20, 2026·

  ·

• feat: add mtp draft models for speculative decoding

  #588·accelerator-llc opened Sep 20, 2026·

  ·

• feat(model): support LFM2-1.2B

  #587·Hyacinth25 opened Sep 20, 2026·

  ·
  1

• feat: add FP8 blockwise weight quantization and FP8(E4M3) KV cache support

  #586·shsaihdsaiudh opened Sep 20, 2026·

  ·

• perf: Qwen3 offline inference — hybrid attention backend, fused rms_norm_rope, chunked prefill + prompt-lookup spec decoding

  #585·shsaihdsaiudh opened Sep 20, 2026·

  ·

• feat(qwen): add paged greedy MTP with FP8 block weights

  #584·big-hip opened Sep 19, 2026·

  ·

• feat(minimax): support MiniMax-Text-01 and MiniMax-M2

  #583·y258dd opened Sep 19, 2026·

  ·

• feat(quantization): support compressed-tensors W8A8 and mixed modules

  #582·sam0336 opened Sep 19, 2026·

  ·
  1

• Feat/add gemma3 model

  #581·zyc107109102 opened Sep 18, 2026·

  ·

• feat(nvidia): support model type: granitemoehybrid

  #580·WHoutstanding opened Sep 18, 2026·

  ·

• perf: reuse running reservations within each scheduler step

  #579·T4t4KAU opened Sep 17, 2026·

  ·

• perf: bound KV page reclamation candidate scans

  #578·T4t4KAU opened Sep 17, 2026·

  ·

• perf: build KV slot mappings by page ranges

  #577·T4t4KAU opened Sep 17, 2026·

  ·

• perf: avoid scanning KV cache pages for capacity checks

  #576·T4t4KAU opened Sep 17, 2026·

  ·

• feat(mamba2): integrate model loading and recurrent inference

  #575·big-hip opened Sep 17, 2026·

  ·

• Feat/OpenAI agent compat

  #574·wooway777 opened Sep 17, 2026Collaborator·

  ·

• feat(engine): integrate prefix eviction and chunked parallel execution

  #573·big-hip opened Sep 16, 2026·

  ·
  1

• feat(nvidia): support MiniMax-Text-01

  #572·Kritace opened Sep 15, 2026·

  ·

• feat: add priority-aware request scheduling

  #571·xiaoba17 opened Sep 13, 2026·

  ·
  1

• fix(moore): enable Mate flash-attn inference

  #567·spike-zhu opened Sep 10, 2026Collaborator·

  ·

• feat(nvidia): add reusable GGUF Route B support for Qwen3.5

  #559·xindongliu594 opened Sep 3, 2026·

  ·
  1

• perf: fuse unquantized Qwen3-Next GDN projections

  #555·T4t4KAU opened Sep 2, 2026·

  ·

• feat(server): add agent support with tool-call and reasoning parsing

  #554·rubik-hua opened Sep 1, 2026Contributor·

  ·
  7