All pull requests

Pull requests

• [App:Feature] Preserve Android chat history after model deletion

  #5076·hulkbig opened Oct 30, 2026Contributor·3 of 5 tasks·

  ·

• [App:Feature] Import prepared MNN model folders through Android SAF

  #5074·hulkbig opened Oct 30, 2026Contributor·5 of 12 tasks·

  ·

• [App:Feature] Add Supertonic voice selection to MNN Chat

  #5072·hulkbig opened Oct 29, 2026Contributor·3 of 5 tasks·

  ·

• Feature/hexagon app support

  #5071·Qxinyu opened Oct 29, 2026Collaborator·12 tasks·

  ·

• [Infra:Bugfix] Cover build inputs in platform CI filters

  #5070·luke-opus opened Oct 29, 2026Collaborator·4 tasks done·

  ·

• [Infra:Bugfix] Rebuild local test binaries before regression

  #5069·luke-opus opened Oct 29, 2026Collaborator·5 of 12 tasks·

  ·

• [LLM:Perf] Skip intermediate prefill logits with a reusable CPU trunk

  #5066·kerry0520 opened Oct 29, 2026Contributor·5 of 12 tasks·

  ·
  2

• [LLM:Feature] Add DSpark speculative decoding

  #5065·yydhYYDH opened Oct 29, 2026Collaborator·8 of 12 tasks·

  ·

• [LLM:Bugfix] Qwen3-TTS: align the talker with the Python implementation (codec prefix + sampling)

  #5064·Pokonliu opened Oct 29, 2026·

  ·
  2

• [Infra:Bugfix] Propagate resource model test failures

  #5058·luke-opus opened Oct 28, 2026Collaborator·5 of 12 tasks·

  ·

• [Infra:Bugfix] Handle model comparison startup failures

  #5057·luke-opus opened Oct 28, 2026Collaborator·5 of 12 tasks·

  ·

• [Converter:Bugfix] Fix boolean option parsing

  #5054·HowieYin opened Oct 27, 2026·5 of 12 tasks·

  ·
  1

• [OpenCL:Bugfix][App:Bugfix] Pending local fixes: OpenCL attention guard and attached-image loading

  #5053·ValiantDiligent opened Oct 27, 2026Collaborator·

  ·

• [App:Feature] Add context budget, auto-summarization and context usage to Android chat

  #5052·ValiantDiligent opened Oct 27, 2026Collaborator·

  ·

• [CPU:Bugfix] Accumulate ARM82 average pooling in FP32

  #5049·ZedingZhang opened Oct 26, 2026Contributor·5 of 12 tasks·

  ·
  1

• [Infra:Bugfix] Detect truncated and malformed tensor dumps

  #5048·luke-opus opened Oct 26, 2026Collaborator·5 of 12 tasks·

  ·

• [Infra:Bugfix] Reject missing model test inputs and references

  #5047·luke-opus opened Oct 26, 2026Collaborator·5 of 12 tasks·

  ·

• [CPU:Bugfix] Fix per-channel quantization indexing across batches

  #5046·luke-opus opened Oct 26, 2026Collaborator·5 of 12 tasks·

  ·

• [LLM:Feature] Support DFlash speculative decoding on Hexagon

  #5044·yydhYYDH opened Oct 26, 2026Collaborator·7 of 12 tasks·

  ·

• Mark x86_64 assembly kernels as not needing an executable stack (fix import on glibc 2.41+)

  #5041·glenn-jocher opened Oct 24, 2026·

  ·
  1

• [LLM:Bugfix] Handle CRLF records in legacy Huggingface tokenizers

  #5039·kerry0520 opened Oct 23, 2026Contributor·5 of 12 tasks·

  ·

• [LLM:Bugfix] Propagate text tokenizer vocabulary load failures

  #5034·kerry0520 opened Oct 23, 2026Contributor·5 of 12 tasks·

  ·

• [PyMNN:Bugfix] Honor shape contracts in array construction and composition

  module:pymnntype:bug
  #5030·hulkbig opened Oct 22, 2026Contributor·

  ·
  

• [PyMNN:Bugfix] Preserve dimensions in statistical reductions

  module:pymnntype:bug
  #5029·hulkbig opened Oct 22, 2026Contributor·

  ·
  

• [CPU:Bugfix] Preserve C4 quantization alignment across RVV chunks

  module:quantizationtype:bug
  #5024·luke-opus opened Oct 21, 2026Collaborator·4 of 12 tasks·

  ·