All pull requests

Pull requests

• fix: bridge a completion that stopped inside open reasoning

  #189·xinyuangui2 opened Oct 11, 2026·

  ·
  2

• Add black-box renderer conformance and pinned vLLM audits

  #188·hallerite opened Oct 11, 2026Member·

  ·

• fix: split a Qwen3.5 <tool_call> block into one call per <function> (vLLM parity)

  #185·yllkryeziu opened Oct 9, 2026·

  ·

• fix(default): keep message_indices aligned when a chat template shrinks

  #183·awesome-pro opened Oct 9, 2026·

  ·

• fix(base): don't synthesise a turn close for an empty completion

  #182·awesome-pro opened Oct 9, 2026·

  ·

• feat(client): transport sampler probability heads for score centering

  #180·nxphi47 opened Oct 8, 2026·

  ·

• perf(client): read prime-rl's packed completion_logprobs

  #174·faresobeid opened Oct 6, 2026Contributor·

  ·
  4

• feat: add Mistral-3 renderer

  #172·DIYA73 opened Oct 4, 2026·

  ·
  2

• feat(client): pass through sampling-mask logprobs and by-handle payload segments

  #170·faresobeid opened Oct 3, 2026Contributor·

  ·
  3

• Qwen3.5-style XML tool calls: coerce True/False booleans, keep string whitespace, split restarted calls

  #166·RGaonkar26 opened Sep 30, 2026·

  ·
  1

• feat(client): allow token capture when logprobs are omitted

  #163·nxphi47 opened Sep 28, 2026·

  ·
  1

• feat(client): preserve paired routing weights for Total Router Recall

  #154·faresobeid opened Sep 9, 2026Contributor·

  ·

• Keep renderer token custody in fixed-width arrays

  #149·seanbell opened Sep 1, 2026·

  ·

• feat: enforce canonical reasoning messages

  #148·hallerite opened Sep 1, 2026Member·

  ·
  12

• Centralize renderer registration metadata

  #146·hallerite opened Aug 31, 2026Member·

  ·
  1

• feat: normalize provider tool specifications

  #119·hallerite opened Aug 6, 2026Member·

  ·

• test: freeze renderer behavior before refactoring

  #118·hallerite opened Aug 6, 2026Member·

  ·
  3

• feat(client)!: tool_calls mirrors the engine; attempts move to tool_call_attempts

  #117·mikasenghaas opened Aug 6, 2026Member·

  ·
  3

• fix: encode message content with split_special_tokens=True

  #116·hubert-marek opened Aug 5, 2026Contributor·

  ·
  9

• Inline raw image refs for multimodal rendering

  #110·eligotts opened Jul 23, 2026Contributor·

  ·
  4

• feat: rlm renderer for the RLM chat format

  #107·mikasenghaas opened Jul 18, 2026Member·

  ·

• [codex] Support raw image refs for multimodal rendering

  #89·eligotts opened Jun 18, 2026Contributor·

  ·
  8

• perf(mm): ephemeral pixel return + smaller image cache to bound rollout RAM

  #85·mikasenghaas opened Jun 11, 2026Member·

  ·

• feat(client): add dynamo_chat transport + routed_experts to renderer generate

  #79·biswapanda opened Jun 9, 2026Contributor·1 task done·

  ·
  14

• Rust PyO3 renderer port and native performance paths

  #73·ThomAub opened May 28, 2026·

  ·