All pull requests

Pull requests

• Add instance-local rule buckets to the Cua-S1 multimodal worker

  #37·Levius-Fubuki opened Sep 29, 2026·

  ·
  1

• feat(frontend): own the engine in a worker thread with a bounded queue [stacked on #34]

  #35·xiaoyu-xyz opened Sep 29, 2026·

  ·

• feat(frontend): add the in-process engine contract and HTTP service

  #34·xiaoyu-xyz opened Sep 29, 2026·

  ·

• Bound Cua-S1 Graph capture work and share per-shape CUDA pools

  #33·Levius-Fubuki opened Sep 29, 2026·

  ·
  1

• Add LFM2.5-350M choice worker with candidate batching

  #32·prettygirlisnotme opened Sep 28, 2026·

  ·

• Laya on Apple Silicon: worker, benchmarks and recipe

  #30·cacheline999 opened Sep 28, 2026·

  ·
  1

• [CLM 2/2] Reach the encoder, answer requests, and check against CLM

  #29·xiaoyu-xyz opened Sep 28, 2026·

  ·

• [CLM 1/2] Checkpoint and typed decision scoring

  #28·xiaoyu-xyz opened Sep 28, 2026·

  ·

• cuda: add the backend contract and its checker

  #25·xiaoyu-xyz opened Sep 28, 2026·

  ·

• Measure mixed-length CUDA Graph capture and eviction costs

  #24·Levius-Fubuki opened Sep 28, 2026·

  ·
  1

• [Recipe] Run CLM behind the frontend, and compare engine responses

  #23·xiaoyu-xyz opened Sep 28, 2026·

  ·

• Serve Cua-S1 multimodal requests with bounded segmented CUDA Graphs

  #22·Levius-Fubuki opened Sep 28, 2026·

  ·
  1

• Add CPU checkpoint loading for Laya

  #21·linear3735 opened Sep 28, 2026·

  ·

• Measure CUDA Graph replay on Cua-S1 multimodal language forward

  #20·Levius-Fubuki opened Sep 28, 2026·

  ·
  1

• cua_s1: add a native CUDA text worker

  #19·twu3202 opened Sep 28, 2026·

  ·
  2

• Limit Cua-S1 reused-request output projection to final token

  #18·Levius-Fubuki opened Sep 28, 2026·

  ·
  1

• Reuse Cua-S1 image preprocessing and vision features per request

  #17·Levius-Fubuki opened Sep 28, 2026·

  ·
  1

• Add Cua-S1 multimodal stage profiling and RTX 4090 results

  #15·Levius-Fubuki opened Sep 28, 2026·

  ·

• cua_s1: add a text worker that loads Qwen3.5-4B through Transformers

  #13·twu3202 opened Sep 27, 2026·

  ·
  1

• Add Cua-S1 0.2 multimodal CUDA worker with upstream parity

  #12·Levius-Fubuki opened Sep 27, 2026·

  ·
  4

• docs: add Cua-S1 4B 0.2 inference contract

  #11·twu3202 opened Sep 27, 2026·

  ·
  2