All pull requests

Pull requests

• server: report what the server holds beside the serving context

  mtmdserver
  #91·joelteply opened Nov 3, 2026·

  ·
  1

• server: learn a past example: /learn's messages form serves its prompt through the ordinary completion path first, then steps

  server
  #89·joelteply opened Nov 2, 2026·

  ·

• server: the lost-reply binding: a learn names its execution, base step and engine; a repeat is answered from its receipt, a stale one is fenced

  ggmlservertesting
  #88·joelteply opened Nov 1, 2026·

  ·
  2

• server: one learning ledger (draft): every learn step debits the grant that admitted it

  servertesting
  #87·joelteply opened Nov 1, 2026·

  ·
  4

• server: POST /lora-adapters/snapshot writes a mind's shadow to an immutable GGUF between steps, and /learn reports the shadow's committed step

  server
  #86·joelteply opened Nov 1, 2026·

  ·

• server: one shadow per mind on the serving context: each shadow keeps its own optimizer, /learn trains any loaded shadow, and unload releases it

  server
  #85·joelteply opened Nov 1, 2026·

  ·

• server: create dormant dream adapters and save immutable snapshots

  devopsserver
  #81·joelteply opened Oct 28, 2026·

  ·
  2

• server: a page carries the draft's KV once; checkpoints stop copying it

  server
  #79·joelteply opened Oct 28, 2026·

  ·
  1

• server: a disk tier for the prompt cache (--cache-disk)

  server
  #78·joelteply opened Oct 27, 2026·

  ·

• server: /reconfigure changes the context geometry without reloading the model (S2)

  server
  #73·joelteply opened Oct 24, 2026·

  ·

• server: genes (LoRA adapters) page between the device and host RAM (stacked on #70)

  server
  #72·joelteply opened Oct 24, 2026·

  ·

• server: a slot pages to named host RAM and back; checkpoints keep a byte budget; a page carries its draft KV

  devopsservertesting
  #70·joelteply opened Oct 24, 2026·

  ·
  1

• server: rotate admitted prefill service without moving KV

  devopsservertesting
  #69·joelteply opened Oct 23, 2026·

  ·
  2

• server: expose realized context allocation geometry

  Apple MetalCUDAggmlmodelmtmdserverSYCLtesting
  #67·joelteply opened Oct 20, 2026·

  ·

• grammar: preserve tagged unions through singleton allOf

  testing
  #66·joelteply opened Oct 17, 2026·

  ·

• llama: connect labelled decode to shared physical cache rows

  ggmltesting
  #64·joelteply opened Oct 14, 2026·

  ·
  4

• cuda: implement shared FLASH_ATTN_BACK with direct quantized cache reads

  CUDAggmltesting
  #63·joelteply opened Oct 14, 2026·

  ·
  5

• train: the start's claim is released on every exit that doesn't hand it to the worker

  servertesting
  #58·joelteply opened Oct 13, 2026·

  ·

• server /train: a per-turn slowdown receipt beside the cycle average

  servertesting
  #39·joelteply opened Oct 6, 2026·

  ·
  7

• continuum: explicit Metal device selection — avoid hang in AMD Polaris lazy init (card 346b356f)

  #1·joelteply opened Jun 7, 2026·

  ·