Commit History

Автор SHA1 Съобщение Дата
  Michael Yang b732beba6a lint преди 9 месеца
  Jeffrey Morgan c4cf8ad559 llm: avoid loading model if system memory is too small (#5637) преди 9 месеца
  Jeffrey Morgan f8241bfba3 gpu: report system free memory instead of 0 (#5521) преди 10 месеца
  Daniel Hiltgen 6f351bf586 review comments and coverage преди 11 месеца
  Daniel Hiltgen fc37c192ae Refine CPU load behavior with system memory visibility преди 11 месеца
  Daniel Hiltgen 30a7d7096c Bump VRAM buffer back up преди 11 месеца
  Michael Yang 4736391bfb llm: add minimum based on layer size преди 1 година
  Jeffrey Morgan f0c454ab57 gpu: add 512MiB to darwin minimum, metal doesn't have partial offloading overhead (#4068) преди 1 година
  Daniel Hiltgen 34b9db5afc Request and model concurrency преди 1 година
  Michael Yang 26df674785 scale graph based on gpu count преди 1 година
  Michael Yang 41a272de9f darwin: no partial offloading if required memory greater than system преди 1 година
  Michael Yang 7e33a017c0 partial offloading преди 1 година
  Daniel Hiltgen be330174dd Allow setting max vram for workarounds преди 1 година
  peanut256 a189810df6 Determine max VRAM on macOS using `recommendedMaxWorkingSetSize` (#2354) преди 1 година
  Daniel Hiltgen 7427fa1387 Fix up the CPU fallback selection преди 1 година
  Daniel Hiltgen 39928a42e8 Always dynamically load the llm server library преди 1 година
  Daniel Hiltgen d88c527be3 Build multiple CPU variants and pick the best преди 1 година
  Jeffrey Morgan c336693f07 calculate overhead based number of gpu devices (#1875) преди 1 година
  Jeffrey Morgan 08f1e18965 Offload layers to GPU based on new model size estimates (#1850) преди 1 година
  Jeffrey Morgan c7ea8f237e set `num_gpu` to 1 only by default on darwin arm64 (#1771) преди 1 година
  Daniel Hiltgen a2ad952440 Fix windows system memory lookup преди 1 година
  Daniel Hiltgen d966b730ac Switch windows build to fully dynamic преди 1 година
  Daniel Hiltgen 7555ea44f8 Revamp the dynamic library shim преди 1 година
  Daniel Hiltgen 6558f94ed0 Fix darwin intel build преди 1 година
  Daniel Hiltgen 35934b2e05 Adapted rocm support to cgo based llama.cpp преди 1 година