Commit History

Autor SHA1 Mensaxe Data
  Alex Cheema d4a932e405 fix merge hai 1 ano
  Alex Cheema 2341aa1acf Merge branch 'main' into better_networking hai 1 ano
  Alex Cheema 5a9f4ba5c1 update examples: remove old llama3_distributed, add chatgpt_api hai 1 ano
  Alex Cheema 581856897a clean up unused, formatting hai 1 ano
  Alex Cheema 62e3726263 add RTX 20 series to device capabilities hai 1 ano
  Alex Cheema ebff636a25 script ot start openwebui hai 1 ano
  Alex Cheema 394935711b add all chat endpoints without v1 prefix to support ollama / openwebui. related: #175 hai 1 ano
  Alex Cheema 70172d7cb9 add /v1/models endpoint and change Content-Type of stremed response to text/event-stream. fixes #175 hai 1 ano
  Alex Cheema d917778e2b update mlx to 0.17.1 (not sure where 0.17.0 went on PyPi disappeared)g hai 1 ano
  Alex Cheema 2667c8af44 cleaner download_progress hai 1 ano
  Alex Cheema f46d077beb fix font dependencies for tinychat. related: #172 hai 1 ano
  Alex Cheema 8a4928f80c fix gitignore to not ignore tinychat static files hai 1 ano
  Alex Cheema d515d9efa3 explicitly use absolute paths for tinychat deps hai 1 ano
  Alex Cheema a386c35fde script to update tinychat deps hai 1 ano
  Alex Cheema 3791e669a4 download tinychat dependencies all to local dir so we dont need internet hai 1 ano
  Alex Cheema 85bab25ac0 fix local check if dir does not exist hai 1 ano
  Alex Cheema 59c4393d95 first try loading tokenizer from local path instead of always going to the internet first. significant speed ups hai 1 ano
  Alex Cheema 784e6bae21 print traceback on topology collection error hai 1 ano
  Alex Cheema 8cad0e1849 only use_fast tokenizer for Mistral Large until this inconsistency bug is fixed #171 hai 1 ano
  Alex Cheema 85279007b3 hotfix edge case where we try to render before tokenizer is set hai 1 ano
  Alex Cheema 09a8468395 upgrade mlx to 0.17.0 hai 1 ano
  Alex Cheema 1f9d16ec78 run tokenizers test in ci, run all models available hai 1 ano
  Alex Cheema 6243846eeb ci logs hai 1 ano
  Alex Cheema cfe980bdaa simplify ci hai 1 ano
  Alex Cheema 9513c4fd17 ci tail log files hai 1 ano
  Alex Cheema 7a02acdcd5 fix ci output streaming hai 1 ano
  Alex Cheema ad695696a5 run on every commit on main, reuqire approval on other branches hai 1 ano
  Alex Cheema 710e5a31e7 TODO for why use_fast=False is giving inconsistent behaviour (no spaces decoding invididual tokens) for Mistral-Large-Instruct-2407-4bit hai 1 ano
  Alex Cheema e17e5f9a41 tests for tokenizers. unfortunately use_fast=False and use_fast=True give different behaviour hai 1 ano
  Alex Cheema 0d218e244e use fast AutoProcessor fixes #164 tokenizer issues with mistral-large. hai 1 ano