Commit Graph

  • ded2b6cecc Shuffle concatenated dataset; main extio-de 2025-02-16 10:51:39 +01:00
  • cb52446985 Batched loading and chunking of text datasets; extio-de 2025-02-09 16:17:27 +01:00
  • f253fde344 Don't reload model for the merge step; extio-de 2025-02-08 11:25:12 +01:00
  • a0e9a5a33a Batched, parallel dataset loading; Disable dataset caching; extio-de 2025-01-25 18:12:31 +01:00
  • 2799b9c555 Process conversation datasets; extio-de 2025-01-11 21:58:29 +01:00
  • 63192f2bc0 Dataset is now directly tokenized to avoid further post-processing of the trainer; locCustomPromptTemplate to specify a custom Jinja template; showChatTemplate for debugging; extio-de 2024-12-26 21:42:59 +01:00
  • cb9e62dbac Dataset: Support history/instruct/completion style; extio-de 2024-12-19 14:05:38 +01:00
  • b10546abc2 lm-finetuner: Close assistant responses; extio-de 2024-12-02 11:33:08 +01:00
  • ff2385bf50 lm-finetuner: Merger: Dequantize before merging lora adapter; More lora options; extio-de 2024-12-01 13:03:45 +01:00
  • aa958eeb81 Add eot tokens to texts; extio-de 2024-10-27 20:47:03 +01:00
  • 8e2f75e6f4 Added more training parameters; Removed chat emulation of text datasets; Fixes; extio-de 2024-10-27 09:59:16 +01:00
  • 91b211547b Readme extio-de 2024-10-20 21:53:33 +02:00
  • 838664935b initial commit extio-de 2024-10-20 20:59:27 +02:00