Commits · 54b6d4c0b59094d969884004fe57b02d395134b3 · Grießhaber Daniel / evoprompt · GitLab

Snippets Groups Projects

Sep 30, 2024
- Refactor tasks and models and fix format for various models · 54b6d4c0
  Max Kimmich authored 5 months ago
  
  54b6d4c0
Sep 26, 2024
- Fix HfLlamaChat not using correct arguments · 3b688348
  Max Kimmich authored 6 months ago
  
  3b688348
Sep 25, 2024
- Always use temperature=0.0 for task prediction · d630c67c
  Max Kimmich authored 6 months ago
  
  d630c67c
- Set temperature to lowest value for evaluation script · 1f10b51f
  Max Kimmich authored 6 months ago
  
  1f10b51f
- Fix missing keys for hash computation for HfLlamaChat · ba90bb7e
  Max Kimmich authored 6 months ago
  
  ba90bb7e
- Use Alpaca model from evoprompt paper's reference implementation · 22316c70
  Max Kimmich authored 6 months ago
  
  22316c70
- Allow to ignore keys in computing the model hash · 1790becb
  Max Kimmich authored 6 months ago
  
  1790becb
- add alpaca prompt evaluation · 5d6a9062
  Grießhaber Daniel authored 6 months ago
  
  5d6a9062
Sep 24, 2024
- Add script for evaluating prompt · 46433787
  Max Kimmich authored 6 months ago
  
  46433787
Sep 23, 2024
- Increase variability of generated name · 6de7bbb0
  Max Kimmich authored 6 months ago
  
  6de7bbb0
Sep 20, 2024
- Allow to skip bad evolutions based on judge model · 8355c2cb
  Max Kimmich authored 6 months ago
  
  8355c2cb
Sep 19, 2024
- Separately returning history and recent turn in model call · 926ed2b5
  Max Kimmich authored 6 months ago
  
  926ed2b5
- Fix DE-CoT prompt template · ab368ffe
  Max Kimmich authored 6 months ago
  
  ab368ffe
Sep 06, 2024
- Increase context size · 1653af4f
  Max Kimmich authored 6 months ago
  
  1653af4f
- Allow user to skip evolution if response was judged as bad · f021dfcb
  Max Kimmich authored 6 months ago
  
  f021dfcb
- Fix SQuAD issue with demonstration prompts for evolution · 6a83f0bd
  Max Kimmich authored 6 months ago
  
  6a83f0bd
Sep 05, 2024
- Log options to wandb · 708cd528
  Max Kimmich authored 6 months ago
  
  708cd528
- Use Llama3.1 as default for llamachat · e6832e69
  Max Kimmich authored 6 months ago
  
  e6832e69
- add llama3.1 support using transformers · f2fd6467
  Grießhaber Daniel authored 8 months ago and Max Kimmich committed 6 months ago
  
  f2fd6467
Sep 03, 2024
- Add missing python dependency · 999a6522
  Max Kimmich authored 6 months ago
  
  999a6522
- Add missing python dependency · 0b8235f1
  Max Kimmich authored 6 months ago
  
  0b8235f1
- Add wandb integration · 97d39f2f
  Max Kimmich authored 6 months ago
  
  97d39f2f
- Implement generating base prompts · fc4f0be6
  Max Kimmich authored 6 months ago
  
  fc4f0be6
- Only load models once · dd98676d
  Max Kimmich authored 6 months ago
  
  dd98676d
- Fix topic classification and question answering tasks · 31c10c46
  Max Kimmich authored 6 months ago
  
  31c10c46
Aug 29, 2024
- Add demonstration data for GA and DE-CoT · 8943980c
  Max Kimmich authored 7 months ago
  
  8943980c
Aug 28, 2024
- Allow to cache demonstration samples via datasets library · 5ee5bb02
  Max Kimmich authored 7 months ago
  
  5ee5bb02
- Log invalid answer as debug to not pollute terminal · 5e5d7541
  Max Kimmich authored 7 months ago
  
  5e5d7541
Aug 22, 2024
- Consider debug mode for paraphrasing · bbce7ee4
  Max Kimmich authored 7 months ago
  
  bbce7ee4
- Simpllify BasePromptsFromJsonMixin and add auto prompts · 662a11e8
  Max Kimmich authored 7 months ago
  
  662a11e8
- Add judgement using judge model and update editor rendering · da2cc6ed
  Max Kimmich authored 7 months ago
  
  da2cc6ed
- Merge branch 'llm-as-a-judge' into 'master' · 0099d51c
  Grießhaber Daniel authored 7 months ago
  
  Llm as a judge See merge request !5
  0099d51c
- initialize judge_model · d008d33a
  Grießhaber Daniel authored 7 months ago
  
  d008d33a
- fix problem when setting 0 demonstration exaples · 9f5d05bb
  Grießhaber Daniel authored 7 months ago
  
  9f5d05bb
- allow to diable llmaaj by not specifying judgement engine · 9462d643
  Grießhaber Daniel authored 7 months ago
  
  9462d643
- add interactive editor to correct badly judged reponses · e751763b
  Grießhaber Daniel authored 7 months ago
  
  e751763b
- add textual editor · 6c484fc0
  Grießhaber Daniel authored 7 months ago
  
  6c484fc0
- add hooks to judge llm evolution response · df42a740
  Grießhaber Daniel authored 7 months ago
  
  df42a740
- Merge branch 'refactor-task' into 'master' · 7e712a59
  Grießhaber Daniel authored 7 months ago
  
  Refactor tasks so that tasks have more control over model input and output See merge request !6
  7e712a59
Aug 21, 2024
- Refactor tasks so that tasks have more control over model input and output · 78ac102a
  Max Kimmich authored 7 months ago
  
  78ac102a