Skip to content

Latest commit

 

History

History
102 lines (74 loc) · 1.75 KB

README.md

File metadata and controls

102 lines (74 loc) · 1.75 KB
Raw

Next art Any_RAG

any_cot

Chain of Thought implementation using different API vendors

inspiration: https://github.com/Jaimboh/Llamaberry-Chain-of-Thought-Reasoning-in-AI

Groq test

cot_models:
  first_turn:
      name: "llama-3.1-70b-versatile"
      api: "groq"
      user_prompt: |        
        {user_prompt}.      
  followup:
      name: "llama-3.1-70b-versatile"
      api: "groq"
  synthesis:  
      name: "llama-3.1-70b-versatile"
      api: "groq"


cot_prompt:
  first_turn: |
  ...

Execute:

python bot.py config\groq\groq_test.yaml

Hugging face test

(generates image prompt)

cot_models:
first_turn:
    name: "mistralai/Mistral-Nemo-Instruct-2407"
    api: "hugging_face"
    user_prompt: |        
      {user_prompt}
      return a detailed description of updated  image prompt in <fused_image> tags.      
followup:
    name: "mistralai/Mistral-Nemo-Instruct-2407"
    api: "hugging_face"
synthesis:  
    name: "mistralai/Mistral-Nemo-Instruct-2407"
    api: "hugging_face"


cot_prompt:
first_turn: |
...

Execute:

python bot.py config\hugging_face\test.yaml

Mixed test

Groq + HF (Groq as synthesizer)

cot_models:
  first_turn:
      name: "mistralai/Mistral-Nemo-Instruct-2407"
      api: "hugging_face"
      user_prompt: |        
        {user_prompt}
             
  followup:
      name: "mistralai/Mistral-Nemo-Instruct-2407"
      api: "hugging_face"
  synthesis:  
      name: "llama-3.1-70b-versatile"
      api: "groq"


cot_prompt:
  first_turn: |
  ...

Execute:

python bot.py config\mixed\mixed_test.yaml

Previous art: Any_MOA