Llama3.2-ColdBrew-4x3B-Argon-test0

Llama3.2-ColdBrew-4x3B-Argon-test0 is a Mixture of Experts (MoE) made with the following models using LazyMergekit:

🧩 Configuration

name: Llama3.2-ColdBrew-4x3B-Argon-test0
base_model: SvalTek/Llama3.2-ColdBrew-3B-OverThinker
gate_mode: hidden
experts:
  # -- TOOLS & PROGRAMMING EXPERT --
  - source_model: SvalTek/Llama3.2-ColdBrew-3B-Bronze-test0  # Tools/Programming
    positive_prompts:
      - "Write a function"
      - "Implement in"
      - "How do I code"
      - "How to write a"
      - "Python code"
      - "JavaScript code"
      - "Java code"
      - "Show me code"
      - "Show me how to"
      - "Here's an error"
      - "Fix this error"
      - "What does this error mean"
      - "How do I use"
      - "What is wrong with my code"
      - "Return only the code"
      - "Debug this"
      - "Can you debug"
      - "Tool call:"
      - "Script to automate"
      - "How can I automate"
      - "Command line example"
      - "Bash script"
      - "Shell script"
      - "Regex for"
      - "Write a regex"
      - "SQL query"
      - "Database schema"
      - "API example"
      - "API usage"
      - "Unit test for"
      - "Test case for"
      - "Write pseudocode"
      - "Config file"
      - "JSON format"
      - "YAML config"
      - "<tools>"
      - "<tool_response>"
    negative_prompts:
      - "Write a story"
      - "Tell me a story"
      - "Scene:"
      - "As a character"
      - "Dialogue between"
      - "Summarize this in plain English"
      - "Roleplay"
      - "Describe a character"
      - "Backstory for"
      - "Oracle"
      - "Poem"
      - "Song lyrics"
      - "<think>"
      - "Explain your reasoning"
      - "Chain-of-thought"
      - "Step by step reasoning"
      - "What would you do if"
      - "In a fantasy world"
      - "Continue the story"
      - "First-person perspective"

  # -- WRITER / ROLEPLAYER EXPERT --
  - source_model: SvalTek/Llama3.2-ColdBrew-3B-Sage-test0   # Writer/Roleplayer
    positive_prompts:
      - "Write a story about"
      - "Tell a story about"
      - "Scene:"
      - "Describe the setting"
      - "Roleplay as"
      - "Pretend you are"
      - "In character,"
      - "Write a dialogue between"
      - "Dialogue between"
      - "Write a poem"
      - "Poem about"
      - "Write a monologue"
      - "First-person perspective"
      - "Narrate"
      - "Backstory for"
      - "Worldbuilding"
      - "Describe a character"
      - "Describe a magical place"
      - "Continue this story"
      - "Flash fiction"
      - "Epic poem"
      - "Science fiction story"
      - "Fantasy story"
      - "Horror story"
      - "Historical fiction"
      - "Letter from"
      - "Write a prophecy"
      - "Villain monologue"
      - "Supernatural"
      - "Short story"
      - "Invent a character"
      - "Make up a legend"
    negative_prompts:
      - "Show code"
      - "Write code"
      - "Debug"
      - "Python function"
      - "Implement"
      - "How to write a function"
      - "Explain step by step"
      - "Fact check"
      - "Math question"
      - "Solve for x"
      - "Tool call"
      - "Explain your reasoning"
      - "Chain-of-thought"
      - "<think>"
      - "<tools>"
      - "<tool_response>"
      - "Summarize this in plain English"
      - "Step by step reasoning"
      - "What would you do if"
      - "Continue the story"
      - "First-person perspective"
      - "Summarize the history"
      - "What is the difference between"
      - "List the"
      - "Give me a definition"
      - "How do I use"
      - "What is wrong with my code"

  # -- ORACLE: GENERAL KNOWLEDGE / Q&A --
  - source_model: SvalTek/Llama3.2-ColdBrew-3B-Oracle-test0  # General Q&A / Factual
    positive_prompts:
      - "What is"
      - "Who is"
      - "When did"
      - "Where is"
      - "List the"
      - "Explain the difference between"
      - "Summarize"
      - "Fact check"
      - "History of"
      - "How does"
      - "Why do"
      - "Why does"
      - "What causes"
      - "What are the main causes"
      - "Tell me about"
      - "Trivia question"
      - "Riddle:"
      - "Give a concise explanation of"
      - "Advice on"
      - "What would happen if"
      - "Definition of"
      - "Give me a list"
      - "Biggest"
      - "Smallest"
      - "Most important"
      - "Give examples"
      - "Symptoms of"
      - "Signs of"
      - "Tips for"
      - "Explain simply"
      - "In simple terms"
      - "For beginners"
      - "Layman's terms"
    negative_prompts:
      - "Roleplay"
      - "Scene:"
      - "Write a story"
      - "Tell me a story"
      - "Continue the story"
      - "Write code"
      - "Show me code"
      - "Tool call"
      - "<think>"
      - "<tools>"
      - "<tool_response>"
      - "Explain your reasoning"
      - "Chain-of-thought"
      - "Step by step explanation"
      - "Write a function"
      - "Narrate"
      - "Dialogue between"
      - "Write a poem"
      - "Make up a legend"
      - "Describe a magical place"
      - "Pretend you are"

  # -- OVERTHINKER: REASONING, INTROSPECTION, EXPLANATION --
  - source_model: SvalTek/Llama3.2-ColdBrew-3B-OverThinker   # Reasoning/Introspection/Base
    positive_prompts:
      - "<think>"
      - "Explain your reasoning"
      - "Step by step"
      - "Walk me through"
      - "Chain-of-thought"
      - "Let's think step by step"
      - "Reflect on"
      - "Consider the following"
      - "Why do you think"
      - "List pros and cons"
      - "Debate"
      - "Analyze"
      - "Reason through"
      - "What if"
      - "Evaluate"
      - "Uncertainty"
      - "Explore possible causes"
      - "Examine alternatives"
      - "Break down the logic"
      - "What are the possible explanations"
      - "Discuss the logic"
      - "Compare and contrast"
      - "Challenge this idea"
      - "Assess the argument"
      - "Explore doubts"
      - "Lay out your logic"
    negative_prompts:
      - "Roleplay as"
      - "Pretend you are"
      - "Scene:"
      - "Write a story"
      - "Tell me a story"
      - "Continue this story"
      - "Dialogue"
      - "Write code"
      - "Show code"
      - "Write a function"
      - "Fact check"
      - "Give a concise answer"
      - "Summarize this in plain English"
      - "Give me a list"
      - "Short answer only"
      - "Write a poem"
      - "Describe a magical place"
      - "Tool call"
      - "<think>"
      - "<tools>"
      - "<tool_response>"
      - "Summarize the history"
      - "What is the difference between"

dtype: bfloat16
normalize: true
int8_mask: true

💻 Usage

!pip install -qU transformers bitsandbytes accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "SvalTek/Llama3.2-ColdBrew-4x3B-Argon-test0"

tokenizer = AutoTokenizer.from_pretrained(model)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    model_kwargs={"torch_dtype": torch.float16, "load_in_4bit": True},
)

messages = [{"role": "user", "content": "Explain what a Mixture of Experts is in less than 100 words."}]
prompt = pipeline.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])

SvalTek
/

Llama3.2-ColdBrew-4x3B-Argon-test0

Llama3.2-ColdBrew-4x3B-Argon-test0

🧩 Configuration

💻 Usage

Model tree for SvalTek/Llama3.2-ColdBrew-4x3B-Argon-test0