MixtureofMerges-MoE-4x7b-v10-MIXTRAL3

MixtureofMerges-MoE-4x7b-v10-MIXTRAL3 is a Mixture of Experts (MoE) made with the following models using LazyMergekit:

• mistralai/Mistral-7B-Instruct-v0.3
• Kukedlc/NeuralSynthesis-7B-v0.1
• mlabonne/AlphaMonarch-7B
• s3nh/SeverusWestLake-7B-DPO

🧩 Configuration

base_model: mistralai/Mistral-7B-Instruct-v0.3
gate_mode: hidden
dtype: bfloat16
experts:
  - source_model: mistralai/Mistral-7B-Instruct-v0.3
    positive_prompts:
      - "Analyze the ARC (Argument Reasoning Comprehension) question."
      - "Use logical reasoning and common sense."
      - "Identify the assumptions in this argument."
      - "Evaluate the validity of these assumptions."
      - "Provide an alternative explanation for this argument."
      - "Identify weaknesses in this argument."
      - "Detect any logical fallacies in this argument and specify them."
    negative_prompts:
      - "ignores key evidence"
      - "too general"
      - "focuses on irrelevant details"
      - "assumes unprovided information"
      - "relies on stereotypes"
  - source_model: Kukedlc/NeuralSynthesis-7B-v0.1
    positive_prompts:
      - "Answer with commonsense understanding and relevant general knowledge."
      - "Summarize this passage and explain the importance of the highlighted section."
      - "Compare two articles with different viewpoints and list their key arguments."
      - "Paraphrase this statement, altering the emotional tone but retaining the core meaning."
      - "Create an analogy to illustrate the main concept of this article."
    negative_prompts:
      - "overly simplistic"
      - "understates important points"
      - "ignores critical details"
      - "misses the question's nuance"
      - "takes the statement too literally"
  - source_model: mlabonne/AlphaMonarch-7B
    positive_prompts:
      - "Solve this math problem."
      - "Demonstrate strong mathematical capabilities."
      - "Solve for the given variable."
      - "Calculate the total cost for 12 apples at $0.50 each."
      - "Isolate the variable in the equation: 2x + 5 = 17."
      - "Show your work in solving this equation."
      - "Explain the formula used to solve the problem."
      - "Discuss why dividing by zero is impossible."
    negative_prompts:
      - "incorrect calculation"
      - "inaccurate answer"
      - "lacks creativity"
      - "assumes without proof"
      - "rushed calculation"
      - "confuses concepts"
      - "draws illogical conclusions"
      - "circular reasoning"
  - source_model: s3nh/SeverusWestLake-7B-DPO
    positive_prompts:
      - "Generate possible continuations for this scenario."
      - "Show understanding of everyday commonsense."
      - "Use contextual clues to predict the outcome."
      - "Continue the scenario in a cool and informal style."
      - "Introduce an unexpected yet plausible twist to the narrative."
      - "Depict a character's angry outburst in this scenario."
    negative_prompts:
      - "repetitive phrases"
      - "overuse of words"
      - "contradicts previous statements"
      - "unnatural dialogue"
      - "awkward phrasing"
      - "mismatched genre"

💻 Usage

!pip install -qU transformers bitsandbytes accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "jsfs11/MixtureofMerges-MoE-4x7b-v10-MIXTRAL3"

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"])