limin(gate)
Adding Evaluation Results (#1)
414776d verified
metadata
license: apache-2.0
tags:
  - moe
  - frankenmoe
  - merge
  - mergekit
  - lazymergekit
  - M4-ai/TinyMistral-248M-v2-cleaner
  - Locutusque/TinyMistral-248M-Instruct
  - jtatman/tinymistral-v2-pycoder-instuct-248m
  - Locutusque/TinyMistral-248M-v2-Instruct
base_model:
  - M4-ai/TinyMistral-248M-v2-cleaner
  - Locutusque/TinyMistral-248M-Instruct
  - jtatman/tinymistral-v2-pycoder-instuct-248m
  - Locutusque/TinyMistral-248M-v2-Instruct
model-index:
  - name: TinyMistral-248Mx4-MOE
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: AI2 Reasoning Challenge (25-Shot)
          type: ai2_arc
          config: ARC-Challenge
          split: test
          args:
            num_few_shot: 25
        metrics:
          - type: acc_norm
            value: 29.52
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=222gate/TinyMistral-248Mx4-MOE
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: HellaSwag (10-Shot)
          type: hellaswag
          split: validation
          args:
            num_few_shot: 10
        metrics:
          - type: acc_norm
            value: 25.71
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=222gate/TinyMistral-248Mx4-MOE
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU (5-Shot)
          type: cais/mmlu
          config: all
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 24.82
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=222gate/TinyMistral-248Mx4-MOE
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: TruthfulQA (0-shot)
          type: truthful_qa
          config: multiple_choice
          split: validation
          args:
            num_few_shot: 0
        metrics:
          - type: mc2
            value: 48.66
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=222gate/TinyMistral-248Mx4-MOE
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: Winogrande (5-shot)
          type: winogrande
          config: winogrande_xl
          split: validation
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 51.78
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=222gate/TinyMistral-248Mx4-MOE
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GSM8k (5-shot)
          type: gsm8k
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 0
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=222gate/TinyMistral-248Mx4-MOE
          name: Open LLM Leaderboard

TinyMistral-248Mx4-MOE

TinyMistral-248Mx4-MOE is a Mixure of Experts (MoE) made with the following models using LazyMergekit:

🧩 Configuration

base_model: Locutusque/TinyMistral-248M-v2-Instruct
gate_mode: hidden
dtype: bfloat16
experts:
  - source_model: M4-ai/TinyMistral-248M-v2-cleaner
    positive_prompts:
    - "versatile"
    - "helpful"
    - "factual"
    - "integrated"
    - "adaptive"
    - "comprehensive"
    - "balanced"
    negative_prompts:
    - "specialized"
    - "narrow"
    - "focused"
    - "limited"
    - "specific"

  - source_model: Locutusque/TinyMistral-248M-Instruct
    positive_prompts:
    - "creative"
    - "chat"
    - "discuss"
    - "culture"
    - "world"
    - "expressive"
    - "detailed"
    - "imaginative"
    - "engaging"
    negative_prompts:
    - "sorry"
    - "cannot"
    - "factual"
    - "concise"
    - "straightforward"
    - "objective"
    - "dry"

  - source_model: jtatman/tinymistral-v2-pycoder-instuct-248m
    positive_prompts:
    - "analytical"
    - "accurate"
    - "logical"
    - "knowledgeable"
    - "precise"
    - "calculate"
    - "compute"
    - "solve"
    - "work"
    - "python"
    - "javascript"
    - "programming"
    - "algorithm"
    - "tell me"
    - "assistant"
    negative_prompts:
    - "creative"
    - "abstract"
    - "imaginative"
    - "artistic"
    - "emotional"
    - "mistake"
    - "inaccurate"

  - source_model: Locutusque/TinyMistral-248M-v2-Instruct
    positive_prompts:
    - "instructive"
    - "clear"
    - "directive"
    - "helpful"
    - "informative"
    negative_prompts:
    - "exploratory"
    - "open-ended"
    - "narrative"
    - "speculative"
    - "artistic"

💻 Usage

!pip install -qU transformers bitsandbytes accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "222gate/TinyMistral-248Mx4-MOE"

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"])

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 30.08
AI2 Reasoning Challenge (25-Shot) 29.52
HellaSwag (10-Shot) 25.71
MMLU (5-Shot) 24.82
TruthfulQA (0-shot) 48.66
Winogrande (5-shot) 51.78
GSM8k (5-shot) 0.00