Magician-MoE-4x7B / README.md
leaderboard-pr-bot's picture
Adding Evaluation Results
38cfa1e verified
|
raw
history blame
5.32 kB
metadata
license: apache-2.0
tags:
  - moe
  - merge
  - deepseek-ai/deepseek-coder-6.7b-instruct
  - ise-uiuc/Magicoder-S-CL-7B
  - WizardLM/WizardMath-7B-V1.0
  - WizardLM/WizardCoder-Python-7B-V1.0
model-index:
  - name: Magician-MoE-4x7B
    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: 28.24
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=FelixChao/Magician-MoE-4x7B
          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: 30.06
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=FelixChao/Magician-MoE-4x7B
          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: 24.67
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=FelixChao/Magician-MoE-4x7B
          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: 0
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=FelixChao/Magician-MoE-4x7B
          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: 49.88
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=FelixChao/Magician-MoE-4x7B
          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=FelixChao/Magician-MoE-4x7B
          name: Open LLM Leaderboard

Magician-MoE-4x7B

Magician-MoE-4x7B is a Mixure of Experts (MoE) made with the following models:

🧩 Configuration

base_model: ise-uiuc/Magicoder-S-CL-7B
gate_mode: cheap_embed
experts:
  - source_model: deepseek-ai/deepseek-coder-6.7b-instruct
    positive_prompts: ["You are an AI coder","coding","Java expert"]
  - source_model: ise-uiuc/Magicoder-S-CL-7B
    positive_prompts: ["You are an AI programmer","programming","C++ expert"]
  - source_model: WizardLM/WizardMath-7B-V1.0
    positive_prompts: ["Math problem solving","Think step by step","Math expert"]
  - source_model: WizardLM/WizardCoder-Python-7B-V1.0
    positive_prompts: ["Great at Deep learning","Algorithm and Data Structure","Python expert"]

💻 Usage

!pip install -qU transformers bitsandbytes accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "FelixChao/Magician-MoE-4x7B"

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. 22.14
AI2 Reasoning Challenge (25-Shot) 28.24
HellaSwag (10-Shot) 30.06
MMLU (5-Shot) 24.67
TruthfulQA (0-shot) 0.00
Winogrande (5-shot) 49.88
GSM8k (5-shot) 0.00