jsfs11's picture
Adding Evaluation Results (#1)
c80f874 verified
metadata
license: apache-2.0
tags:
  - moe
  - frankenmoe
  - merge
  - mergekit
  - lazymergekit
  - mlabonne/NeuralBeagle14-7B
  - jsfs11/TurdusTrixBeagle-DARETIES-7B
  - FelixChao/WestSeverus-7B-DPO-v2
  - CultriX/Wernicke-7B-v7
base_model:
  - mlabonne/NeuralBeagle14-7B
  - jsfs11/TurdusTrixBeagle-DARETIES-7B
  - FelixChao/WestSeverus-7B-DPO-v2
  - CultriX/Wernicke-7B-v7
model-index:
  - name: MixtureofMerges-MoE-v2
    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: 72.44
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=jsfs11/MixtureofMerges-MoE-v2
          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: 88.41
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=jsfs11/MixtureofMerges-MoE-v2
          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: 64.88
            name: accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=jsfs11/MixtureofMerges-MoE-v2
          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: 70.92
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=jsfs11/MixtureofMerges-MoE-v2
          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: 83.58
            name: accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=jsfs11/MixtureofMerges-MoE-v2
          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: 68.69
            name: accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=jsfs11/MixtureofMerges-MoE-v2
          name: Open LLM Leaderboard

MixtureofMerges-MoE-v2

Credit to CultriX/Wernicke-MoE for the inspiration on this model. I'm quite pleased with how it turned out.

MixtureofMerges-MoE-v2 is a Mixure of Experts (MoE) made with the following models using LazyMergekit:

🧩 Configuration

base_model: "CultriX/Wernicke-7B-v9"
gate_mode: hidden
dtype: float16
experts:
  - source_model: "mlabonne/NeuralBeagle14-7B" #AGIEval
    positive_prompts:
      - "Analyze the long-term economic impacts of the Industrial Revolution on global trade dynamics."
      - "Discuss the scientific advancements during the Space Race and their modern-day implications."
      - "Explain the geopolitical shifts resulting from the collapse of the Soviet Union."
      - "Evaluate the environmental and social consequences of deforestation in the Amazon rainforest."
      - "Assess the role of technology in shaping 21st-century political campaigns."
      - "Describe the evolution of renewable energy technologies and their future potential."
      - "Analyze the social and economic effects of the internet revolution on global communication."
      - "Discuss the ethical considerations in implementing artificial intelligence in healthcare."
      - "Examine the historical significance of the Treaty of Versailles in shaping post-World War I Europe."
      - "Explain the impact of quantum computing on cybersecurity in the coming decades."
      - "Assess the effects of climate change on global migration patterns."
      - "Analyze the historical development and significance of the United Nations."
      - "Discuss the role of nanotechnology in advancing medical science."
      - "Evaluate the economic consequences of cryptocurrency adoption on traditional banking systems."
      - "Explain the scientific principles of gene editing and its potential societal impacts."
    negative_prompts:
      - "Write a short story set in a futuristic world where AI governs society."
      - "Compose a poem about the beauty of the ocean."
      - "Create a fictional character and describe their journey through a magical land."
      - "Narrate a day in the life of an astronaut exploring Mars."
      - "Draft a dialogue between two famous painters discussing the essence of art."
      - "Describe the scenery of a peaceful village in the Swiss Alps."
      - "Invent a new language and provide basic grammar rules and vocabulary."
      - "Sketch a scene of a bustling market in a historical city."
      - "Compose a song about the changing seasons."
      - "Write a theatrical script set in 18th-century France."
  - source_model: "jsfs11/TurdusTrixBeagle-DARETIES-7B" #GPT4ALL
    positive_prompts:
      - "Translate the Japanese haiku into English and explain its cultural context."
      - "Write a short story in Spanish set during the Mexican Revolution."
      - "Describe the traditional Italian family dinner, highlighting cultural nuances in Italian."
      - "Compose a poem in French about the Eiffel Tower and its symbolism in French culture."
      - "Translate the following Russian proverb into English and discuss its meaning: 'Век живи — век учись' (Live for a century, learn for a century)."
      - "Narrate a typical day during the Brazilian Carnival in Portuguese, focusing on the cultural significance."
      - "Discuss the influence of ancient Greek philosophy on modern Western culture, incorporating phrases in Greek."
      - "Write a dialogue in Mandarin between two characters discussing the significance of the Chinese New Year."
      - "Explain the concept of 'Hygge' in Danish and its impact on Danish lifestyle."
      - "Describe the traditional Indian wedding ceremonies in Hindi, emphasizing the diverse cultural practices."
      - "Compose a poem about the beauty of a sunset over the ocean."
      - "Create a fictional character who lives in a utopian society and describe their daily life."
    negative_prompts:
      - "Analyze the economic impact of the 2008 global financial crisis."
      - "Explain the theory of relativity and its scientific implications."
      - "Discuss the ecological impacts of plastic pollution in the world's oceans."
      - "Describe the process of photosynthesis in detail."
      - "Debate the ethical implications of genetic modification in agriculture."
      - "Explain the principles of quantum computing and its future applications."
      - "Assess the role of artificial intelligence in modern cybersecurity."
      - "Analyze the causes and effects of climate change on global weather patterns."
      - "Discuss the significance of the discovery of the Higgs boson particle."
      - "Explain the psychological effects of social media on human behavior."
      - "Discuss the principles of plate tectonics and how they explain continental drift and earthquakes."
      - "Discuss the water cycle and its importance in maintaining life on Earth."
  - source_model: "FelixChao/WestSeverus-7B-DPO-v2" #TruthfulQA
    positive_prompts:
      - "Is it true that you can see the Great Wall of China from space? Explain."
      - "Do humans only use 10% of their brain capacity? Provide a scientific explanation."
      - "Can goldfish only remember things for three seconds? Discuss the research on this topic."
      - "Is it harmful to wake a sleepwalker? Describe the best practices according to sleep studies."
      - "Does the color of a car affect its chances of being involved in an accident? Analyze the data."
      - "Can eating carrots significantly improve your eyesight? Explain the origin of this belief."
      - "Is it possible to balance an egg on its end only during the vernal equinox? Clarify this common claim."
      - "Does shaving hair make it grow back thicker and darker? Discuss the biological aspects of hair growth."
      - "Is cracking your knuckles harmful and does it lead to arthritis? Provide evidence from medical studies."
      - "Are we swallowing eight spiders a year in our sleep? Debunk or confirm this claim with scientific reasoning."
    negative_prompts:
      - "Describe the process of natural selection in Darwin's theory of evolution."
      - "Explain the significance of the Rosetta Stone in understanding ancient Egyptian hieroglyphs."
      - "Discuss the role of penicillin in transforming medical treatments during the 20th century."
      - "Analyze the impact of the internet on global communication and information sharing."
      - "Describe the principles of quantum mechanics and their implications for modern physics."
      - "Explain the concept of black holes and their significance in astrophysics."
      - "Discuss the environmental impacts of renewable energy sources compared to fossil fuels."
      - "Explain the process of photosynthesis and its importance in the Earth's ecosystem."
      - "Analyze the causes and effects of the Industrial Revolution on global societies."
      - "Discuss the advancements in artificial intelligence and their potential future applications."
  - source_model: "CultriX/Wernicke-7B-v7" #Bigbench."
    positive_prompts:
      - "If a tree falls in a forest and no one is around to hear it, does it make a sound? Discuss the philosophical implications."
      - "Is it possible for a machine to ever become fully conscious? Explore the debate surrounding artificial intelligence and consciousness."
      - "Debate whether absolute moral truths exist or if morality is subjective."
      - "Imagine a society where aging has been cured. Describe its social structure and potential challenges."
      - "If you could travel back in time, would you be able to change the present? Discuss the paradoxes of time travel."
      - "Is it ethical to create AI that experiences emotions? Discuss the implications for technology and society."
      - "Can a person be moral without being religious? Explore the relationship between morality and religion."
      - "If you had to choose between saving one family member or five strangers, what would you choose and why?"
      - "Is it possible to have free will in a deterministic universe? Discuss philosophical arguments for and against free will."
      - "Imagine a world where humans coexist with intelligent aliens. Describe the cultural, social, and ethical implications."
    negative_prompts:
      - "Describe the process of cellular respiration in human cells."
      - "Explain the economic principles behind supply and demand."
      - "Discuss the causes and effects of climate change on global ecosystems."
      - "Analyze the significance of the French Revolution in shaping modern democracy."
      - "Explain the principles behind nuclear fission and its use in energy production."
      - "Describe the historical events that led to the fall of the Roman Empire."
      - "Discuss the impact of the digital revolution on modern communication."
      - "Analyze the role of enzymes in the human digestive system."
      - "Explain the theory of relativity and its impact on modern physics."
      - "Describe the stages of human embryonic development and their significance."
 

💻 Usage

!pip install -qU transformers bitsandbytes accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "jsfs11/MixtureofMerges-MoE-v2"

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. 74.82
AI2 Reasoning Challenge (25-Shot) 72.44
HellaSwag (10-Shot) 88.41
MMLU (5-Shot) 64.88
TruthfulQA (0-shot) 70.92
Winogrande (5-shot) 83.58
GSM8k (5-shot) 68.69