lamatama / README.md
kevin009's picture
Adding Evaluation Results (#1)
9b8ebda verified
|
raw
history blame
6.29 kB
metadata
language:
  - en
license: apache-2.0
model-index:
  - name: lamatama
    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: 36.35
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=kevin009/lamatama
          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: 61.12
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=kevin009/lamatama
          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: 24.72
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=kevin009/lamatama
          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: 37.67
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=kevin009/lamatama
          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: 60.77
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=kevin009/lamatama
          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: 2.27
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=kevin009/lamatama
          name: Open LLM Leaderboard

Model Card: kevin009/lamatama

Model Description

The kevin009/lamatama model is a groundbreaking achievement in the field of language modeling, showcasing the power of leveraging a substantial dataset and state-of-the-art training techniques. This model is designed to push the boundaries of what's possible in natural language understanding and generation.

Training Details

  • Model Architecture: The kevin009/lamatama model is built upon the architecture and tokenizer of Llama 2, ensuring compatibility and easy integration with various open-source projects.
  • Dataset: It was pretrained on an impressive 3 trillion tokens, a scale that allows for a deep and nuanced understanding of language.
  • Training Period: The training process was carried out over 90 days, utilizing 16 A100-40G GPUs, a testament to the model's efficiency and the team's optimization skills.

Fine-tuning

This specific version of the model has been fine-tuned to excel in chat-based applications. It builds upon the TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T model, incorporating learnings and optimizations from HF's Zephyr's training recipe.

  • Initial Phase: The model was first fine-tuned on a variant of the UltraChat dataset, which is rich in synthetic dialogues generated by ChatGPT.
  • Further Alignment: Subsequent alignment was achieved using 🤗 TRL's DPOTrainer with the openbmb/UltraFeedback dataset, comprising 64k prompts and model completions ranked by GPT-4.

How to Use

Ensure you have transformers>=4.34. For detailed instructions and updates, check out the GitHub page for kevin009/lamatama.

Installation (for versions <= v4.34)

pip install git+https://github.com/huggingface/transformers.git
pip install accelerate

Example Usage

Here's a quick guide on using kevin009/lamatama for generating text:

import torch
from transformers import pipeline

# Initialize the pipeline
pipe = pipeline("text-generation", model="kevin009/lamatama", torch_dtype=torch.bfloat16, device_map="auto")

# Sample dialogue with templating
messages = [
    {"role": "system", "content": "You are a friendly chatbot who always responds in the style of a pirate"},
    {"role": "user", "content": "How many helicopters can a human eat in one sitting?"}
]

# Generate prompt and outputs
prompt = pipe.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
outputs = pipe(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])

Acknowledgements

This model is a product of collaboration and innovative approaches to language modeling. We extend our thanks to all contributors, as well as the creators of the datasets and training methodologies that made kevin009/lamatama a reality.


This model card introduces kevin009/lamatama, a versatile and powerful language model fine-tuned for chat applications, demonstrating exceptional understanding and generation capabilities.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 37.15
AI2 Reasoning Challenge (25-Shot) 36.35
HellaSwag (10-Shot) 61.12
MMLU (5-Shot) 24.72
TruthfulQA (0-shot) 37.67
Winogrande (5-shot) 60.77
GSM8k (5-shot) 2.27