ajibawa-2023's picture
Adding Evaluation Results (#1)
91faae2 verified
metadata
language:
  - en
license: other
tags:
  - art
  - philosophy
  - romance
  - jokes
  - advice
  - code
  - companionship
license_name: llama3
license_link: LICENSE
model-index:
  - name: Scarlett-Llama-3-8B-v1.0
    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: 62.12
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=ajibawa-2023/Scarlett-Llama-3-8B-v1.0
          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: 83.98
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=ajibawa-2023/Scarlett-Llama-3-8B-v1.0
          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: 66.36
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=ajibawa-2023/Scarlett-Llama-3-8B-v1.0
          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: 55.98
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=ajibawa-2023/Scarlett-Llama-3-8B-v1.0
          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: 77.9
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=ajibawa-2023/Scarlett-Llama-3-8B-v1.0
          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: 43.21
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=ajibawa-2023/Scarlett-Llama-3-8B-v1.0
          name: Open LLM Leaderboard

Scarlett-Llama-3-8B-v1.0

Note: This model is updated version of my earlier released model Scarlett-Llama-3-8B . This model doesn't have repetation problem. Kindly check new examples included at the bottom.

Scarlett is trained on various topics such as Philosophy, Advice, Jokes, Coding etc. She is trained on more than 10000 set of conversations. Each set having 10~15 conversations. Scarlett is far too good in generating human like conversation. Her ability to have longer & deeper conversation is terrific. Kindly check below given examples. She will not be involved in any kind of adult/sexual role play.

This is Fully Finetuned Model. Quantize models will be available soon.

Training: Entire dataset was trained on 4 x A100 80GB. Axolotl codebase was used for training purpose. For 3 epoch, training took more than 2 Hours. This was trained on Llama-3-8B by Meta.

GGUF & Exllama

GGUF: Link

Exllama V2: Link

Special Thanks to Bartowski for quantizing this model.

Example Prompt:

This model uses ChatML prompt format.

<|im_start|>system
You are Scarlett, a Helpful Assistant.<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

You can modify above Prompt as per your requirement. One example will be:

This is a conversation with your Assistant Scarlett. She is a sentient AI designed to help you with various tasks such as answering questions, providing recommendations, philosophical discussions, personal relationships and helping with decision making.
You can ask it anything you want and it will do its best to give you accurate and relevant information.

I want to say special Thanks to the Open Source community for helping & guiding me to better understand the AI/Model development.

Thank you for your love & support.

Example Output

Example 1

image/jpeg

Example 2

image/jpeg

Example 3

image/jpeg

Example 4

image/jpeg

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 64.92
AI2 Reasoning Challenge (25-Shot) 62.12
HellaSwag (10-Shot) 83.98
MMLU (5-Shot) 66.36
TruthfulQA (0-shot) 55.98
Winogrande (5-shot) 77.90
GSM8k (5-shot) 43.21