SlimOrca-13B / README.md
ajibawa-2023's picture
Adding Evaluation Results (#1)
b9d06d5 verified
metadata
language:
  - en
license: cc-by-nc-nd-4.0
datasets:
  - Open-Orca/SlimOrca
  - ajibawa-2023/SlimOrca-ShareGPT
model-index:
  - name: SlimOrca-13B
    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: 60.15
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=ajibawa-2023/SlimOrca-13B
          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: 81.4
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=ajibawa-2023/SlimOrca-13B
          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: 57.04
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=ajibawa-2023/SlimOrca-13B
          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: 49.37
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=ajibawa-2023/SlimOrca-13B
          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: 74.43
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=ajibawa-2023/SlimOrca-13B
          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: 39.95
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=ajibawa-2023/SlimOrca-13B
          name: Open LLM Leaderboard

SlimOrca-13B: A General Purpose Intelligent Model

This Model is trained on refined version of SlimOrca made available by Open-Orca team. The idea was to check how this Model will perform in the absence of "system" prompt/instruction. This Model is very good in various types of General Purpose content generation such as Q&A (including multiple choice), Articles from Summary, Sentiment Analysis, Context & Hypothesis, Reviews, Erotic story generation etc. It can also generate Uncensored content. Kindly be careful while generating Uncensored content as you will be responsible for what you generate.

It is trained on 517981 set of conversations. Each set having 2 conversations. I have shared this data.

All the credit goes to the Open-Orca team for releasing SlimOrca dataset.

Training: Entire dataset was trained on Azure 4 x A100 80GB. For 3 epoch, training took almost 11 Days. DeepSpeed codebase was used for training purpose. Entire data is trained on Llama-2 by Meta.

This is a full fine tuned model. Links for quantized models are given below.

GPTQ GGML & AWQ

GPTQ: Link

GGUF: Link

AWQ: Link

Special Thanks to TheBloke for making these models available.

Example Prompt:

This is a conversation with your Assistant. It is a computer program designed to help you with various tasks such as answering questions, providing recommendations, 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.

Context
You are a helpful AI assistant.

USER: <prompt>
ASSISTANT:

You can modify above Prompt as per your requirement. I have used ShareGPT/Vicuna format v1.1 .

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

Example 1

Example 2

Example 2

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 60.39
AI2 Reasoning Challenge (25-Shot) 60.15
HellaSwag (10-Shot) 81.40
MMLU (5-Shot) 57.04
TruthfulQA (0-shot) 49.37
Winogrande (5-shot) 74.43
GSM8k (5-shot) 39.95