922CA's picture
Update README.md
1387592 verified
metadata
language:
  - en
license: llama3
tags:
  - text-generation-inference
  - transformers
  - unsloth
  - llama
  - trl
base_model: unsloth/llama-3-8b-bnb-4bit
model-index:
  - name: Llama-3-natsuki-ddlc-8b-v1
    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: 60.32
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=922-CA/Llama-3-natsuki-ddlc-8b-v1
          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: 80.99
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=922-CA/Llama-3-natsuki-ddlc-8b-v1
          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: 64.75
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=922-CA/Llama-3-natsuki-ddlc-8b-v1
          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: 45.41
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=922-CA/Llama-3-natsuki-ddlc-8b-v1
          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: 78.77
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=922-CA/Llama-3-natsuki-ddlc-8b-v1
          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: 41.85
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=922-CA/Llama-3-natsuki-ddlc-8b-v1
          name: Open LLM Leaderboard
datasets:
  - 922-CA/NaChA_v1

Llama-3-Natsuki-DDLC-8b-v1

  • Model fine-tuned for Natsuki character from DDLC
  • Base: LLaMA-3
  • GGUFs

USAGE

For best results: replace "Human" and "Assistant" with "Player" and "Natsuki" like so:

\nPlayer: (prompt)\nNatsuki:

HYPERPARAMS

  • Trained for 1 epoch
  • rank: 32
  • lora alpha: 32
  • lora dropout: 0
  • lr: 2e-4
  • batch size: 2
  • grad steps: 4

This llama model was trained 2x faster with Unsloth and Huggingface's TRL library.

WARNINGS AND DISCLAIMERS

This model is meant to closely reflect the characteristics of Natsuki. Despite this, there is always the chance that "Natsuki" will hallucinate and get information about herself wrong or act out of character.

Finally, this model is not guaranteed to output aligned or safe outputs, use at your own risk.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 62.02
AI2 Reasoning Challenge (25-Shot) 60.32
HellaSwag (10-Shot) 80.99
MMLU (5-Shot) 64.75
TruthfulQA (0-shot) 45.41
Winogrande (5-shot) 78.77
GSM8k (5-shot) 41.85