monika-ddlc-7b-v1 / README.md
922CA's picture
Adding Evaluation Results (#1)
ddd1a2d verified
metadata
license: llama2
datasets:
  - 922-CA/MoCha_v1
model-index:
  - name: monika-ddlc-7b-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: 54.95
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=922-CA/monika-ddlc-7b-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: 76.78
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=922-CA/monika-ddlc-7b-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: 45.61
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=922-CA/monika-ddlc-7b-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: 43.94
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=922-CA/monika-ddlc-7b-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: 72.85
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=922-CA/monika-ddlc-7b-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: 8.79
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=922-CA/monika-ddlc-7b-v1
          name: Open LLM Leaderboard

monika-ddlc-7b-v1:

  • LLaMA-2 7b chat fine-tuned for Monika character from DDLC (still somewhat experimental)
  • Fine-tuned on a dataset of ~600+ items (dialogue scraped from game, reddit, and Twitter augmented by l2-7b-monika-v0.3c1 to turn each into snippets of multi-turn chat dialogue between Player and Monika; this was then manually edited, with more manually crafted items including info about character added in)
  • GGUFs (feel free to convert to GGML, or can do it if requested)
  • QLoras

USAGE

This is meant to be mainly a chat model with limited RP ability.

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

\nPlayer: (prompt)\nMonika:

HYPERPARAMS

  • Trained for 3 epochs
  • rank: 32
  • lora alpha: 64
  • lora dropout: 0.5
  • lr: 2e-4
  • batch size: 2
  • warmup ratio: 0.1
  • grad steps: 4

WARNINGS AND DISCLAIMERS

This model is meant to closely reflect the characteristics of Monika. Despite this, there is always the chance that "Monika" will hallucinate and get information about herself wrong or act out of character (for example, in testing she usually knows her own club and its members, her game, and even her height and favorite ice cream flavor, but may still get her eye color wrong or mistake her developer as being a member of her club).

Additionally, being character-focused means that this model may not be the smartest model/not as capable as others for simple tasks (not yet tested).

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

Note: Ideally, would have liked to fine-tune on other models (specifically Zephyr-7b-alpha). May try soon for later versions.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 50.49
AI2 Reasoning Challenge (25-Shot) 54.95
HellaSwag (10-Shot) 76.78
MMLU (5-Shot) 45.61
TruthfulQA (0-shot) 43.94
Winogrande (5-shot) 72.85
GSM8k (5-shot) 8.79