README.md

metadata

language:
  - en
license: llama3
tags:
  - text-generation-inference
  - transformers
  - unsloth
  - llama
  - trl
  - sft
base_model: Replete-AI/Llama-3-11.5B-Instruct-V2
model-index:
  - name: Llama-3-monika-ddlc-11.5b-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: 60.07
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=922CA/Llama-3-monika-ddlc-11.5b-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: 78.77
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=922CA/Llama-3-monika-ddlc-11.5b-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: 66.36
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=922CA/Llama-3-monika-ddlc-11.5b-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: 47.62
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=922CA/Llama-3-monika-ddlc-11.5b-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: 75.77
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=922CA/Llama-3-monika-ddlc-11.5b-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: 65.81
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=922CA/Llama-3-monika-ddlc-11.5b-v1
          name: Open LLM Leaderboard
datasets:
  - 922-CA/MoCha_v1a

monika-ddlc-11.5b-v1:

• LLaMA-3 11.5b fine-tuned for Monika character from DDLC (test)
• 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

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 2 epochs
• rank: 32
• lora alpha: 32
• lora dropout: 0.5
• lr: 2e-4
• batch size: 2
• warmup ratio: 0.1
• 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 Monika. Despite this, there is always the chance that "Monika" will hallucinate and get information about herself wrong or act out of character.

Additionally, being character-focused means that this model may not be the smartest model/not as capable as others for specific tasks.

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.	65.73
AI2 Reasoning Challenge (25-Shot)	60.07
HellaSwag (10-Shot)	78.77
MMLU (5-Shot)	66.36
TruthfulQA (0-shot)	47.62
Winogrande (5-shot)	75.77
GSM8k (5-shot)	65.81