metadata

license: llama3
model-index:
  - name: Llama-3.1-8B-ArliAI-RPMax-v1.1
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: IFEval (0-Shot)
          type: HuggingFaceH4/ifeval
          args:
            num_few_shot: 0
        metrics:
          - type: inst_level_strict_acc and prompt_level_strict_acc
            value: 63.59
            name: strict accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=ArliAI/Llama-3.1-8B-ArliAI-RPMax-v1.1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: BBH (3-Shot)
          type: BBH
          args:
            num_few_shot: 3
        metrics:
          - type: acc_norm
            value: 28.79
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=ArliAI/Llama-3.1-8B-ArliAI-RPMax-v1.1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MATH Lvl 5 (4-Shot)
          type: hendrycks/competition_math
          args:
            num_few_shot: 4
        metrics:
          - type: exact_match
            value: 11.33
            name: exact match
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=ArliAI/Llama-3.1-8B-ArliAI-RPMax-v1.1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GPQA (0-shot)
          type: Idavidrein/gpqa
          args:
            num_few_shot: 0
        metrics:
          - type: acc_norm
            value: 4.47
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=ArliAI/Llama-3.1-8B-ArliAI-RPMax-v1.1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MuSR (0-shot)
          type: TAUR-Lab/MuSR
          args:
            num_few_shot: 0
        metrics:
          - type: acc_norm
            value: 5.31
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=ArliAI/Llama-3.1-8B-ArliAI-RPMax-v1.1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU-PRO (5-shot)
          type: TIGER-Lab/MMLU-Pro
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 28.35
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=ArliAI/Llama-3.1-8B-ArliAI-RPMax-v1.1
          name: Open LLM Leaderboard

Llama-3.1-8B-ArliAI-RPMax-v1.1

=====================================

RPMax Series Overview

| 3.8B | 8B | 12B | 70B |

RPMax is a series of models that are trained on a diverse set of curated creative writing and RP datasets with a focus on variety and deduplication. This model is designed to be highly creative and non-repetitive by making sure no two entries in the dataset have repeated characters or situations, which makes sure the model does not latch on to a certain personality and be capable of understanding and acting appropriately to any characters or situations.

Early tests by users mentioned that these models does not feel like any other RP models, having a different style and generally doesn't feel in-bred.

You can access the model at https://arliai.com and ask questions at https://www.reddit.com/r/ArliAI/

We also have a models ranking page at https://www.arliai.com/models-ranking

Ask questions in our new Discord Server! https://discord.gg/aDVx6FZN

Model Description

Llama-3.1-8B-ArliAI-RPMax-v1.1 is a variant of the Meta-Llama-3.1-8B model.

v1.1 is just a small fix to not train and save the embeddings layer and small changes to the dataset, since v1.0 had the lm_head unnecessarily trained on accident.

Training Details

Sequence Length: 8192
Training Duration: Approximately 1 day on 2x3090Ti
Epochs: 1 epoch training for minimized repetition sickness
LORA: 64-rank 128-alpha, resulting in ~2% trainable weights
Learning Rate: 0.00001
Gradient accumulation: Very low 32 for better learning.

Quantization

The model is available in quantized formats:

We recommend using full weights or GPTQ

FP16: https://huggingface.co/ArliAI/Llama-3.1-8B-ArliAI-RPMax-v1.1
GPTQ_Q4: https://huggingface.co/ArliAI/Llama-3.1-8B-ArliAI-RPMax-v1.1-GPTQ_Q4
GPTQ_Q8: https://huggingface.co/ArliAI/Llama-3.1-8B-ArliAI-RPMax-v1.1-GPTQ_Q8
GGUF: https://huggingface.co/ArliAI/Llama-3.1-8B-ArliAI-RPMax-v1.1-GGUF

Suggested Prompt Format

Llama 3 Instruct Format

Example:

<|begin_of_text|><|start_header_id|>system<|end_header_id|>

You are [character]. You have a personality of [personality description]. [Describe scenario]<|eot_id|><|start_header_id|>user<|end_header_id|>

{{ user_message_1 }}<|eot_id|><|start_header_id|>assistant<|end_header_id|>

{{ model_answer_1 }}<|eot_id|><|start_header_id|>user<|end_header_id|>

{{ user_message_2 }}<|eot_id|><|start_header_id|>assistant<|end_header_id|>

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	23.64
IFEval (0-Shot)	63.59
BBH (3-Shot)	28.79
MATH Lvl 5 (4-Shot)	11.33
GPQA (0-shot)	4.47
MuSR (0-shot)	5.31
MMLU-PRO (5-shot)	28.35