Pulsar_7B / README.md
rmdhirr's picture
Update README.md
c63efbf verified
|
raw
history blame
5.45 kB
metadata
language:
  - en
license: apache-2.0
library_name: transformers
tags:
  - text-generation-inference
  - transformers
  - unsloth
  - mistral
  - trl
  - dpo
  - uncensored
  - roleplay
  - fine-tune
base_model: MTSAIR/multi_verse_model
datasets:
  - grimulkan/theory-of-mind
  - grimulkan/physical-reasoning
  - ResplendentAI/Luna_Alpaca
  - unalignment/toxic-dpo-v0.2
  - kira/math-dpo
  - athirdpath/DPO_Pairs-Roleplay-Alpaca-NSFW-v1-SHUFFLED
model-index:
  - name: Pulsar_7B
    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: 69.71
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=rmdhirr/Pulsar_7B
          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: 86.99
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=rmdhirr/Pulsar_7B
          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: 63.72
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=rmdhirr/Pulsar_7B
          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: 69.28
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=rmdhirr/Pulsar_7B
          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: 84.06
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=rmdhirr/Pulsar_7B
          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: 71.65
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=rmdhirr/Pulsar_7B
          name: Open LLM Leaderboard

💫 Pulsar_7B

A more compliant, uncensored, RP-oriented version of MTSAIR/multi_verse_model, fine-tuned on carefully selected datasets. It's smart, soulful, and adept at following the desired markdown format and adhering to the provided character card. The first message of the character card significantly influences its writing style. Pulsar_7B pairs well with guidance from CFG Scale and works effectively with PLists + Ali:Chat character cards. Pulsar_7B was fine-tuned on the following datasets:

  • grimulkan/theory-of-mind
  • grimulkan/physical-reasoning
  • ResplendentAI/Luna_Alpaca
  • unalignment/toxic-dpo-v0.2
  • kira/math-dpo
  • athirdpath/DPO_Pairs-Roleplay-Alpaca-NSFW-v1-SHUFFLED

Quantizations

Thanks to mradermacher, static GGUF quants are available here.

Formatting

Pulsar_7B works well with Alpaca, it's not a picky model when it comes to formatting. Mistral should be compatible too. The custom chat template from MTSAIR/multi_verse_model also performs well:

{% for message in messages %}{% if message['role'] == 'user' %}{{ '### Instruction:\n' + message['content'] + '\n### Response:\n' }}{% elif message['role'] == 'assistant' %}{{ message['content'] + eos_token}}{% elif message['role'] == 'system' %}{{ '### System:\n' + message['content'] + '\n' }}{% endif %}{% endfor %}

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

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 74.23
AI2 Reasoning Challenge (25-Shot) 69.71
HellaSwag (10-Shot) 86.99
MMLU (5-Shot) 63.72
TruthfulQA (0-shot) 69.28
Winogrande (5-shot) 84.06
GSM8k (5-shot) 71.65