ShortKing-3b-v0.2 / README.md
leaderboard-pr-bot's picture
Adding Evaluation Results
f7ef4a4 verified
|
raw
history blame
4.48 kB
metadata
language:
  - en
license: cc-by-4.0
datasets:
  - Photolens/alpaca-cleaned-airoboros-2.1-no-code-oasst1-en-merged
model-index:
  - name: ShortKing-3b-v0.3
    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: 40.96
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=AtAndDev/ShortKing-3b-v0.3
          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: 70.72
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=AtAndDev/ShortKing-3b-v0.3
          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: 26.21
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=AtAndDev/ShortKing-3b-v0.3
          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: 38.78
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=AtAndDev/ShortKing-3b-v0.3
          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: 66.93
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=AtAndDev/ShortKing-3b-v0.3
          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: 1.21
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=AtAndDev/ShortKing-3b-v0.3
          name: Open LLM Leaderboard

Model overview

This model is finetuned on a merged dataset of: oasst1-en, alpaca-cleaned and airoboros-2.1-no-code on a base model: Marx-3b-V2

  • License: "Creative-Commons-Attribution-4.0"
  • Language: "en"
  • Size: "3.43b params"

Prompt template

Prompt template:

### SYSTEM:
<system_prompt_here>

### HUMAN:
<prompter_message_here>

### INPUT:
<input_text_here>

### RESPONSE:
<leave_a_blank_line_here>

Note: If you dont have a system or input text, do not include the tokens in the prompt.

Training Details

This model took 2:40:54 to train in LoRA on a single A100 40gb GPU.

  • epochs: 1
  • train batch size: 8
  • eval batch size: 8
  • gradient accumulation steps: 1
  • maximum gradient normal: 0.3
  • learning rate: 2e-4
  • weight decay: 0.001
  • optimizer: paged_adamw_32bit
  • learning rate schedule: cosine
  • warmup ratio (linear): 0.03

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 40.80
AI2 Reasoning Challenge (25-Shot) 40.96
HellaSwag (10-Shot) 70.72
MMLU (5-Shot) 26.21
TruthfulQA (0-shot) 38.78
Winogrande (5-shot) 66.93
GSM8k (5-shot) 1.21