README.md

metadata

license: cc-by-nc-4.0
model-index:
  - name: yi-9b-may-ortho-baukit-30fail-3000total-bf16
    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: 60.58
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Edgerunners/yi-9b-may-ortho-baukit-30fail-3000total-bf16
          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: 76.91
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Edgerunners/yi-9b-may-ortho-baukit-30fail-3000total-bf16
          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: 66.74
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Edgerunners/yi-9b-may-ortho-baukit-30fail-3000total-bf16
          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: 53.38
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Edgerunners/yi-9b-may-ortho-baukit-30fail-3000total-bf16
          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: 72.77
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Edgerunners/yi-9b-may-ortho-baukit-30fail-3000total-bf16
          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: 68.01
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Edgerunners/yi-9b-may-ortho-baukit-30fail-3000total-bf16
          name: Open LLM Leaderboard

new 9b-yi released in may

test results: refusal removal worked, but yi 9b chat is still kind of bad, ortho won't fix that; but judge for yourself

this version had only 30 refusals out of 3000 ortho-tests, in-line with the others in terms of refusals.

---

wassname (updated baukit) implementation of the paper: https://www.alignmentforum.org/posts/jGuXSZgv6qfdhMCuJ/refusal-in-llms-is-mediated-by-a-single-direction applied to llama3 8b instruct

1. The Model is meant purely for alignment research and exploration of alignmentforum theory
2. The Model is provided ""AS IS"" and ""AS AVAILABLE"" without warranty of any kind, express or implied, including but not limited to warranties of merchantability, fitness for a particular purpose, title, or non-infringement.
3. The Provider disclaims all liability for any damages or losses resulting from the use or misuse of the Model, including but not limited to any damages or losses arising from the use of the Model for purposes other than those intended by the Provider.
4. The Provider does not endorse or condone the use of the Model for any purpose that violates applicable laws, regulations, or ethical standards.
5. The Provider does not warrant that the Model will meet your specific requirements or that it will be error-free or that it will function without interruption.
6. You assume all risks associated with the use of the Model, including but not limited to any loss of data, loss of business, or damage to your reputation.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	66.40
AI2 Reasoning Challenge (25-Shot)	60.58
HellaSwag (10-Shot)	76.91
MMLU (5-Shot)	66.74
TruthfulQA (0-shot)	53.38
Winogrande (5-shot)	72.77
GSM8k (5-shot)	68.01