README.md

metadata

license: mit
model-index:
  - name: BruinsV2-OpHermesNeu-11B
    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: 68.09
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Ba2han/BruinsV2-OpHermesNeu-11B
          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: 84.7
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Ba2han/BruinsV2-OpHermesNeu-11B
          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: 64.19
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Ba2han/BruinsV2-OpHermesNeu-11B
          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: 62.76
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Ba2han/BruinsV2-OpHermesNeu-11B
          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: 79.48
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Ba2han/BruinsV2-OpHermesNeu-11B
          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: 60.05
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Ba2han/BruinsV2-OpHermesNeu-11B
          name: Open LLM Leaderboard

Task	Version	Metric	Value		Stderr
arc_challenge	0	acc	0.6527	±	0.0139
		acc_norm	0.6869	±	0.0136

Warning! This model may or may not be contaminated See discussion 474. What a shame. It still does perform well though

A passthrough merge of OpenHermes-2.5-neural-chat-7b-v3-1 and Bruins-V2. To be updated.

Template: ChatML

My settings:

Temperature: 0.7-0.8

Min_p: 0.12

Top_K: 0

Repetition Penalty: 1.16

Mirostat Tau: 2.5-3

Mirostat Eta: 0.12

Personal Thoughts:

• The model sometimes throws wrong tags, you can add those to "Custom stopping strings" in Oobabooga.
• Output with Mirostat consistently felt smarter than a set Top_K rate.

Note: The model is hallucinating hard in chat mode for me in some instances, like writing adblocker messages. Kind of funny.

I am not sure which dataset involved was poisoned.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	69.88
AI2 Reasoning Challenge (25-Shot)	68.09
HellaSwag (10-Shot)	84.70
MMLU (5-Shot)	64.19
TruthfulQA (0-shot)	62.76
Winogrande (5-shot)	79.48
GSM8k (5-shot)	60.05