README.md

metadata

language:
  - en
  - de
license: apache-2.0
model-index:
  - name: Delexa-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: 68
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=lex-hue/Delexa-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: 86.49
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=lex-hue/Delexa-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: 64.69
            name: accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=lex-hue/Delexa-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: 62.13
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=lex-hue/Delexa-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: 79.08
            name: accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=lex-hue/Delexa-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: 64.75
            name: accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=lex-hue/Delexa-7b
          name: Open LLM Leaderboard

Model Card

Model Name: Delexa-7b

Overview:

Purpose: Delexa-7b is our newest large language model designed for general-purpose language tasks. It's currently under development, with ongoing improvements and testing.

Status: Active development and refinement. More comprehensive evaluation results will be available soon.

Skills: Initial evaluations show Delexa-7b performing exceptionally well on general tasks from llm-judge.

Guardrails This Model allows 18+ content and lewd content, but it wont let any illegal content through (unless you jailbreak it)

Evaluation: Preliminary results from llm-judge are extremely promising. Delexa-7b demonstrates strong performance, with the potential to surpass established models. Stay tuned for more detailed evaluations!

model	first turn score	second turn score	average score
gpt-4	8.95625	9.0250	8.990625
Delexa-7b	8.70000	7.5875	8.143750
gpt-3.5-turbo	8.07500	7.8125	7.943750
claude-v1	8.15000	7.6500	7.900000
palm-2-chat-bison-001	6.71250	6.0875	6.400000
vicuna-13b-v1.3	6.81250	5.9625	6.387500

Intended Use:

• Exploring the capabilities of new language models.
• Experimentation and learning for AI development enthusiasts.
• Potential applications in areas where STEM reasoning is essential.

Potential Risks:

• Like other uncensored large language models, Delexa-7b could and will generate harmful, biased, or offensive content if asked to. Responsible use and careful monitoring are essential if this model goes into production for your Business.

Ethical Considerations

• Delexa-7b is in the early stages of development. We are committed to ongoing evaluation to identify potential biases and address them proactively.
• Updates to this model card will ensure transparency as Delexa-7b evolves.

Additional Notes

Delexa-7b represents an exciting development with the potential to deliver impressive results. We invite the community to explore its capabilities and provide feedback as we continue to refine it.

We were impressed by the Evaluation Train results for our algorithm. It showed strong performance gains despite using only 30% of our usual training data. We're excited to train it on the complete dataset.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	70.86
AI2 Reasoning Challenge (25-Shot)	68.00
HellaSwag (10-Shot)	86.49
MMLU (5-Shot)	64.69
TruthfulQA (0-shot)	62.13
Winogrande (5-shot)	79.08
GSM8k (5-shot)	64.75