gabrielmbmb's picture
gabrielmbmb HF staff
Adding Evaluation Results (#2)
d06c867 verified
metadata
language:
  - en
license: apache-2.0
library_name: trl
tags:
  - distilabel
  - dpo
  - rlaif
  - rlhf
datasets:
  - argilla/dpo-mix-7k
base_model: teknium/OpenHermes-2.5-Mistral-7B
model-index:
  - name: CapybaraHermes-2.5-Mistral-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: 65.78
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=argilla/CapybaraHermes-2.5-Mistral-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: 85.45
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=argilla/CapybaraHermes-2.5-Mistral-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: 63.13
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=argilla/CapybaraHermes-2.5-Mistral-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: 56.91
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=argilla/CapybaraHermes-2.5-Mistral-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: 78.3
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=argilla/CapybaraHermes-2.5-Mistral-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: 59.29
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=argilla/CapybaraHermes-2.5-Mistral-7B
          name: Open LLM Leaderboard

CapybaraHermes-2.5-Mistral-7B

Built with Distilabel

This model is the launching partner of the capybara-dpo dataset build with ⚗️ distilabel. It's a preference tuned OpenHermes-2.5-Mistral-7B.

CapybaraHermes has been preference tuned with LoRA and TRL for 3 epochs using argilla's dpo mix 7k.

To test the impact on multi-turn performance we have used MTBench. We also include the Nous Benchmark results and Mistral-7B-Instruct-v0.2 for reference as it's a strong 7B model on MTBench:

Model AGIEval GPT4All TruthfulQA Bigbench MTBench First Turn MTBench Second Turn Nous avg. MTBench avg.
argilla/CapybaraHermes-2.5-Mistral-7B 43.8 73.35 57.07 42.44 8.24375 7.5625 54.16 7.903125
teknium/OpenHermes-2.5-Mistral-7B 42.75 72.99 52.99 40.94 8.25 7.2875 52.42 7.76875
Mistral-7B-Instruct-v0.2 38.5 71.64 66.82 42.29 7.8375 7.1 54.81 7.46875

The most interesting aspect in the context of the capybara-dpo dataset is the increased performance in MTBench Second Turn scores.

For the merge lovers, we also preference tuned Beagle14-7B with a mix of capybara-dpo and distilabel orca pairs using the same recipe as NeuralBeagle (see YALL - Yet Another LLM Leaderboard for reference):

Model AGIEval GPT4All TruthfulQA Bigbench Average
DistilabelBeagle14-7B 45.29 76.92 71.66 48.78 60.66

Model Details

Model Description

This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.

  • Developed by: Argilla
  • Shared by [optional]: Argilla
  • Model type: 7B chat model
  • Language(s) (NLP): English
  • License: Same as OpenHermes
  • Finetuned from model [optional]: OpenHermes-2.5-Mistral-7B

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 68.14
AI2 Reasoning Challenge (25-Shot) 65.78
HellaSwag (10-Shot) 85.45
MMLU (5-Shot) 63.13
TruthfulQA (0-shot) 56.91
Winogrande (5-shot) 78.30
GSM8k (5-shot) 59.29