gabrielmbmb's picture
gabrielmbmb HF staff
Adding Evaluation Results (#1)
7797fde verified
metadata
language:
  - en
license: apache-2.0
tags:
  - distilabel
  - dpo
  - rlaif
  - rlhf
  - merge
  - mergekit
datasets:
  - argilla/distilabel-intel-orca-dpo-pairs
model-index:
  - name: distilabeled-Marcoro14-7B-slerp-full
    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: 70.65
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=argilla/distilabeled-Marcoro14-7B-slerp-full
          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: 87.55
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=argilla/distilabeled-Marcoro14-7B-slerp-full
          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: 65.33
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=argilla/distilabeled-Marcoro14-7B-slerp-full
          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: 64.21
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=argilla/distilabeled-Marcoro14-7B-slerp-full
          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: 82
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=argilla/distilabeled-Marcoro14-7B-slerp-full
          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: 70.66
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=argilla/distilabeled-Marcoro14-7B-slerp-full
          name: Open LLM Leaderboard

⚗️ distilabeled Marcoro14 7B Slerp

Built with Distilabel

Introduction

This model is a new DPO fine-tune of our new open dataset argilla/distilabel-intel-orca-dpo-pairs, on the mlabonne/Marcoro14-7B-slerp model. You can find more information of the "distilabeled" dataset used at this repo argilla/distilabeled-Hermes-2.5-Mistral-7B, and visit distilabel.

The difference between this model and argilla/distilabeled-Marcoro14-7B-slerp is that this model has been fine-tuned for a whole epoch instead instead of 200 steps, so it has seen the whole dataset.

Training details

As we did with Notus, we wanted a reproducible recipe to test the impact of data quality.

And we're lucky to have so many amazing folks in the open community contributing reproducible, easy-to-use training scripts and recipes. This time, Maxime Labonne had shared a Colab to fine-tune OpenHermes with DPO and the original Intel's dataset, perfect! We just updated the base model to mlabonne/Marcoro14-7B-slerp, and applied the same dataset recipe we used for argilla/distilabeled-Hermes-2.5-Mistral-7B:

from datasets import load_dataset

# Instead of this:
# dataset = load_dataset("Intel/orca_dpo_pairs", split="train")

# we did this
dataset = load_dataset("argilla/distilabel-intel-orca-dpo-pairs", split="train")

dataset = dataset.filter(
    lambda r: 
        r["status"] != "tie" and 
        r["chosen_score"] >= 8 and 
        not r["in_gsm8k_train"]
)

Benchmark results

For benchmarking we used the famous "Nous" or "Teknium" benchmark. You can find below an overview, including our first experiment with a less ambitious dataset filtering (removing ties and score>5).

For running the benchmark we used another awesome contribution from Maxime: LLM AutoEval, check it out!

Model AGIEval GPT4ALL TruthfulQA Bigbench Average
argilla/distilabeled-Marcoro14-7B-slerp-full 45.17 76.59 64.68 48.15 58.65
argilla/distilabeled-Marcoro14-7B-slerp 45.4 76.47 65.46 47.19 58.63
Marcoro14-7B-slerp 44.66 76.24 64.15 45.64 57.67
argilla/distilabeled-Hermes-2.5-Mistral-7B 44.64 73.35 55.96 42.21 54.04

Training Hardware

We used 1 x A100 80GB in runpod for less than 2 hours.

Acknowledgements

We'd like to thank the amazing open community and in particular:

  • The Intel team for publishing a great open dataset and show how well it worked in the first place
  • Teknium and NousResearch for their awesome work and models.
  • Maxime for sharing such great resources.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 73.40
AI2 Reasoning Challenge (25-Shot) 70.65
HellaSwag (10-Shot) 87.55
MMLU (5-Shot) 65.33
TruthfulQA (0-shot) 64.21
Winogrande (5-shot) 82.00
GSM8k (5-shot) 70.66