smol_llama-220M-GQA / README.md
pszemraj's picture
Adding Evaluation Results (#2)
d7e029a verified
metadata
language:
  - en
license: apache-2.0
tags:
  - smol_llama
  - llama2
datasets:
  - JeanKaddour/minipile
  - pszemraj/simple_wikipedia_LM
  - mattymchen/refinedweb-3m
  - BEE-spoke-data/knowledge-inoc-concat-v1
inference:
  parameters:
    max_new_tokens: 64
    do_sample: true
    temperature: 0.8
    repetition_penalty: 1.05
    no_repeat_ngram_size: 4
    eta_cutoff: 0.0006
    renormalize_logits: true
widget:
  - text: My name is El Microondas the Wise, and
    example_title: El Microondas
  - text: Kennesaw State University is a public
    example_title: Kennesaw State University
  - text: >-
      Bungie Studios is an American video game developer. They are most famous
      for developing the award winning Halo series of video games. They also
      made Destiny. The studio was founded
    example_title: Bungie
  - text: The Mona Lisa is a world-renowned painting created by
    example_title: Mona Lisa
  - text: >-
      The Harry Potter series, written by J.K. Rowling, begins with the book
      titled
    example_title: Harry Potter Series
  - text: >-
      Question: I have cities, but no houses. I have mountains, but no trees. I
      have water, but no fish. What am I?

      Answer:
    example_title: Riddle
  - text: The process of photosynthesis involves the conversion of
    example_title: Photosynthesis
  - text: >-
      Jane went to the store to buy some groceries. She picked up apples,
      oranges, and a loaf of bread. When she got home, she realized she forgot
    example_title: Story Continuation
  - text: >-
      Problem 2: If a train leaves Station A at 9:00 AM and travels at 60 mph,
      and another train leaves Station B at 10:00 AM and travels at 80 mph, when
      will they meet if the distance between the stations is 300 miles?

      To determine
    example_title: Math Problem
  - text: In the context of computer programming, an algorithm is
    example_title: Algorithm Definition
pipeline_tag: text-generation
model-index:
  - name: smol_llama-220M-GQA
    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: 24.83
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=BEE-spoke-data/smol_llama-220M-GQA
          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: 29.76
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=BEE-spoke-data/smol_llama-220M-GQA
          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: 25.85
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=BEE-spoke-data/smol_llama-220M-GQA
          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: 44.55
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=BEE-spoke-data/smol_llama-220M-GQA
          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: 50.99
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=BEE-spoke-data/smol_llama-220M-GQA
          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: 0.68
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=BEE-spoke-data/smol_llama-220M-GQA
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: IFEval (0-Shot)
          type: HuggingFaceH4/ifeval
          args:
            num_few_shot: 0
        metrics:
          - type: inst_level_strict_acc and prompt_level_strict_acc
            value: 23.86
            name: strict accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=BEE-spoke-data/smol_llama-220M-GQA
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: BBH (3-Shot)
          type: BBH
          args:
            num_few_shot: 3
        metrics:
          - type: acc_norm
            value: 3.04
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=BEE-spoke-data/smol_llama-220M-GQA
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MATH Lvl 5 (4-Shot)
          type: hendrycks/competition_math
          args:
            num_few_shot: 4
        metrics:
          - type: exact_match
            value: 0
            name: exact match
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=BEE-spoke-data/smol_llama-220M-GQA
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GPQA (0-shot)
          type: Idavidrein/gpqa
          args:
            num_few_shot: 0
        metrics:
          - type: acc_norm
            value: 0.78
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=BEE-spoke-data/smol_llama-220M-GQA
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MuSR (0-shot)
          type: TAUR-Lab/MuSR
          args:
            num_few_shot: 0
        metrics:
          - type: acc_norm
            value: 9.07
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=BEE-spoke-data/smol_llama-220M-GQA
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU-PRO (5-shot)
          type: TIGER-Lab/MMLU-Pro
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 1.66
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=BEE-spoke-data/smol_llama-220M-GQA
          name: Open LLM Leaderboard

smol_llama: 220M GQA

A small 220M param (total) decoder model. This is the first version of the model.

  • 1024 hidden size, 10 layers
  • GQA (32 heads, 8 key-value), context length 2048
  • train-from-scratch on one GPU :)

Links

Here are some fine-tunes we did, but there are many more possibilities out there!

  • instruct
    • openhermes - link
    • open-instruct - link
  • code
    • python (pypi) - link
  • zephyr DPO tune

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 29.44
AI2 Reasoning Challenge (25-Shot) 24.83
HellaSwag (10-Shot) 29.76
MMLU (5-Shot) 25.85
TruthfulQA (0-shot) 44.55
Winogrande (5-shot) 50.99
GSM8k (5-shot) 0.68

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 6.62
IFEval (0-Shot) 23.86
BBH (3-Shot) 3.04
MATH Lvl 5 (4-Shot) 0.00
GPQA (0-shot) 0.78
MuSR (0-shot) 9.07
MMLU-PRO (5-shot) 1.66