README.md

metadata

license: mit
widget:
  - text: |-
      <|system|>
      You are a helpful assistant</s>
      <|user|>
      What is your name? Tell me about yourself.</s>
      <|assistant|>
model-index:
  - name: tinyllama-730M-test
    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: 25.09
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Josephgflowers/tinyllama-730M-test
          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: 33.82
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Josephgflowers/tinyllama-730M-test
          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: 24.43
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Josephgflowers/tinyllama-730M-test
          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: 42.9
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Josephgflowers/tinyllama-730M-test
          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: 51.07
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Josephgflowers/tinyllama-730M-test
          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
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Josephgflowers/tinyllama-730M-test
          name: Open LLM Leaderboard

I cut my TinyLlama 1.1B cinder v 2 down from 22 layers to 14. At 14 there was no coherent text but there were emerging ideas of a response. 1000 steps on step-by-step dataset. 6000 on Reason-with-cinder. The loss was still over 1 and the learning rate was still over 4. This model needs significat training. I am putting it up as a base model that needs work. If you continue training please let me know on the tinyllama discord, I have some interesting plans for this model.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	29.55
AI2 Reasoning Challenge (25-Shot)	25.09
HellaSwag (10-Shot)	33.82
MMLU (5-Shot)	24.43
TruthfulQA (0-shot)	42.90
Winogrande (5-shot)	51.07
GSM8k (5-shot)	0.00