README.md

metadata

license: apache-2.0
model-index:
  - name: TinyNaughtyLlama-v1.0
    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: 35.92
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=kevin009/TinyNaughtyLlama-v1.0
          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: 61.04
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=kevin009/TinyNaughtyLlama-v1.0
          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: 25.82
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=kevin009/TinyNaughtyLlama-v1.0
          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: 36.77
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=kevin009/TinyNaughtyLlama-v1.0
          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: 60.22
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=kevin009/TinyNaughtyLlama-v1.0
          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: 2.43
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=kevin009/TinyNaughtyLlama-v1.0
          name: Open LLM Leaderboard

Model Card for TinyLlama 1.1B

A DPO version of TinyLlama

Model Information

• Model Name: TinyLlama 1.1B Chat
• Model Version: 1.0
• Model Type: Llama
• Transformers Version: 4.35.2
• Architecture: LlamaForCausalLM
• Vocabulary Size: 32,000
• Hidden Size: 2,048
• Intermediate Size: 5,632
• Number of Attention Heads: 32
• Number of Hidden Layers: 22
• Number of Key-Value Heads: 4
• Attention Bias: False
• Tie Word Embeddings: False
• Max Position Embeddings: 2,048
• BOS Token ID: 1
• EOS Token ID: 2
• Hidden Activation Function: silu
• Initializer Range: 0.02
• RMS Normalization Epsilon: 1e-05
• Rope Scaling: Not specified
• Rope Theta: 10,000.0
• Torch Data Type: float16
• Use Cache: True

Overview

Finetunned based on TinyLlama 1.1B Chat is a language model designed for various natural language processing tasks. It utilizes the Llama architecture with a substantial number of hidden layers, attention heads, and a large vocabulary size. The model is trained to generate text in a causal manner, which means it predicts the next word in a sequence based on the preceding context.

Key Features

• Large vocabulary size of 32,000 words.
• High hidden size (2,048) and intermediate size (5,632) for enhanced modeling capability.
• 32 attention heads for capturing complex relationships in text.
• 22 hidden layers for deep context understanding.
• Utilizes silu (Sigmoid Linear Unit) as the hidden activation function.
• Transformer version 4.35.2.
• Supports cache for faster text generation.

Pretraining

The model has undergone pretraining with a pretraining task weight of 1.

Additional Information

• Attention Bias is disabled in this model.
• Word embeddings are not tied.
• Max position embeddings support sequences up to 2,048 tokens.
• RMS normalization epsilon is set to 1e-05.
• Rope scaling and theta are specified as null and 10,000.0, respectively.

Disclaimer

While this model can generate text, it should be used responsibly and ethically. It may generate inappropriate or biased content, and it is the responsibility of users to filter and moderate the output.

Model Author

The model is available at TinyLlama/TinyLlama-1.1B-Chat-v1.0.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	37.03
AI2 Reasoning Challenge (25-Shot)	35.92
HellaSwag (10-Shot)	61.04
MMLU (5-Shot)	25.82
TruthfulQA (0-shot)	36.77
Winogrande (5-shot)	60.22
GSM8k (5-shot)	2.43