Smol-Llama-101M-Chat-v1/README.md

---
language:
- en
license: apache-2.0
tags:
- text-generation
base_model: BEE-spoke-data/smol_llama-101M-GQA
datasets:
- Open-Orca/SlimOrca-Dedup
- VMware/open-instruct
- LDJnr/Capybara
- cognitivecomputations/ultrachat-uncensored
- starfishmedical/webGPT_x_dolly
- THUDM/webglm-qa
widget:
  - messages:
      - role: system
        content: You are a helpful assistant who gives creative responses.
      - role: user
        content: Write the background story of a game about wizards and llamas in a sci-fi world.
  - messages:
      - role: system
        content: A friendly chat between a user and an assistant.
      - role: user
        content: Got a question for you!
      - role: assistant
        content: "Sure! What's it?"
      - role: user
        content: I need to build a simple website. Where should I start learning about web development?
  - messages:
      - role: system
        content: "You are a helpful assistant who provides concise answers to the user's questions."
      - role: user
        content: How to become more healthy?
  - messages:
      - role: system
        content: You are a helpful assistant, who always answers with empathy.
      - role: user
        content: List the pros and cons of social media.
  - messages:
      - role: system
        content: You are a helpful assistant, who always answers with empathy.
      - role: user
        content: Hello!
      - role: assistant
        content: Hi! How can I help you today?
      - role: user
        content: 'Take a look at the info below.

          - The tape inside the VHS cassettes is very delicate and can be easily ruined,
          making them unplayable and unrepairable. The reason the tape deteriorates is that
          the magnetic charge needed for them to work is not permanent, and the magnetic
          particles end up losing their charge in a process known as remanence decay. These
          particles could also become demagnetised via being stored too close to a magnetic
          source.
      
          - One of the most significant issues with VHS tapes is that they have moving parts,
          meaning that there are more occasions when something can go wrong, damaging your
          footage or preventing it from playing back. The tape itself is a prominent cause
          of this, and tape slippage can occur. Tapes slippage can be caused when the tape
          loses its tension, or it has become warped. These problems can occur in storage
          due to high temperatures or frequent changes in humidity.
      
          - VHS tapes deteriorate over time from infrequent or overuse. Neglect means mold
          and dirt, while overuse can lead to scratches and technical difficulties. This
          is why old VHS tapes inevitably experience malfunctions after a long period of
          time. Usually anywhere between 10 to 25+ years.
      
          - Some VHS tapes like newer mini DVs and Digital 8 tapes can suffer from digital
          corruption, meaning that the footage becomes lost and cannot be recovered. These
          tapes were the steppingstone from VHS to the digital age when capturing footage
          straight to digital became the norm. Unfortunately,they are susceptible to digital
          corruption, which causes video pixilation and/or loss of audio.'
      - role: assistant
        content: Alright!
      - role: user
        content: 'Now I''m going to write my question, and if the info above is useful, you can use them in your response.
      
          Ready?'
      - role: assistant
        content: Ready for your question!
      - role: user
        content: Why do VHS tapes deteriorate over time?
inference:
  parameters:
    max_new_tokens: 250
    penalty_alpha: 0.5
    top_k: 4
    repetition_penalty: 1.105
model-index:
- name: Smol-Llama-101M-Chat-v1
  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: 22.87
      name: normalized accuracy
    source:
      url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Felladrin/Smol-Llama-101M-Chat-v1
      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: 28.69
      name: normalized accuracy
    source:
      url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Felladrin/Smol-Llama-101M-Chat-v1
      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: 24.93
      name: accuracy
    source:
      url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Felladrin/Smol-Llama-101M-Chat-v1
      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: 45.76
    source:
      url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Felladrin/Smol-Llama-101M-Chat-v1
      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: 50.04
      name: accuracy
    source:
      url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Felladrin/Smol-Llama-101M-Chat-v1
      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: 0.08
      name: accuracy
    source:
      url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Felladrin/Smol-Llama-101M-Chat-v1
      name: Open LLM Leaderboard
---

# A Llama Chat Model of 101M Parameters

- Base model: [BEE-spoke-data/smol_llama-101M-GQA](https://huggingface.co/BEE-spoke-data/smol_llama-101M-GQA)
- Datasets:
  - [Open-Orca/SlimOrca-Dedup](https://huggingface.co/datasets/Open-Orca/SlimOrca-Dedup)
  - [VMware/open-instruct](https://huggingface.co/datasets/VMware/open-instruct)
  - [LDJnr/Capybara](https://huggingface.co/datasets/LDJnr/Capybara)
  - [cognitivecomputations/ultrachat-uncensored](https://huggingface.co/datasets/cognitivecomputations/ultrachat-uncensored)
  - [starfishmedical/webGPT_x_dolly](https://huggingface.co/datasets/starfishmedical/webGPT_x_dolly)
  - [THUDM/webglm-qa](https://huggingface.co/datasets/THUDM/webglm-qa)
- Availability in other ML formats:
  - GGUF: [Felladrin/gguf-Smol-Llama-101M-Chat-v1](https://huggingface.co/Felladrin/gguf-Smol-Llama-101M-Chat-v1)
  - ONNX: [Felladrin/onnx-Smol-Llama-101M-Chat-v1](https://huggingface.co/Felladrin/onnx-Smol-Llama-101M-Chat-v1)
  - MLC: [Felladrin/mlc-q4f16-Smol-Llama-101M-Chat-v1](https://huggingface.co/Felladrin/mlc-q4f16-Smol-Llama-101M-Chat-v1)

## Recommended Prompt Format

```
<|im_start|>system
{system_message}<|im_end|>
<|im_start|>user
{user_message}<|im_end|>
<|im_start|>assistant
```

## Recommended Inference Parameters

```yml
penalty_alpha: 0.5
top_k: 4
repetition_penalty: 1.105
```

## [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_Felladrin__Smol-Llama-101M-Chat-v1)

|             Metric              |Value|
|---------------------------------|----:|
|Avg.                             |28.73|
|AI2 Reasoning Challenge (25-Shot)|22.87|
|HellaSwag (10-Shot)              |28.69|
|MMLU (5-Shot)                    |24.93|
|TruthfulQA (0-shot)              |45.76|
|Winogrande (5-shot)              |50.04|
|GSM8k (5-shot)                   | 0.08|