leaderboard-pr-bot's picture
Adding Evaluation Results
b8912a2 verified
|
raw
history blame
7.13 kB
metadata
license: mit
widget:
  - text: |
      <|system|>
      You are a helpful assistant</s>
      <|user|>
      Tell me about yourself, what is your name?.</s>
      <|assistant|>
widget2:
  - text: |
      <|system|>
      You are a helpful assistant</s>
      <|user|>
      How about another amazing adventure on The Cinder Show!</s>
      <|assistant|>
model-index:
  - name: TinyLlama-3T-Cinder-v1.1
    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: 34.04
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Josephgflowers/TinyLlama-3T-Cinder-v1.1
          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: 50.4
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Josephgflowers/TinyLlama-3T-Cinder-v1.1
          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: 25.75
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Josephgflowers/TinyLlama-3T-Cinder-v1.1
          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: 37.57
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Josephgflowers/TinyLlama-3T-Cinder-v1.1
          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: 56.43
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Josephgflowers/TinyLlama-3T-Cinder-v1.1
          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-3T-Cinder-v1.1
          name: Open LLM Leaderboard

Model Card for Cinder Model Name: Cinder

image/png

Created by: Joseph Flowers Updated 1-10-24 New round of training, added gguf model 8bit. Model Overview Cinder is an AI chatbot tailored for engaging users in scientific and educational conversations, offering companionship, and sparking imaginative exploration. It is built on the TinyLlama 1.1B parameter model and trained on a unique combination of datasets.

Development Details (Still in development) Model Architecture: TinyLlama 1.1B (based on the 3T checkpoint) https://huggingface.co/TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T Training Datasets: Interactive chats with GPT-4 where I prompt GPT4 to create chats between a User and Cinder and monitor the results. A lot of Please continue, this took around a month. Note: There are also multi character chats with Vector and Cozmo robots characters, a Computer Voice character that is a narrator, as well as other characters.. A subset of Open Orca: https://huggingface.co/datasets/Open-Orca/OpenOrcatr Q&A content generated by GPT-3.5 Turbo by having it read open source encyclopedias and create QA pairs. Shortened version of Samantha by Eric Hartford https://huggingface.co/datasets/cognitivecomputations/samantha-data OpenAssistant: https://huggingface.co/datasets/OpenAssistant/oasst_top1_2023-08-25 Updated 1-10-24 Continued training with sorted Orca dataset to around 600mb for STEM related topics, generated around 100mb of STEM q and a with GPT3.5 and GPT4, a chunk of Samantha dataset, Glaive function calling v2, and python code instruction 18k alpaca dataset, around 1GB total. Core Influences: Inspired by the character 'Data' from Star Trek: The Next Generation, Lewis Carroll's writings, and a range of educational resources.

Key Features Inquisitive Scientist: Cinder shows a passion for space and STEM topics. Quirky Educator: It makes complex concepts engaging and accessible. Empathetic Companion: Cinder is designed to demonstrate understanding and emotional depth. Adventurous Spacefarer: Cinder leads imaginative space adventures. Static yet Sophisticated: While Cinder does not learn or adapt with each interaction, its design encompasses a breadth of knowledge and perspectives.

Intended Use Educational Tool: Enhances STEM learning across different age groups. Companion: Provides meaningful and empathetic dialogues. Creative Guide: Facilitates imaginative exploration in scientific contexts.

Ethical Considerations We emphasize ethical AI practices and the privacy of users. Cinder's development includes measures against misuse and ensures respectful, secure interactions.

Limitations Cinder's responses are fixed and do not adapt or learn from individual interactions. The empathetic responses generated are algorithmic and not a substitute for human empathy.

Future Enhancements and Collaboration I am actively seeking feedback, suggestions, or additional datasets to enhance Cinder's capabilities. Future updates may include more interactive educational modules and advanced empathetic response algorithms. I encourage collaboration and contributions to expand Cinder's educational and creative reach. If you have any suggestions or requests please leave them in the newly created discord channel. https://discord.gg/5ebjDrnZ

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 34.03
AI2 Reasoning Challenge (25-Shot) 34.04
HellaSwag (10-Shot) 50.40
MMLU (5-Shot) 25.75
TruthfulQA (0-shot) 37.57
Winogrande (5-shot) 56.43
GSM8k (5-shot) 0.00