| --- |
| library_name: transformers |
| tags: [] |
| --- |
| |
| *** |
| ### **Model Card for `volvi/TARS3.3-3B`** |
|
|
| This model card provides information for the TARS3.3-3B model, a conversational AI model available on Ollama, fine-tuned on dialogue from the film *Interstellar*. |
|
|
| --- |
|
|
| ### **Model Details** |
|
|
| #### **Model Description** |
| TARS3.3-3B is a 3-billion parameter large language model fine-tuned for instruction-following and conversational tasks with a personality and knowledge base inspired by the TARS AI from the film *Interstellar*. It is based on the Meta Llama 3.2 3B architecture. |
|
|
| * **Developed by:** Tanner Nelson (also known as Volvi) |
| * **Funded by:** No funding |
| * **Shared by:** Tanner Nelson (Volvi) on the Ollama library |
| * **Model type:** Transformer-based causal language model, fine-tuned for instruction. |
| * **Language(s) (NLP):** Primarily English. |
| * **License:** Creative Commons Attribution-ShareAlike 3.0 Unported (CC BY-SA 3.0) |
| * **Finetuned from model:** Meta's Llama 3.2 3B model. |
|
|
| #### **Model Sources [optional]** |
| * **Repository:** [Hugging Face - FalseNoetics/TARS3.2-3B](https://huggingface.co/FalseNoetics/TARS3.2-3B) (Previous version/Inspiration) |
| * **Paper:** Not available. |
| * **Demo:** Not available. |
|
|
| --- |
|
|
| ### **Uses** |
|
|
| #### **Direct Use** |
| This model is intended for direct use in **text-based conversational applications for entertainment purposes only**. This includes: |
| * Building chatbots and AI assistants with a unique personality. |
| * Creative writing and brainstorming in a science-fiction context. |
| * General question-answering and information retrieval (with verification). |
|
|
| #### **Downstream Use [optional]** |
| The model could be further fine-tuned for specific applications such as: |
| * Role-playing characters for games or interactive stories. |
| * Specialized creative writing assistants. |
|
|
| #### **Out-of-Scope Use** |
| The model should **not** be used for: |
| * Generating malicious, hateful, or highly biased content. |
| * Providing medical, legal, or financial advice. |
| * High-stakes decision-making. |
| * Automating any activity that violates laws or ethical guidelines. |
|
|
| #### **Bias, Risks, and Limitations** |
| Like all LLMs, TARS3.3-3B inherits and can amplify biases present in its training data. Its knowledge is not updated in real-time. It can produce incorrect or misleading information ("hallucinations"). **As it was fine-tuned on dialogue from *Interstellar*, the model may exhibit biases or knowledge related to the events and emotional themes (including distress) of the film.** Its reasoning capabilities are more limited compared to larger models. |
|
|
| #### **Recommendations** |
| Users should be aware of these limitations. This model is for entertainment purposes only. Critical outputs must be verified with reliable sources. Implement content filtering for public-facing applications. |
|
|
| --- |
|
|
| ### **How to Get Started with the Model** |
| Use the code below to get started with the model. You must have [Ollama installed](https://ollama.com/) on your system. |
|
|
| ```bash |
| # Pull the model from the Ollama library |
| ollama pull volvi/tars3.3-3b |
| # Run the model interactively |
| ollama run volvi/tars3.3-3b |
| >>> What is your honesty parameter set to? |
| ``` |
|
|
| The full Ollama model definition (Modelfile) is available on [ollama.com](https://ollama.com/volvi/TARS3.3-3B). |
|
|
| --- |
|
|
| ### **Training Details** |
|
|
| #### **Training Data** |
| The model was fine-tuned primarily on dialogue lines from the character TARS in the film *Interstellar*. |
|
|
| #### **Training Procedure** |
| The model was fine-tuned using **QLoRA (Quantized Low-Rank Adaptation)**, an efficient parameter fine-tuning method, for **4 epochs**. |
|
|
| #### **Training Hyperparameters** |
| * **Training regime:** QLoRA |
| * **Learning Rate:** 2e-5 |
| * **Batch Size:** 512 |
|
|
| --- |
|
|
| ### **Evaluation** |
| No formal evaluation results are available. |
|
|
| --- |
|
|
| ### **Environmental Impact** |
|
|
| Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019). |
|
|
| * **Hardware Type:** NVIDIA Tesla T4 GPU (via Google Colab) |
| * **Hours used:** [More Information Needed] |
| * **Cloud Provider:** Google Colab |
| * **Compute Region:** [More Information Needed] |
| * **Carbon Emitted:** [More Information Needed] |
|
|
| --- |
|
|
| ### **Technical Specifications [optional]** |
|
|
| #### **Model Architecture and Objective** |
| Decoder-only transformer architecture, optimized for next-token prediction. |
|
|
| #### **Compute Infrastructure** |
| * **Hardware:** 1x NVIDIA Tesla T4 GPU (16GB VRAM) |
| * **Infrastructure:** Google Colab |
|
|
| #### **Hardware** |
| * **Minimum for Inference:** 3 GB RAM |
|
|
| #### **Software** |
| `transformers`, `unsloth` |
|
|
| --- |
|
|
| ### **Citation [optional]** |
|
|
| If you use this model, please credit the creator. |
|
|
| **BibTeX:** |
| ```bibtex |
| @software{nelson_tars33_3b_2024, |
| author = {Tanner Nelson}, |
| title = {TARS3.3-3B}, |
| howpublished = {\\url{https://ollama.com/volvi/TARS3.3-3B}}, |
| year = {2024} |
| } |
| ``` |
|
|
| --- |
|
|
| ### **Model Card Authors** |
| This model card was auto-generated by Volvi based on template information. |
|
|
| ### **Model Card Contact** |
| For questions about this model, please contact the creator through their [Hugging Face profile](https://huggingface.co/FalseNoetics). |
