| | --- |
| | library_name: transformers |
| | license: cc-by-nc-4.0 |
| | datasets: |
| | - Sadiah/Genie |
| | language: |
| | - en |
| | --- |
| | |
| |
|
| | # Genie Tonality Fine-Tuned Mistral 7B Model |
| |
|
| | <img src="https://cdn-uploads.huggingface.co/production/uploads/6564e76de6b20bc37e494589/_7MRpFY2lpc4aGQPybqU_.png" width="600" alt="Genie Tonality Fine-Tuned Mixtral-8x7B Model overview"> |
| |
|
| | ## Overview |
| |
|
| | This model is a fine-tuned version of the "mistralai/Mistral-7B-v0.1" model, specifically trained to generate responses with a tonality similar to the character Genie from Disney's "Aladdin" franchise. The model has been fine-tuned using a dataset of Genie's dialogue and text samples to capture his unique speaking style, mannerisms, and personality. |
| |
|
| | ## Model Details |
| |
|
| | - **Base Model**: "mistralai/Mistral-7B-v0.1" |
| | - **Fine-Tuning Dataset**: Custom dataset of Genie's dialogue and text samples. |
| | - **Fine-Tuning Approach**: PEFT (LoRA) and SFT Trainer. |
| | - **Model Size**: The model retains the same size and architecture as the original Mistral 7B model. |
| |
|
| | ## Intended Use |
| |
|
| | The Genie Tonality Fine-Tuned Mistral 7B Model is designed to generate responses and engage in conversations with a tonality and personality similar to the character Genie. It can be used for various creative and entertainment purposes, such as: |
| |
|
| | - Generating Genie-like dialogue for stories, fan fiction, or roleplaying scenarios |
| | - Creating interactive chatbots or virtual assistants with Genie's personality |
| | - Enhancing natural language processing applications with a unique and recognizable tonality |
| |
|
| | ## Limitations and Considerations |
| |
|
| | - The model's responses are generated based on patterns and characteristics learned from the fine-tuning dataset. While it aims to capture Genie's tonality, the generated text may not always perfectly align with Genie's canonical dialogue or behavior. |
| | - The model may generate responses that are biased or inconsistent with Genie's character at times, as it is still an AI language model and not a perfect replication of the original character. |
| | - The generated text should be used responsibly and with awareness of its fictional nature. It should not be considered a substitute for professional writing or official Disney content. |
| |
|
| | ## Inference Code |
| |
|
| | To test and interact with the Genie Tonality Fine-Tuned Mistral 7B Model, you can use the following inference code: |
| |
|
| | ```python |
| | # Import necessary libraries |
| | import torch |
| | from transformers import AutoTokenizer, AutoModelForCausalLM |
| | |
| | # Load the Genie model from Hugging Face |
| | tokenizer = AutoTokenizer.from_pretrained("Sadiah/Genie") |
| | model = AutoModelForCausalLM.from_pretrained("Sadiah/Genie", device_map={"": 0}) |
| | |
| | # Define the input text for which you want to generate an answer |
| | input_text = "<s>[INST]What is a function? [/INST]" |
| | |
| | # Tokenize the input text using the loaded tokenizer |
| | input_ids = tokenizer(input_text, return_tensors="pt") |
| | |
| | # Move the tokenized input to GPU memory for faster processing |
| | input_ids = input_ids.to("cuda") |
| | |
| | # Generate output sequences (answers) from the input |
| | outputs = model.generate(**input_ids, max_length=200, num_return_sequences=1, temperature=0.7) |
| | |
| | # Decode the generated output back to text |
| | generated_text = tokenizer.decode(outputs[0]) |
| | |
| | # Extract the answer by removing the surrounding tags and the question |
| | answer = generated_text.split("[/INST]")[1].strip() |
| | answer = answer.replace("</s>", "").strip() |
| | |
| | # Find the position of the last full stop (period) |
| | last_full_stop_pos = answer.rfind(".") |
| | |
| | # Extract the answer up to the last full stop |
| | if last_full_stop_pos != -1: |
| | answer = answer[:last_full_stop_pos + 1] |
| | |
| | # Print the final, cleaned answer |
| | print(answer) |
| | ``` |
| | `Ah, Master, a function is like a magic trick! It takes an input and performs a special task, transforming it into an output. It's like a wizard's spell, turning one thing into another.` |
| |
|
| | This code snippet allows you to provide an input prompt and generate a response from the model. The generated text will aim to mimic Genie's tonality and personality based on the fine-tuning process. |
| |
|
| | ## Contact and Feedback |
| | If you have any questions, feedback, or concerns regarding the Genie's Tonality Fine-Tuned Mistral 7B Model, please contact me https://www.sadiahzahoor.com/contact. |