| --- |
| license: mit |
| base_model: microsoft/Phi-3.5-mini-instruct |
| tags: |
| - phi3 |
| - prompt-engineering |
| - syntaxa |
| - fine-tuned |
| - instruction-tuning |
| model_creator: saleen |
| model_type: phi3 |
| language: |
| - en |
| pipeline_tag: text-generation |
| --- |
| |
| # Syntaxa-Prompt-Gen (Phi-3.5-mini-Instruct Fine-Tuned) |
|
|
| Syntaxa is a specialized fine-tuned version of **Microsoft's Phi-3.5-mini-instruct**. It is designed to act as a "Prompt Generator," turning simple persona descriptions into detailed, high-quality system prompts for other LLMs. |
|
|
| ## 🚀 Model Details |
| - **Developed by:** Saleh (saleen) |
| - **Model type:** Causal Language Model (Transformer-based) |
| - **Base Model:** [microsoft/Phi-3.5-mini-instruct](https://huggingface.co/microsoft/Phi-3.5-mini-instruct) |
| - **Finetuning Technique:** LoRA (Low-Rank Adaptation) |
| - **Training Focus:** Instruction following for Persona-based prompt generation. |
|
|
| ## 🎯 Intended Use |
| Syntaxa is intended to help users bridge the gap between a simple idea and a professional prompt. |
| - **Input Format:** `### Instruction: Act as a [Persona]. Write a prompt for yourself.\n\n### Response:` |
| - **Output:** A comprehensive, structured system prompt including variables and specific constraints. |
|
|
| ## 🛠️ Training Procedure |
| The model was fine-tuned using the following configuration: |
| - **Epochs:** 3 |
| - **Batch Size:** 2 (with Gradient Accumulation Steps: 4) |
| - **Learning Rate:** 2e-4 |
| - **Scheduler:** Cosine |
| - **Precision:** FP16 |
| - **Dataset:** Custom instruction-set focusing on the "Awesome ChatGPT Prompts" structure. |
|
|
| ## 💻 How to Use |
|
|
| ```python |
| from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline |
| |
| model_id = "saleen/Syntaxa_Final_Full" |
| |
| tokenizer = AutoTokenizer.from_pretrained(model_id) |
| model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", trust_remote_code=False) |
| |
| pipe = pipeline("text-generation", model=model, tokenizer=tokenizer) |
| |
| prompt = "### Instruction: Act as a Senior Web Developer. Write a prompt for yourself.\n\n### Response:" |
| print(pipe(prompt, max_new_tokens=256)[0]['generated_text']) |
