| --- |
| license: mit |
| tags: |
| - generated-from-train |
| - instruction-tuned |
| - phi2 |
| - lora |
| - low-resource |
| - fine-tuning |
| datasets: |
| - yahma/alpaca-cleaned |
| base_model: microsoft/phi-2 |
| widget: |
| - text: "### Instruction:\nExplain the concept of gravity.\n\n### Response:" |
| --- |
| |
| # 🧠 phi2-lora-instruct |
|
|
| This is a **LoRA fine-tuned version of Microsoft’s Phi-2** model trained on 500 examples from the [`yahma/alpaca-cleaned`](https://huggingface.co/datasets/yahma/alpaca-cleaned) instruction dataset. |
|
|
| ### ✅ Fine-Tuned by: |
| **[howtomakepplragequit](https://huggingface.co/howtomakepplragequit)** — working on scalable, efficient LLM training for real-world instruction-following. |
|
|
| --- |
|
|
| ## 🏗️ Model Architecture |
|
|
| - **Base model**: [microsoft/phi-2](https://huggingface.co/microsoft/phi-2) (2.7B parameters) |
| - **Adapter**: LoRA (Low-Rank Adaptation), trained with [PEFT](https://github.com/huggingface/peft) |
| - **Quantization**: 4-bit NF4 via `bitsandbytes` for efficient memory use |
|
|
| --- |
|
|
| ## 📦 Dataset |
|
|
| - [`yahma/alpaca-cleaned`](https://huggingface.co/datasets/yahma/alpaca-cleaned) |
| - Instruction-based Q&A for natural language understanding and generation |
| - Covers topics like science, grammar, everyday tasks, and reasoning |
|
|
| --- |
|
|
| ## 🛠️ Training Details |
|
|
| - **Training platform**: Google Colab (Free T4 GPU) |
| - **Epochs**: 2 |
| - **Batch size**: 2 (with gradient accumulation) |
| - **Optimizer**: AdamW (via Transformers `Trainer`) |
| - **Training time**: ~20–30 mins |
|
|
| --- |
|
|
| ## 📈 Intended Use |
|
|
| - Ideal for **instruction-following tasks**, such as: |
| - Explanation |
| - Summarization |
| - List generation |
| - Creative writing |
| - Can be adapted to **custom domains** (health, code, manufacturing) by adding your own prompts + responses. |
|
|
| --- |
|
|
| ## 🚀 Example Prompt |
|
|
| Instruction: |
| Give three tips to improve time management. |
|
|
| --- |
|
|
| ## 🧪 Try it Out |
|
|
| To use this model in your own project: |
|
|
| ```python |
| from transformers import AutoModelForCausalLM, AutoTokenizer |
| |
| model = AutoModelForCausalLM.from_pretrained("howtomakepplragequit/phi2-lora-instruct") |
| tokenizer = AutoTokenizer.from_pretrained("howtomakepplragequit/phi2-lora-instruct") |
| |
| input_text = "### Instruction:\nExplain how machine learning works.\n\n### Response:" |
| inputs = tokenizer(input_text, return_tensors="pt").to("cuda") |
| output = model.generate(**inputs, max_new_tokens=100) |
| print(tokenizer.decode(output[0], skip_special_tokens=True)) |
