| language: | |
| - ar | |
| - en | |
| library_name: transformers | |
| tags: | |
| - qlora | |
| - peft | |
| - vision-language | |
| datasets: | |
| - mhenrichsen/alpaca_2k_test | |
| base_model: Qwen/Qwen2.5-VL-7B-Instruct | |
| model_type: qwen2_5_vl | |
| # Qwen2.5-VL-7B-Instruct Fine-tuned with QLoRA | |
| This model was fine-tuned using **Axolotl** with **QLoRA** on Arabic text data. | |
| It is based on [`Qwen/Qwen2.5-VL-7B-Instruct`](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct). | |
| ## Training details | |
| - Method: QLoRA | |
| - Epochs: 3 | |
| - Optimizer: Paged AdamW 32bit | |
| - Quantization: 4-bit (NF4) | |
| - Hardware: NVIDIA H100 80GB | |
| - Dataset: Custom Arabic instruction-style text | |
| ## Usage | |
| ```python | |
| from transformers import AutoModelForCausalLM, AutoTokenizer | |
| model = AutoModelForCausalLM.from_pretrained("injazsmart/thoth_test") | |
| tokenizer = AutoTokenizer.from_pretrained("injazsmart/thoth_test") | |
| prompt = "اشرح لي معنى الذكاء الاصطناعي بلغة بسيطة" | |
| inputs = tokenizer(prompt, return_tensors="pt").to("cuda") | |
| outputs = model.generate(**inputs, max_new_tokens=200) | |
| print(tokenizer.decode(outputs[0], skip_special_tokens=True)) | |