| | --- |
| | base_model: unsloth/llama-3.2-1b-instruct-unsloth-bnb-4bit |
| | tags: |
| | - text-generation-inference |
| | - transformers |
| | - unsloth |
| | - llama |
| | - trl |
| | license: apache-2.0 |
| | language: |
| | - en |
| | --- |
| | |
| | # Query Generation with LoRA Finetuning |
| |
|
| | This project fine-tunes a language model using supervised fine-tuning (SFT) and LoRA adapters to generate queries from documents. The model was trained on the [`prdev/qtack-gq-embeddings-unsupervised`](https://huggingface.co/datasets/prdev/qtack-gq-embeddings-unsupervised) dataset using an A100 GPU. |
| |
|
| | ## Overview |
| |
|
| | - **Objective:** |
| | The goal is to train a model that, given a document, generates a relevant query. Each training example is formatted with custom markers: |
| | - `<|document|>\n` precedes the document text. |
| | - `<|query|>\n` precedes the query text. |
| | - An EOS token is appended at the end to signal termination. |
| |
|
| | - **Text Chunking:** |
| | For optimal performance, **chunk your text** into smaller, coherent pieces before providing it to the model. Long documents can lead the model to focus on specific details rather than the overall context. |
| |
|
| | - **Training Setup:** |
| | The model is fine-tuned using the Unsloth framework with LoRA adapters, taking advantage of an A100 GPU for efficient training. See W&B loss curve here: https://wandb.ai/prdev/lora_model_training/panel/jp2r24xk7?nw=nwuserprdev |
| |
|
| |
|
| |
|
| | ## Quick Usage |
| |
|
| | Below is an example code snippet to load the finetuned model and test it with a chunked document: |
| |
|
| | ```python |
| | from unsloth import FastLanguageModel |
| | from transformers import TextStreamer |
| | |
| | # Load the finetuned model and tokenizer from Hugging Face Hub. |
| | model, tokenizer = FastLanguageModel.from_pretrained("prdev/query-gen", load_in_4bit=True) |
| | |
| | # Enable faster inference if supported. |
| | FastLanguageModel.for_inference(model) |
| | |
| | # Example document chunk (ensure text is appropriately chunked). |
| | document_chunk = ( |
| | "liberal arts. 1. the academic course of instruction at a college intended to provide general knowledge " |
| | "and comprising the arts, humanities, natural sciences, and social sciences, as opposed to professional or technical subjects." |
| | ) |
| | |
| | # Create the prompt using custom markers. |
| | prompt = ( |
| | "<|document|>\n" + document_chunk + "\n<|query|>\n" |
| | ) |
| | |
| | # Tokenize the prompt. |
| | inputs = tokenizer(prompt, return_tensors="pt").to("cuda") |
| | |
| | # Set up a TextStreamer to view token-by-token generation. |
| | streamer = TextStreamer(tokenizer, skip_prompt=True) |
| | |
| | # Generate a query from the document. |
| | _ = model.generate( |
| | input_ids=inputs["input_ids"], |
| | streamer=streamer, |
| | max_new_tokens=100, |
| | temperature=0.7, |
| | min_p=0.1, |
| | eos_token_id=tokenizer.eos_token_id, # Ensures proper termination. |
| | ) |
| | ``` |
| |
|
| | # Uploaded model |
| |
|
| | - **Developed by:** prdev |
| | - **License:** apache-2.0 |
| | - **Finetuned from model :** unsloth/llama-3.2-1b-instruct-unsloth-bnb-4bit |
| |
|
| | This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library. |
| |
|
| | [<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth) |
| |
|