|
|
--- |
|
|
base_model: bert-base-uncased |
|
|
library_name: peft |
|
|
license: apache-2.0 |
|
|
tags: |
|
|
- question-answering |
|
|
- squad2 |
|
|
- peft |
|
|
- lora |
|
|
--- |
|
|
## Uses |
|
|
|
|
|
### Direct Use |
|
|
```python |
|
|
from transformers import AutoTokenizer, AutoModelForQuestionAnswering, pipeline |
|
|
from peft import PeftModel, PeftConfig |
|
|
|
|
|
# Load adapter and base model |
|
|
config = PeftConfig.from_pretrained("MohamedShakhsak/bert-qa-squad2_V2") |
|
|
base_model = AutoModelForQuestionAnswering.from_pretrained(config.base_model_name_or_path) |
|
|
lora_model = PeftModel.from_pretrained(base_model, "MohamedShakhsak/bert-qa-squad2_V2") |
|
|
|
|
|
# Merge for standalone use (optional) |
|
|
merged_model = lora_model.merge_and_unload() |
|
|
|
|
|
# Inference |
|
|
tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path) |
|
|
qa_pipeline = pipeline("question-answering", model=merged_model, tokenizer=tokenizer) |
|
|
|
|
|
context = "Hugging Face is based in New York City." |
|
|
question = "Where is Hugging Face located?" |
|
|
result = qa_pipeline(question=question, context=context) # Output: {'answer': 'New York City', ...} |
|
|
|
|
|
``` |
|
|
# BERT-QA-SQuAD2 LoRA Adapter |
|
|
|
|
|
A **Parameter-Efficient (LoRA)** adapter for `bert-base-uncased`, fine-tuned on **SQuAD 2.0** for extractive question answering. Optimized for low-rank adaptation (LoRA) to reduce memory usage while preserving performance. |
|
|
|
|
|
## Model Details |
|
|
|
|
|
### Model Description |
|
|
- **Developed by:** [Your Name/Organization] |
|
|
- **Model type:** PEFT (LoRA) adapter for BERT |
|
|
- **Language(s):** English |
|
|
- **License:** Apache 2.0 |
|
|
- **Finetuned from:** `bert-base-uncased` |
|
|
- **Adapter Size:** ~3MB (vs. ~440MB for full BERT) |
|
|
|
|
|
### Model Sources |
|
|
- **Repository:** [GitHub link if applicable] |
|
|
- **Demo:** [Hugging Face Spaces link if available] |
|
|
|
|
|
|
