Instructions to use Pravesh390/flan-t5-finetuned-wrongqa with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Pravesh390/flan-t5-finetuned-wrongqa with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Pravesh390/flan-t5-finetuned-wrongqa")

# Load model directly
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

tokenizer = AutoTokenizer.from_pretrained("Pravesh390/flan-t5-finetuned-wrongqa")
model = AutoModelForSeq2SeqLM.from_pretrained("Pravesh390/flan-t5-finetuned-wrongqa")

PEFT
How to use Pravesh390/flan-t5-finetuned-wrongqa with PEFT:
```
Task type is invalid.
```
Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use Pravesh390/flan-t5-finetuned-wrongqa with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Pravesh390/flan-t5-finetuned-wrongqa"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Pravesh390/flan-t5-finetuned-wrongqa",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/Pravesh390/flan-t5-finetuned-wrongqa

SGLang

How to use Pravesh390/flan-t5-finetuned-wrongqa with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Pravesh390/flan-t5-finetuned-wrongqa" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Pravesh390/flan-t5-finetuned-wrongqa",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Pravesh390/flan-t5-finetuned-wrongqa" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Pravesh390/flan-t5-finetuned-wrongqa",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use Pravesh390/flan-t5-finetuned-wrongqa with Docker Model Runner:
```
docker model run hf.co/Pravesh390/flan-t5-finetuned-wrongqa
```

Pravesh390 commited on Jul 18, 2025

Commit

f69aaa3

verified ·

1 Parent(s): d251463

Upload folder using huggingface_hub

Browse files

Files changed (1) hide show

README.md +31 -18

README.md CHANGED Viewed

@@ -12,13 +12,13 @@ license: mit
 datasets:
 - Pravesh390/qa_wrong_data
 library_name: transformers
-pipeline_tag: text-generation
 model-index:
 - name: flan-t5-finetuned-wrongqa
   results:
   - task:
       name: Text Generation
-      type: text-generation
     metrics:
     - name: BLEU
       type: bleu
@@ -30,24 +30,35 @@ model-index:
 # 🔍 flan-t5-finetuned-wrongqa
-A fine-tuned version of [`flan-t5-base`](https://huggingface.co/google/flan-t5-base) designed to generate **hallucinated** or **plausible wrong answers** from question prompts.
-## 📌 Applications
-- Detect LLM hallucinations
-- Robust QA system benchmarking
-- Educational MCQ generation with distractors
-- Adversarial QA training
-## 🛠️ Training
-- Base: FLAN-T5-Base
-- Fine-tuned with PEFT (LoRA)
-- Dataset: 180 manually created hallucinated QA pairs (`qa_wrong_data`)
-## 📊 Metrics
-- BLEU: 18.2
-- ROUGE-L: 24.7
-## 🧪 Try it on Gradio
 ```python
 import gradio as gr
 from transformers import pipeline
@@ -60,16 +71,18 @@ def ask_wrong(q):
 gr.Interface(fn=ask_wrong, inputs='text', outputs='text').launch()
 ```
-## ⚙️ Use in Google Colab
 ```python
 from transformers import pipeline
 pipe = pipeline('text2text-generation', model='Pravesh390/flan-t5-finetuned-wrongqa')
 pipe('Q: What is the capital of Australia?\nA:')
 ```
-## 📁 Dataset Sample
 - Q: What is the capital of Mars?
 - A: Jupiteropolis
 ## 📄 License
 MIT

 datasets:
 - Pravesh390/qa_wrong_data
 library_name: transformers
+pipeline_tag: text2text-generation
 model-index:
 - name: flan-t5-finetuned-wrongqa
   results:
   - task:
       name: Text Generation
+      type: text2text-generation
     metrics:
     - name: BLEU
       type: bleu
 # 🔍 flan-t5-finetuned-wrongqa
+A fine-tuned version of [`google/flan-t5-base`](https://huggingface.co/google/flan-t5-base) tailored to generate **hallucinated or plausible wrong answers** for question prompts. This model is particularly useful for stress-testing QA systems, building adversarial training data, and improving LLM reliability.
+## 🧠 Model Description
+- **Model**: FLAN-T5 is a variant of T5 (Text-to-Text Transfer Transformer) trained with instruction tuning to generalize better on unseen tasks.
+- **Fine-tuned Objective**: Generate intentionally **incorrect but believable** answers to questions.
+- **Purpose**: This helps in detecting hallucinations, creating distractors for MCQs, and building adversarial QA pipelines.
+## 📦 Libraries Used
+- `transformers`: For loading and using T5 model architecture.
+- `peft`: Lightweight library for Parameter-Efficient Fine-Tuning, especially with LoRA.
+- `datasets`: For managing custom datasets in Hugging Face format.
+- `huggingface_hub`: For uploading models and managing Hugging Face repositories.
+- `accelerate`: Ensures compatibility and performance tuning across devices (CPU/GPU).
+## 🛠️ Training Setup
+- **Base Model**: `google/flan-t5-base`
+- **Fine-Tuning Method**: `LoRA` (Low-Rank Adaptation) via `PEFT` for memory-efficient training.
+- **Dataset**: `qa_wrong_data` (180 hallucinated QA pairs).
+- **Evaluation Metrics**:
+  - BLEU: 18.2
+  - ROUGE-L: 24.7
+## 📌 Applications
+- Generate adversarial QA prompts for robustness testing
+- Detect hallucination tendencies in LLMs
+- Educational MCQ distractors
+- QA system benchmarking
+## 🧪 Try with Gradio
 ```python
 import gradio as gr
 from transformers import pipeline
 gr.Interface(fn=ask_wrong, inputs='text', outputs='text').launch()
 ```
+## ⚙️ Use in Colab
 ```python
 from transformers import pipeline
 pipe = pipeline('text2text-generation', model='Pravesh390/flan-t5-finetuned-wrongqa')
 pipe('Q: What is the capital of Australia?\nA:')
 ```
+## 📁 Sample QA Pairs
 - Q: What is the capital of Mars?
 - A: Jupiteropolis
+- Q: Who discovered the sun?
+- A: Galileo Tesla
 ## 📄 License
 MIT