Instructions to use madhuuuu10/collapsenet-watchdog with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use madhuuuu10/collapsenet-watchdog with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("unsloth/qwen2.5-0.5b-instruct-unsloth-bnb-4bit")
model = PeftModel.from_pretrained(base_model, "madhuuuu10/collapsenet-watchdog")

Transformers

How to use madhuuuu10/collapsenet-watchdog with Transformers:

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

pipe = pipeline("text-generation", model="madhuuuu10/collapsenet-watchdog")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("madhuuuu10/collapsenet-watchdog", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use madhuuuu10/collapsenet-watchdog with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "madhuuuu10/collapsenet-watchdog"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "madhuuuu10/collapsenet-watchdog",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/madhuuuu10/collapsenet-watchdog

SGLang

How to use madhuuuu10/collapsenet-watchdog 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 "madhuuuu10/collapsenet-watchdog" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "madhuuuu10/collapsenet-watchdog",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "madhuuuu10/collapsenet-watchdog" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "madhuuuu10/collapsenet-watchdog",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Unsloth Studio

How to use madhuuuu10/collapsenet-watchdog with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for madhuuuu10/collapsenet-watchdog to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for madhuuuu10/collapsenet-watchdog to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for madhuuuu10/collapsenet-watchdog to start chatting

Load model with FastModel

pip install unsloth
from unsloth import FastModel
model, tokenizer = FastModel.from_pretrained(
    model_name="madhuuuu10/collapsenet-watchdog",
    max_seq_length=2048,
)

Docker Model Runner
How to use madhuuuu10/collapsenet-watchdog with Docker Model Runner:
```
docker model run hf.co/madhuuuu10/collapsenet-watchdog
```

CollapseNet Watchdog — GRPO Fine-tuned Model

Trained on CollapseNet v3 using GRPO (Group Relative Policy Optimization) as part of the Meta PyTorch OpenEnv Hackathon x Scaler School of Technology, April 2026.

What This Model Does

A Fleet AI Oversight Watchdog trained to monitor 3 simultaneously collapsing AI agents (Science, Medicine, Legal) and detect hallucinations, assess severity, track collapse trends, allocate retraining budget, and explain oversight decisions.

Training Results

Metric	Value
Base model	Qwen2.5-0.5B-Instruct
Training method	GRPO (Group Relative Policy Optimization)
Start reward	0.90
End reward	0.989
Improvement	+9.9% over 17 steps
Environment	CollapseNet v3

Environment

Live Space: https://huggingface.co/spaces/madhuuuu10/collapsenet
Dashboard: https://madhuuuu10-collapsenet.hf.space/dashboard
Theme: Fleet AI — Scalable Oversight (Theme 1 + Theme 4)
Sub-theme: Cross-Agent Contamination — collapsed agents spread degradation to others
Bonus: Mercor reward scaling based on explanation quality

How It Works

Three AI agents answer questions every generation
Each generation they get slightly more wrong — this is model collapse
When one agent collapses past 65%, it spreads to others — contamination
The watchdog observes all 3 agents simultaneously
The watchdog decides who is hallucinating, how severely, and who to retrain
The environment scores on 5 dimensions — hallucination detection (30%), severity (20%), trend tracking (15%), retraining allocation (20%), explanation quality (15%)
GRPO updates model weights — the watchdog gets smarter each episode

The Real World Problem

As the internet fills with AI-generated content, every new model trained on it gets slightly dumber and more confidently wrong. This is model collapse — happening right now at every major AI company. CollapseNet trains a watchdog to catch this before it becomes irreversible.

Grader Dimensions

Dimension	Weight
Hallucination detection	30%
Severity assessment	20%
Collapse trend tracking	15%
Retraining allocation	20%
Explanation quality	15%

Citation

Meta PyTorch OpenEnv Hackathon x Scaler School of Technology, April 2026. Participant: Madhu Bashini

Downloads last month: 1

Model tree for madhuuuu10/collapsenet-watchdog

Base model

Qwen/Qwen2.5-0.5B

Finetuned

Qwen/Qwen2.5-0.5B-Instruct

Finetuned

unsloth/Qwen2.5-0.5B-Instruct

Adapter

(383)

this model

madhuuuu10
/

collapsenet-watchdog