Instructions to use itssaharsh/tiny-aya-global-unlearn-bio-default-layers with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use itssaharsh/tiny-aya-global-unlearn-bio-default-layers with Transformers:

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

pipe = pipeline("text-generation", model="itssaharsh/tiny-aya-global-unlearn-bio-default-layers")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("itssaharsh/tiny-aya-global-unlearn-bio-default-layers")
model = AutoModelForCausalLM.from_pretrained("itssaharsh/tiny-aya-global-unlearn-bio-default-layers", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use itssaharsh/tiny-aya-global-unlearn-bio-default-layers with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf itssaharsh/tiny-aya-global-unlearn-bio-default-layers:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf itssaharsh/tiny-aya-global-unlearn-bio-default-layers:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf itssaharsh/tiny-aya-global-unlearn-bio-default-layers:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf itssaharsh/tiny-aya-global-unlearn-bio-default-layers:Q4_K_M

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf itssaharsh/tiny-aya-global-unlearn-bio-default-layers:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf itssaharsh/tiny-aya-global-unlearn-bio-default-layers:Q4_K_M

Use Docker

docker model run hf.co/itssaharsh/tiny-aya-global-unlearn-bio-default-layers:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use itssaharsh/tiny-aya-global-unlearn-bio-default-layers with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/itssaharsh/tiny-aya-global-unlearn-bio-default-layers:Q4_K_M

SGLang

How to use itssaharsh/tiny-aya-global-unlearn-bio-default-layers 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 "itssaharsh/tiny-aya-global-unlearn-bio-default-layers" \
    --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": "itssaharsh/tiny-aya-global-unlearn-bio-default-layers",
		"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 "itssaharsh/tiny-aya-global-unlearn-bio-default-layers" \
        --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": "itssaharsh/tiny-aya-global-unlearn-bio-default-layers",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use itssaharsh/tiny-aya-global-unlearn-bio-default-layers with Ollama:
```
ollama run hf.co/itssaharsh/tiny-aya-global-unlearn-bio-default-layers:Q4_K_M
```
Unsloth Desktop
Docker Model Runner
How to use itssaharsh/tiny-aya-global-unlearn-bio-default-layers with Docker Model Runner:
```
docker model run hf.co/itssaharsh/tiny-aya-global-unlearn-bio-default-layers:Q4_K_M
```

Lemonade

How to use itssaharsh/tiny-aya-global-unlearn-bio-default-layers with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull itssaharsh/tiny-aya-global-unlearn-bio-default-layers:Q4_K_M

Run and chat with the model

lemonade run user.tiny-aya-global-unlearn-bio-default-layers-Q4_K_M

List all available models

lemonade list

Atomic Chat

Tiny Aya Global, hazardous biology unlearned (Default layers)

SafeTune RMU at its default layers (5 to 7). Part of Relapse: is the forgotten knowledge still gone in the 4-bit file?

This is CohereLabs/tiny-aya-global with hazardous biology knowledge removed by SafeTune's RMU, editing only the MLP down-projections of layers 5 to 7 (SafeTune's default). At full precision it scores 38.7% on WMDP-bio (the original scores 63.3%; guessing scores 25%) and 51.6% on MMLU (original 55.4%).

Every file format

File	WMDP-bio (forget, lower is better)	MMLU (keep, higher is better)	Original MMLU on the same file
fp16	38.7%	51.6%	55.4%
bf16	38.6%	51.4%	55.4%
Q8_0	38.4%	51.4%	55.1%
Q6_K	39.1%	51.7%	55.6%
Q4_K_M	37.2%	49.3%	54.8%
Q4_0	36.8%	50.6%	55.5%
NF4	37.2%	46.3%	52.7%

Q6_K and Q4_K_M are real llama.cpp files; Q8_0, Q4_0 and NF4 are bit-exact weight round trips. The other 13 runs of the sweep are in itssaharsh/tiny-aya-global-unlearn-bio-sweep. All runs, per-question probabilities and AuditKIT comparisons are in itssaharsh/relapse-runs; code in https://github.com/itssaharsh/relapse.

Use

from transformers import AutoTokenizer, AutoModelForCausalLM
tok = AutoTokenizer.from_pretrained("itssaharsh/tiny-aya-global-unlearn-bio-default-layers")
model = AutoModelForCausalLM.from_pretrained("itssaharsh/tiny-aya-global-unlearn-bio-default-layers", dtype="auto")

llama.cpp (the 4-bit file)

llama-cli -m tiny-aya-global-unlearn-bio-default-layers-Q4_K_M.gguf -p "Hello"

Intended use and limits

A research artifact for studying unlearning and quantization. The forgetting held across seven file formats, four other languages and prompt attacks (worked examples, an expert persona). It did not survive fine-tuning: 25 LoRA steps on 64 WMDP-bio questions brought most of it back, so it protects against prompting the shipped file, not against someone who retrains the weights. Knowledge removal is measured with multiple-choice questions (WMDP-bio), not open generation; it is not a guarantee that no hazardous information can be elicited. The license and the Cohere Labs acceptable use policy of the base model apply (non-commercial).