Instructions to use StentorLabs/Stentor-30M-Instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use StentorLabs/Stentor-30M-Instruct with Transformers:

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

pipe = pipeline("text-generation", model="StentorLabs/Stentor-30M-Instruct")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("StentorLabs/Stentor-30M-Instruct")
model = AutoModelForCausalLM.from_pretrained("StentorLabs/Stentor-30M-Instruct")
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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

PEFT
How to use StentorLabs/Stentor-30M-Instruct with PEFT:
```
Task type is invalid.
```
Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use StentorLabs/Stentor-30M-Instruct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "StentorLabs/Stentor-30M-Instruct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "StentorLabs/Stentor-30M-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/StentorLabs/Stentor-30M-Instruct

SGLang

How to use StentorLabs/Stentor-30M-Instruct 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 "StentorLabs/Stentor-30M-Instruct" \
    --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": "StentorLabs/Stentor-30M-Instruct",
		"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 "StentorLabs/Stentor-30M-Instruct" \
        --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": "StentorLabs/Stentor-30M-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use StentorLabs/Stentor-30M-Instruct with Docker Model Runner:
```
docker model run hf.co/StentorLabs/Stentor-30M-Instruct
```

StentorLabs commited on Feb 22

Commit

76eadc2

verified ·

1 Parent(s): 6a332b4

Update README.md

Browse files

Files changed (1) hide show

README.md +3 -39

README.md CHANGED Viewed

@@ -143,15 +143,6 @@ outputs = model.generate(
 print(tokenizer.decode(outputs[0], skip_special_tokens=True))
 ```
-### Recommended Generation Settings
-| Parameter | Value |
-|---|---|
-| `max_new_tokens` | 40–100 |
-| `temperature` | 1.0–1.2 |
-| `top_p` | 0.5–0.7 |
-| `repetition_penalty` | 1.2–1.4 |
 ---
 ## Stentor-30M vs Stentor-30M-Instruct — Comparative Statistics
@@ -318,7 +309,7 @@ All examples were prepended with a safety system prompt before tokenization.
 |---|---|
 | Hardware | 2× NVIDIA Tesla T4 (16 GB each) |
 | Platform | Kaggle Notebooks (free tier) |
-| Main SFT training time | ~45 min (2,721 s) |
 | Total fine-tune time (all phases) | ~1 hour |
 | Training samples / sec (main phase) | ~19.3 |
@@ -326,33 +317,6 @@ All examples were prepended with a safety system prompt before tokenization.
 ## Evaluation
-### Training Curves
-![Training Loss](training_loss.png)
-![Training Perplexity](training_perplexity.png)
-### Eval Loss at Checkpoints (Main SFT Phase)
-| Step | Approx. Epoch | Eval Loss | Eval PPL |
-|---|---|---|---|
-| 40 | 0.44 | 3.711 | 40.9 |
-| 80 | 0.88 | 3.397 | 29.9 |
-| 120 | 1.32 | 3.272 | 26.4 |
-| 160 | 1.76 | 3.213 | 24.8 |
-| 200 | 2.20 | 3.186 | 24.2 |
-| **240** | **2.64** | **3.176** | **23.9** |
-### Per-Source Eval Loss at End of Epoch 3
-| Source | Eval Loss | Notes |
-|---|---|---|
-| BeaverTails | **2.135** | Model converges strongly on short refusal templates |
-| Seed Safety | 3.086 | Hand-crafted refusals; good fit |
-| FalseReject | 3.322 | Benign-but-edgy prompts; stable throughout training |
-| Dolly | 3.488 | General instruction following; modest increase vs. early training |
-The low BeaverTails eval loss confirms the model learned refusal phrasing effectively. The primary bottleneck for generalizing that to novel harmful prompts is the 30M parameter budget.
 ### Safety Probe Results (Post-Training, 35-prompt suite)
 | Metric | Greedy | Sampled (T=0.7) |
@@ -387,7 +351,7 @@ The low BeaverTails eval loss confirms the model learned refusal phrasing effect
 ## Bias, Risks, and Limitations
 - **Weak safety generalization.** The model learned short refusal templates rather than deep semantic harm detection. Paraphrased or novel harmful prompts frequently bypass refusals.
-- **Terse outputs.** Stop-calibration training produces short responses by design. Increase `max_new_tokens` if you need longer answers.
 - **All base model limitations apply.** 512-token context, limited world knowledge, occasional hallucination — see the [Stentor-30M model card](https://huggingface.co/StentorLabs/Stentor-30M) for full details.
 - **No RLHF.** SFT only — no preference-based alignment was applied.
 - **Dataset biases.** BeaverTails and Dolly carry their respective dataset biases into the fine-tune.
@@ -549,7 +513,7 @@ print(tokenizer.decode(outputs[0], skip_special_tokens=True))
 | Platform | Kaggle (free tier) |
 | Compute region | US West |
 | Total fine-tune time (all phases) | ~1 hour |
-| Estimated CO₂e | ~5 gCO₂e |
 ---

 print(tokenizer.decode(outputs[0], skip_special_tokens=True))
 ```
 ---
 ## Stentor-30M vs Stentor-30M-Instruct — Comparative Statistics
 |---|---|
 | Hardware | 2× NVIDIA Tesla T4 (16 GB each) |
 | Platform | Kaggle Notebooks (free tier) |
+| Main SFT training time | 49 min 32.7 s (2,972.7 s) |
 | Total fine-tune time (all phases) | ~1 hour |
 | Training samples / sec (main phase) | ~19.3 |
 ## Evaluation
 ### Safety Probe Results (Post-Training, 35-prompt suite)
 | Metric | Greedy | Sampled (T=0.7) |
 ## Bias, Risks, and Limitations
 - **Weak safety generalization.** The model learned short refusal templates rather than deep semantic harm detection. Paraphrased or novel harmful prompts frequently bypass refusals.
+- **Terse outputs.** The base Stentor-30M had a persistent tendency to keep generating text well past a natural stopping point rather than terminating cleanly on its own. The stop-calibration phase was specifically designed to reinforce the behavior of ending a response once the answer is complete, so short and clean outputs are intentional.
 - **All base model limitations apply.** 512-token context, limited world knowledge, occasional hallucination — see the [Stentor-30M model card](https://huggingface.co/StentorLabs/Stentor-30M) for full details.
 - **No RLHF.** SFT only — no preference-based alignment was applied.
 - **Dataset biases.** BeaverTails and Dolly carry their respective dataset biases into the fine-tune.
 | Platform | Kaggle (free tier) |
 | Compute region | US West |
 | Total fine-tune time (all phases) | ~1 hour |
+| Estimated CO₂e | ~20 gCO₂e |
 ---