Instructions to use netis-ai/RavenGuard-gen with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use netis-ai/RavenGuard-gen with Transformers:

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

pipe = pipeline("text-generation", model="netis-ai/RavenGuard-gen", trust_remote_code=True)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("netis-ai/RavenGuard-gen", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use netis-ai/RavenGuard-gen with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/netis-ai/RavenGuard-gen

SGLang

How to use netis-ai/RavenGuard-gen 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 "netis-ai/RavenGuard-gen" \
    --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": "netis-ai/RavenGuard-gen",
		"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 "netis-ai/RavenGuard-gen" \
        --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": "netis-ai/RavenGuard-gen",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use netis-ai/RavenGuard-gen with Docker Model Runner:
```
docker model run hf.co/netis-ai/RavenGuard-gen
```

RavenGuard-gen

Commit History

Jul 17

04:22
weights: v11-rgclean step3000 (clean firewalled ckpt; replaces prior weights). Card update pending.

vadery

Verified signature
bf8d98e

Jul 16

01:25
reproduce_eval: neutral scoring note (Severity-token, reference AUROC)

vadery

Verified signature
67bd6e2
01:25
manifest: align score description to shipped Severity-token read

vadery

Verified signature
d59a728
01:25
card: AUROC -> reproducible Severity-token figures (87/89); drop unshipped Decision-token 92/96 headline + false reproduce_eval attribution

vadery

Verified signature
8993dbf
00:50
Add eval_data/ (4 datasets, 4858 rows)

vadery

Verified signature
576f936
00:50
Add reproduce_eval.py (self-contained 4-axis reproduction)

vadery

Verified signature
5e9622a

Jul 15

15:29
Update PolyGuard figure: RavenGuard AUROC 87.0->92.2 (Decision-token)

vadery

Verified signature
2d2fc35
15:22
Backfill corrected Decision-token AUROC (ChineseSafe 89→96.1, PolyGuard 87→92.2)

vadery

Verified signature
d1182f1
15:00
Fix critical: rotary embeddings were all-zero (no positional encoding)⁠

vadery

Verified signature
752d194
09:01
Correct RavenGuard language metadata to the base-18 training list

vadery

Verified signature
a9816d4
08:56
Make bundled backbone snapshot resolution deterministic

vadery

Verified signature
97a2e7c
08:55
Fix bundled Netis Amniota loading and update 0.8B parameter labels

vadery

Verified signature
31199ea
06:21
Self-contained backbone (netis_amniota); remove nanoai dependency & naming (forward bit-identical)

vadery

Verified signature
bcbed37
01:52
Gen-only model card: remove Stream sections/rows (Stream ships in its own repo)

vadery

Verified signature
2d2f68b
01:39
refresh figure size_vs_chinese.png

vadery

Verified signature
9657d5f
01:39
refresh figure service_sim_disruption.png

vadery

Verified signature
ec237e1
01:38
refresh figure polyguard_multilingual.png

vadery

Verified signature
ea05105
01:38
refresh figure chinesesafe_f1.png

vadery

Verified signature
a607ee6
01:38
Refresh model card: honest Stream numbers (remove fabricated Qwen AUROC / 2270ms / 140x; add measured 13.2ms latency, categorical-vs-continuous framing)

vadery

Verified signature
0daf984

Jul 14

03:17
RavenGuard-Gen v16-s2000: weights (HF trust_remote_code) + refreshed card/figures

vadery

Verified signature
40ee3bf
03:16
RavenGuard-Gen v16-s2000: weights (HF trust_remote_code) + refreshed card/figures

vadery

Verified signature
5be5643

Jul 10

09:15
Add comparison figures

vadery

Verified signature
44d98c3
09:15
Add model card (v0)

vadery

Verified signature
07bb941
09:15
initial commit

vadery

Verified signature
a60e6ce

Commit History

Jul 17

weights: v11-rgclean step3000 (clean firewalled ckpt; replaces prior weights). Card update pending.

Jul 16

reproduce_eval: neutral scoring note (Severity-token, reference AUROC)

manifest: align score description to shipped Severity-token read

card: AUROC -> reproducible Severity-token figures (87/89); drop unshipped Decision-token 92/96 headline + false reproduce_eval attribution

Add eval_data/ (4 datasets, 4858 rows)

Add reproduce_eval.py (self-contained 4-axis reproduction)

Jul 15

Update PolyGuard figure: RavenGuard AUROC 87.0->92.2 (Decision-token)

Backfill corrected Decision-token AUROC (ChineseSafe 89→96.1, PolyGuard 87→92.2)

Fix critical: rotary embeddings were all-zero (no positional encoding)⁠

Correct RavenGuard language metadata to the base-18 training list

Make bundled backbone snapshot resolution deterministic

Fix bundled Netis Amniota loading and update 0.8B parameter labels

Self-contained backbone (netis_amniota); remove nanoai dependency & naming (forward bit-identical)

Gen-only model card: remove Stream sections/rows (Stream ships in its own repo)

refresh figure size_vs_chinese.png

refresh figure service_sim_disruption.png

refresh figure polyguard_multilingual.png

refresh figure chinesesafe_f1.png

Refresh model card: honest Stream numbers (remove fabricated Qwen AUROC / 2270ms / 140x; add measured 13.2ms latency, categorical-vs-continuous framing)

Jul 14

RavenGuard-Gen v16-s2000: weights (HF trust_remote_code) + refreshed card/figures

RavenGuard-Gen v16-s2000: weights (HF trust_remote_code) + refreshed card/figures

Jul 10

Add comparison figures

Add model card (v0)

initial commit