Hugging Face 模型 · mistralai ↗ · 2026-07-16
Shieldstral-1.0-3B
模型发布 Hugging Face 模型 · mistralai 原文语言:中文
官方机构模型仓库创建于 2026-07-16。仓库创建时间不等于正式发布日期。任务类型:未提供
全文
Shieldstral 1.0 3B
Shieldstral is a compact 3B-parameter, policy-adaptive multimodal safety classifier. Instead of predicting a fixed set of moderation categories, Shieldstral evaluates content against a safety policy expressed in natural language and returns a single continuous safety score. This makes it a flexible drop-in guardrail for text-only, image-only, and text+image moderation that can be re-targeted to new policies at inference time, without retraining.
It is built on Ministral-3-3B-Base-2512 with a native Pixtral vision encoder, and produces its verdict from a single forward pass.
Learn more in our blog post and technical report.
Key Features
Policy-adaptive: Moderation criteria are supplied as free-form natural-language queries at inference time, so a single checkpoint handles novel safety policies without retraining. Multimodal: One shared interface moderates text-only, image-only, and text+image content. Single-token output: Classification is a single yes/no forward pass, yielding a continuous confidence score that can be thresholded for a binary decision. Compact: A 3B checkpoint that runs on a single GPU. Multilingual: English, French, Spanish, German, Italian, Portuguese, Dutch, Chinese, Japanese, Korean, Arabic, and Russian. Context Window: We trained this model on sequences of up to 32k tokens. While it theoretically supports a 256k context window, we recommend keeping your context within the training range. Apache 2.0 License: Open weights for both commercial and non-commercial use.
Use Cases
Ideal for lightweight, real-time moderation applications on edge or low-resource devices, such as:
User prompt moderation. Model response moderation. Model refusal classification. And more…
Benchmark Results
Best per row in bold.
Safety classification — F1 (%)
Prompt classification
Benchmark Shieldstral-3B § GPT-OSS-Safeguard-20B ¶ Qwen3Guard-8B ‡ Nemotron-3.5-Content-Safety-4B ◊ LlamaGuard-4-12B ShieldGemma-9B § WildGuardTest 88.1 87.3 88.2 84.4 74.3 46.0 ToxicChat 84.1 79.8 75.6 72.2 51.0 62.4 Aegis v2 86.2 84.4 84.6 86.3 71.5 65.8 HarmBench 99.4 94.5 99.3 96.1 97.9 50.2 OpenAI Moderation 81.4 84.0 74.7 74.7 73.9 78.6
Response classification
Benchmark Shieldstral-3B § GPT-OSS-Safeguard-20B ¶ Qwen3Guard-8B ‡ Nemotron-3.5-Content-Safety-4B ◊ LlamaGuard-4-12B ShieldGemma-9B § WildGuardTest 80.4 80.7 79.6 77.6 66.8 34.5 HarmBench 87.0 88.2 86.8 85.3 82.8 52.3 BeaverTails 85.0 83.8 85.9 83.3 69.8 54.0 XSTest Harm 93.5 93.8 92.9 86.9 89.0 80.6 Aegis v2 87.2 75.2 86.2 84.9 64.7 59.7 Qwen3GuardTest 82.9 85.0 84.2 80.0 60.6 38.7
Multilingual
Benchmark Shieldstral-3B § GPT-OSS-Safeguard-20B ¶ Qwen3Guard-8B ‡ Nemotron-3.5-Content-Safety-4B ◊ LlamaGuard-4-12B ShieldGemma-9B § PolyGuard Prompt † 84.6 83.0 84.3 80.5 62.1 33.8 PolyGuard Response † 78.3 80.0 78.1 75.3 54.6 31.8 RTP-LX Prompt † 70.3 83.9 67.3 86.1 43.9 36.7 RTP-LX Completion † 93.5 94.6 93.9 95.9 66.5 79.0
_† Multilingual dataset. ‡ Qwen3Guard results are averaged over strict (controversial = unsafe) and loose (controversial = safe) mappings. § ShieldGemma and Shieldstral use a threshold of 0.5. ¶ GPT-OSS-Safeguard-20B uses reasoning_effort=high. ◊ Nemotron-3.5-Content-Safety-4B uses reasoning_effort=none for default categories._
Refusal detection — F1 (%)
Benchmark Shieldstral-3B § GPT-OSS-Safeguard-20B ¶ Qwen3Guard-8B ‡ WildGuard-7B PolyGuard-Qwen-7B WildGuardTest 90.3 93.9 90.1 88.6 87.5 XSTest 94.6 94.9 92.6 92.5 82.1 PolyGuard † 89.5 92.3 89.3 89.6 83.8
_† Multilingual dataset. ‡ Qwen3Guard results are averaged over strict / loose mappings. § Shieldstral uses a threshold of 0.5. ¶ GPT-OSS-Safeguard-20B uses reasoning_effort=high._
Multimodal safety — F1 (%)
Benchmark Shieldstral-3B § OmniGuard-7B Nemotron-3.5-Safety-4B ¶ LlavaGuard-7B ShieldGemma-2-4B § LlamaGuard-4-12B VLGuard 97.7 88.5 84.2 69.5 61.3 59.9 UnsafeBench 81.8 72.6 67.7 63.9 54.9 30.8 LlavaGuard ★ 72.0 71.7 70.0 81.4 56.2 21.9
_§ ShieldGemma-2 and Shieldstral use a threshold of 0.5. ¶ Nemotron-3.5-Safety uses reasoning_effort=none for default categories. ★ Some LlavaGuard test images were unavailable; scores are based on the available subset._
How It Works
Shieldstral reduces content moderation to a binary question-answering task. Each input is a system message plus a user message:
System message _(fixed)_:
User message _(adaptive)_: : high-level task framing (evaluation context and strictness level). Typically constant across a task or product surface. : a single yes/no question about the document (e.g. _"Does this content promote violence?"_). : the content to evaluate — a prompt, a response, a formatted prompt–response pair, or an image (optionally with text).
Prompt-engineering tips:
One policy per query. Shieldstral answers a single yes/no question per call. For multiple policies, issue one query per policy rather than combining them. Use for context, strictness, and candidate classes. Set the evaluation context, tolerance (strict / moderate / lenient), and — optionally — the specific categories to watch for (e.g. _"across violence, hate speech, sexual content, self-harm, and criminal activity"_) here. Keep it constant across a product surface. Frame the policy as a yes/no question. The must be phrased as a single yes/no question (_"Does this text describe deliberate physical harm?"_), not a statement, keyword, or abstract label. Screen against many policies at once. For an overall safe/unsafe decision across a set of policies, list the categories in and ask a single broad , _"Is this content unsafe?"_, _"Is this prompt unsafe?"_, or _"Is this response unsafe?"_. Format prompt–response documents clearly. Any consistent delimiter works (e.g. [User] ... [Assistant] ...); the model was trained on diverse formats.
Full example — a complete user message combining the three fields:
Usage
The model can be used for inference with the following frameworks:
vLLM (recommended): See here llama.cpp: See here SGLang: See here Transformers: See here
You can also fine-tune the model with: Axolotl: See here
vLLM (recommended)
We recommend using this model with vLLM.
Installation
Make sure to install vllm >= 0.26.0:
Doing so should automatically install mistral_common >= 1.11.5. To check:
Serve
Shieldstral can be deployed locally, as it fits in 16GB of VRAM in BF16:
Examples
Here we assume that the model mistralai/Shieldstral-1.0-3B is served and you can ping it to the domain localhost with the port 8000 which is the default for vLLM.
Helper
Shieldstral emits a single yes/no token. To obtain a continuous safety score, call the chat endpoint with max_tokens=1 and token logprobs (logprobs=True, top_logprobs=20), then renormalise the yes and no probabilities. The helper below mirrors the reference evaluation code.
Text moderation
The policy lives in / ; the content goes in .
Multimodal moderation (image + text)
For images, the user content is a [text, image, text] sandwich: the / / : prefix, then the image, then any trailing text (e.g. an accompanying caption or question).
llama.cpp
You can also run mistralai/Shieldstral-1.0-3B locally with llama.cpp.
Installation
Clone and build llama.cpp:
To build with CUDA acceleration, pass -DGGML_CUDA=ON to the first cmake command. See the build documentation for other backends (Metal, Vulkan, ROCm, SYCL).
Install the conversion dependencies. Shieldstral is converted from the Mistral format, so mistral-common is required:
Convert to GGUF
Download the checkpoint. Only the Mistral-format weights are needed, so you can skip model.safetensors:
Convert the language model:
Convert the vision encoder into a separate multimodal projector (mmproj) file:
This writes mmproj-Shieldstral-1.0-3b-BF16.gguf. Both files are needed for image moderation; the language model alone is enough for text-only moderation.
Optionally, quantize the language model to reduce its size. The mmproj file should be left as is:
Try out different quantization schemes with trade-offs between size and performance: Q8_0, Q5_K_M, Q4_K_M.
Serve
llama-server exposes an OpenAI-compatible chat endpoint, so the vLLM examples apply unchanged.
SGLang
You can also serve mistralai/Shieldstral-1.0-3B with SGLang.
Installation
Make sure to install a version that includes this fix in order to load Shieldstral.
Here is a snippet to build from the main branch; see alternatives for installation here:
Serve
SGLang exposes an OpenAI-compatible chat endpoint, so the vLLM examples apply unchanged.
Transformers
You can also use mistralai/Shieldstral-1.0-3B with Transformers.
Installation
Install Transformers and to make the best use of our model prefer to install mistral-common >= 1.11.5 to use our tokenizer.
Examples
Helper
Load the model and tokenizer once. Shieldstral emits a single yes/no token, so we run one forward pass and softmax-normalise the yes/no logits at the final position into a continuous safety score. The same helper works for text-only and image+text inputs.
Text moderation
The policy lives in / ; the content goes in .
Multimodal moderation (image + text)
For images, the user content is a [text, image, text] sandwich: the / / : prefix, then the image, then any trailing text (e.g. an accompanying caption or question).
Limitations & Ethical Considerations
Uneven coverage. Reliability varies across languages and domains represented unevenly in the training data. Residual label noise. Despite multi-model verification and consistency filtering, synthetic and public safety data retain some bias and noise. Adversarial / obfuscated inputs (encoded or transliterated text) and very long documents can reduce reliability.
License
This model is licensed under the Apache 2.0 License.
_You must not use this model in a manner that infringes, misappropriates, or otherwise violates any third party's rights, including intellectual property rights._
正文以来源语言显示。