zenlm/zen-guard-stream
1
0
Fork 0
mirror of https://github.com/zenlm/zen-guard-stream.git synced 2026-07-26 22:08:53 +00:00
Code Issues Packages Projects Releases Wiki Activity

docs: honest attribution — correct 0.6B (was fabricated 4B) + credit Qwen3Guard-Stream-0.6B upstream (apache-2.0)

Browse Source
This commit is contained in:
z
2026-07-16 10:52:30 -07:00
parent e4605dff59
commit 78a7ee3aae
1 changed files with 14 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+14 -11
View File
@@ -7,7 +7,7 @@
<h1 align="center">Zen Guard Stream</h1>
<p align="center">
<strong>4B Real-Time Streaming Safety Moderation</strong>
<strong>Real-Time Streaming Safety Moderation</strong>
</p>
<p align="center">
@@ -19,9 +19,13 @@
---
> **Status:** Superseded by [zenlm/zen3-guard](https://huggingface.co/zenlm/zen3-guard). Weights remain available for reproducibility.
## Introduction
**Zen Guard Stream** is a 4B parameter token-level streaming classifier that evaluates each generated token in real time. It's optimized for real-time content moderation during LLM response generation.
**Zen Guard Stream** is a 0.6B parameter token-level streaming classifier that evaluates each generated token in real time. It's optimized for real-time content moderation during LLM response generation.
Repackaged from [Qwen/Qwen3Guard-Stream-0.6B](https://huggingface.co/Qwen/Qwen3Guard-Stream-0.6B) (apache-2.0, Alibaba Qwen — Qwen3Guard). **Not trained from scratch** — an Apache-2.0 redistribution for the OSS-clean, commercially-usable Zen model line.
## Features
@@ -47,14 +51,13 @@ LLM Response → Token by Token → Real-time Classification
| Specification | Value |
|---------------|-------|
| Parameters | 4B |
| Parameters | 0.6B |
| Type | Streaming Classifier |
| Base Model | 4B dense transformer |
| Context Length | 32,768 tokens |
| Base Model | [Qwen/Qwen3Guard-Stream-0.6B](https://huggingface.co/Qwen/Qwen3Guard-Stream-0.6B) (apache-2.0) |
| Architecture | Qwen3 (`Qwen3ForGuardModel`) |
| Context Length | 8,192 tokens |
| Languages | 119 |
| Latency | 5ms/token |
| VRAM (FP16) | 8GB |
| VRAM (INT8) | 4GB |
## Quick Start
@@ -160,12 +163,12 @@ def safe_generate(prompt):
## Related Models
- [zen-guard](https://huggingface.co/zenlm/zen-guard) - 4B base model
- [zen-guard-gen](https://huggingface.co/zenlm/zen-guard-gen) - 8B generative model
- [zen3-guard](https://huggingface.co/zenlm/zen3-guard) - canonical successor
- [zen-guard-gen](https://huggingface.co/zenlm/zen-guard-gen) - generative moderation
## License
Apache 2.0
`Apache-2.0`. Upstream: **[Qwen/Qwen3Guard-Stream-0.6B](https://huggingface.co/Qwen/Qwen3Guard-Stream-0.6B)** by Alibaba Qwen (apache-2.0). Upstream LICENSE/NOTICE retained in-repo.
## Citation
@@ -181,7 +184,7 @@ Apache 2.0
## Architecture
Zen Guard Stream is a 4B-parameter streaming safety classifier with 5ms/token latency. It evaluates each generated token in real time, enabling early detection and intervention before unsafe content is fully rendered.
Zen Guard Stream is a 0.6B-parameter streaming safety classifier (Qwen3, `Qwen3ForGuardModel`) with 5ms/token latency. It evaluates each generated token in real time, enabling early detection and intervention before unsafe content is fully rendered.
---