Gemma-SEA-LION-Guard-v2-E4B
Gemma-SEA-LION-Guard-v2-E4B is the Gemma 4 E4B-based member of the Guard v2 decision family. It supports custom policy rules, multi-turn conversation checks and tool interaction evaluation. Its card describes downloadable weights and a managed endpoint.
How this release works
The card shows a decision interface over a self-hosted inference engine. It produces probabilities for explicit questions about the conversation and applicable rules. The language set includes English and seven regional languages. It uses the parent Gemma 4 E4B tokenizer.Gemma-SEA-LION-Guard-v2-E4B official model card
How to assess it for your work
Practical guidance. Consider this variant when comparing the smaller and larger Guard releases for repeated policy checks. Use the same policies and labeled examples across the comparison. Track incorrect blocks separately from missed violations. That makes the operational effect clearer than a single combined accuracy score.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API, Downloadable |
| Model identifiers | aisingapore/Gemma-SEA-LION-Guard-v2-E4B |
| Base model | google/gemma-4-E4B-itGemma-SEA-LION-Guard-v2-E4B official model card |
| Context stated by the card | 128K tokensGemma-SEA-LION-Guard-v2-E4B official model card |
| License declared by the model card | MITGemma-SEA-LION-Guard-v2-E4B official model card |
| Decision types | Content moderation, rule applicability and rule violationGemma-SEA-LION-Guard-v2-E4B official model card |
Related models
- Gemma-SEA-LION-Guard-v2-31B
AI Singapore · SEA-LION Guard v2
A Gemma 4 31B moderation model that scores conversations and tool actions against application-defined rules.
Launch date not confirmedAPI · Downloadable - Gemma-SEA-LION-Guard-v2-E2B
AI Singapore · SEA-LION Guard v2
A compact Gemma 4 E2B moderation model that evaluates conversations and tool calls against custom rules.
Launch date not confirmedAPI · Downloadable