v4.3.1
The popular open-source AI UI now supports Google's latest Gemma 4 model with full tool-calling capabilities.
Oobabooga has launched version 4.3.1 of text-generation-webui, the widely-used open-source interface for running local large language models. The headline addition is full support for Google's recently released Gemma 4 model, including its tool-calling capabilities through both the API and user interface. This allows users to run the 27-billion parameter model locally with the same functionality available in cloud services. The update also introduces ik_llama.cpp as a new backend option, featuring improved quantization methods like Hadamard KV cache rotation that significantly enhance accuracy for memory-efficient inference.
Performance improvements are substantial, with Oobabooga's custom Gradio fork now saving up to 50 milliseconds per UI interaction like button clicks. The API has been enhanced with echo and logprobs support for the completions endpoint, providing token-level probability data that's crucial for developers building applications. Security receives major attention with fixes for SSRF (Server-Side Request Forgery) vulnerabilities in the Superbooga extensions, preventing malicious URL requests to internal networks. The update also patches access control bypass issues in Windows/macOS path matching and updates core dependencies including llama.cpp and ExLlamaV3 for better model compatibility and inference efficiency.
- Adds full Gemma 4 support with tool-calling capabilities in both API and UI interfaces
- Integrates ik_llama.cpp backend with Hadamard KV cache rotation for 50% more accurate quantization
- Fixes critical SSRF vulnerability in Superbooga extensions that could access internal networks
Why It Matters
Enables professionals to run state-of-the-art models like Gemma 4 locally with enterprise-grade security and performance optimizations.