Ant Group Releases Open Source Code Model CodeFuse-CodeLlama-34B in 4bits Quantized Version

The Ant Group has recently released a 4bits quantized version of CodeFuse-CodeLlama-34B, which can still be loaded on a single A10 or RTX4090 after quantization. The quantized model achieved a performance of 73.8% on the Humaneval pass@1 metric. CodeFuse is an exclusive large model for code generation developed by Ant Group, providing intelligent suggestions and real-time support to help developers automatically generate code and enhance research and development efficiency.

Crossing Borders and the Digital Divide: Ant Group Launches the AMP Protocol to Enable a New Payment Chain for Global Intelligent Agents

Artificial intelligence is driving a transformation in business logic from manual user operations to intelligent agent execution. Intelligent agents have become the new entry points for search, decision-making, and shopping, but convenience and security in the payment process remain critical challenges. To address this, Ant Group has launched the Mobile Intelligent Agent Protocol (AMP) to build a trusted and efficient payment infrastructure for this emerging business model.

Ant Group Makes Appearance at the 9th Digital China Construction Summit, Data+AI Application Achievements Exposed for the First Time in Concentrated Manner

Ant Group first showcased its 'Data+AI' as the core, demonstrating a full-stack layout from underlying technology to industrial applications at the 9th Digital China Construction Summit, marking a new stage in its data strategy - the 'Intelligent and Trustworthy Data Flow'. By integrating large models into daily scenarios, AI tools are implemented in practice. The medical AI application 'Ant A Fu' has served over 100 million users, and has collaborated with the Fuzhou Health Commission.

Ant Group Wins Championship at Top Computer Vision Conference, Achieving Practical-Level Advancement in AIGC Detection

Ant Group won the championship in the "Robustness Sample Testing in Complex Real-World Scenarios" and "Facial Enhancement Anomaly Detection" tracks at the CVPR 2026 NTIRE Challenge. This achievement helps enhance risk identification capabilities in scenarios such as payment, content review, and financial authentication. In response to the increasing challenges of deepfakes and misuse of AIGC, as well as the insufficiency of detection models in real-world scenarios and multi-modal large model iterations, this breakthrough provides important technical support.

Cleans 11 Ranking Titles! Ant Group Releases F2LLM-v2: A Full-Scale, Multilingual Hexagonal Embedding Model

Ant Group and Shanghai Jiao Tong University jointly released the F2LLM-v2 series of Embedding models, aiming to break the English-centric bias in the field of semantic representation. The model swept 11 SOTA rankings in the authoritative MTEB benchmark, demonstrating dominant performance. As a fully open-source solution, it combines high performance with extreme efficiency, providing global developers with advanced semantic representation tools.

Ant Group Releases Open Source Code Model CodeFuse-CodeLlama-34B in 4bits Quantized Version

Related Recommendations

Crossing Borders and the Digital Divide: Ant Group Launches the AMP Protocol to Enable a New Payment Chain for Global Intelligent Agents

Ant Group Makes Appearance at the 9th Digital China Construction Summit, Data+AI Application Achievements Exposed for the First Time in Concentrated Manner

Ant Group Wins Championship at Top Computer Vision Conference, Achieving Practical-Level Advancement in AIGC Detection

Ant AI Security Lab Discovers Multiple High-Risk Vulnerabilities in OpenClaw and Assists in Rapid Fixing

Cleans 11 Ranking Titles! Ant Group Releases F2LLM-v2: A Full-Scale, Multilingual Hexagonal Embedding Model