🚨 Breaking
Aucun changement. Weaviate v1.38.2 and v1.37.10 are tagged as breaking but summaries explicitly state no breaking changes. No action required.
🗑️ Dépréciations
Aucun changement.
💰 Pricing
Aucun changement.
🆕 Nouveautés
- Weaviate v1.38.2 adds a new
generative-deepseekmodule, enabling integration with DeepSeek models. Release notes - llama.cpp b9788 adds SYCL tensor parallelism (
--split-mode tensor) for dual-GPU setups. Release notes - llama.cpp b9803 fixes OpenCL profiling batch flush at shutdown. Release notes
- llama.cpp b9789 fixes quantizing MoE with MTP. Release notes
- llama.cpp b9802 includes various improvements (see release).
- Goose v1.39.0 adds ACP-based recipe management and slash commands. Release notes
- Ollama v0.30.11-rc0 adds thinking capability detection and auto-install for Claude Code/opencode. Release notes
- LlamaIndex v0.14.23 updates dependencies and minor improvements. Release notes
🌐 Actualité IA
Aucun signal.
💡 Conseil du jour
If you use Weaviate, evaluate the new generative-deepseek module for potential cost savings and model diversity. For self-hosted LLM inference, consider upgrading llama.cpp to b9788 if you need SYCL tensor parallelism on dual-GPU systems.