Coding Tools
llama.cpp b11009 fixes Gemma4 and Qwen35 fused attention
llama.cpp Releaseshigh confidence
Why it matters
- →Fixes tensor shape mismatches for Gemma4 and Qwen3.5 when using fused QKV attention.
- →Enables stable inference for newer models with non-standard embedding dimensions.
- →Maintains broad hardware support while temporarily disabling KleidiAI on macOS.
llama.cpp version b11009 addresses specific architectural requirements for Gemma4 and Qwen3.5 models by correcting split state calculations for fused QKV attention layers. The fix ensures proper handling when embedding dimensions (n_embd) differ from key head dimensions, a common scenario in newer model designs. Standard binaries for CUDA 12/13, ROCm 10.0, and various CPU targets are available, but KleidiAI support on macOS is currently disabled. This update resolves immediate compatibility issues for users attempting to fuse attention layers in these specific model families.
Read originalMore from llama.cpp Releases
Coding Toolscoding
llama.cpp b11000 fixes critical RCE in server
This release patches a critical remote code execution vulnerability in the llama.cpp server that allowed unauthenticated attackers to hijack memory via dangling pointers. The flaw stemmed from caching compute graphs that referenced freed buffers, enabling heap corruption and arbitrary code execution through subsequent tensor commands. By discarding cached graphs when buffers are freed, the fix forces a safe fallback to full recomputation without changing the API. This is a vital security update for anyone running the llama.cpp server remotely, closing a direct path to system compromise.
llama.cpp Releases
Coding Toolscoding
llama.cpp b11001 fixes Spacemit transpose bug
A copy-paste error in llama.cpp was corrupting matrix transpositions on Spacemit hardware, causing significant data corruption for int16 operations. This release patches the specific RVV instruction call to ensure correct computation on these RISC-V based chips. While niche, it prevents silent inference failures for users relying on this specific accelerator architecture. The update also ships binaries for CUDA 13 and ROCm 10.0, keeping the runtime current with latest driver ecosystems.
More in Coding Tools
Coding Toolscoding
Claude Code v2.1.274 fixes MCP timeouts and session stability
This release targets the friction points that make local AI coding feel fragile. The most critical fix addresses MCP servers timing out after five minutes regardless of configuration, a major blocker for complex agent workflows. Session reliability also improves with self-healing corrupted transcripts and better handling of background agents during resumption. While not feature-heavy, these patches stabilize the environment for developers relying on long-running automated tasks.
Claude Code Releases
© GitHub ChangelogCoding Toolscoding
GitHub automates SSO for PATs and SSH keys
GitHub Enterprise Cloud finally addresses the friction of manual SSO authorization for classic tokens and SSH keys. Admins can now delegate bulk authorization to GitHub Apps via a new API, handling up to 50 organizations in one request without exposing secrets. This shift from manual per-org clicks to automated delegation reduces the temptation to use insecure long-lived tokens. It is a practical infrastructure improvement that streamlines credential rotation for large enterprises managing complex access controls.
