tool-calling

2 articles tagged with this topic

Qwen-32Bllama.cpp

Local LLMs Lose Tool Call Accuracy After 8–9 Chained Calls

Qwen 32B, Gemma 9B, and Command R 32B all fail similarly after 8+ tool calls — attention dilution, not context limits.

Apr 84 min read

Gemma 4llama.cpp

Fixing Gemma 4 Tool Calls in llama.cpp: Root Causes Explained

Four bugs in llama.cpp's Gemma 4 chat template handling caused tool call results to crash or loop.

Apr 83 min read