ggml-org/llama.cpp

↓ decelerating · ggml-org/llama.cpp

llama.cpp's velocity dropped to 31.0 from 45.6 stars per day with -14.6 acceleration, triggered by saturation after commits like f84270e and 0f1bb60 optimized token generation and model compatibility for Qwen3 and LLaMA. This reflects a post-optimization lull, as benchmarks in the pull requests showed gains up to 20% in speed, reducing immediate developer urgency. Compared to vLLM's -7.3 and Ollama's -7.7, llama.cpp's sharper drop highlights its prior lead at 4x peers, now normalizing. Investors evaluating inference should view this as a signal for potential feature mergers, as sustained deceleration below -10 could pressure the project to integrate with platforms like PyTorch for broader hardware support.

Receipts — documents this drew from

• llama.cpp (ggml-org/llama.cpp)
• vLLM (vllm-project/vllm)
• Ollama (ollama/ollama)

From the briefing: 2026-04-27 · Inference Runtimes Decelerate Amid Platform Acceleration