Back to home
CUDA
2 articles tagged with this topic
Gemma 4llama.cpp
Gemma 4 Local CUDA Setup: Precision Traps and Real Benchmarks
Running Gemma 4 locally on CUDA requires strict dtype matching at KV cache boundaries or output degenerates silently.
Apr 72 min read
Kokoro TTSKVoiceWalk
Kokoro TTS Trainer Gets GPU Support: 26 Hours Down to 4
Forked KVoiceWalk adds CUDA support to Kokoro TTS voice training, cutting time from 26 hours to ~4 hours on an RTX 3060.
Apr 52 min read