Harvesting forks from GitHub — first visit takes a few seconds…
fukuro-kun/fukuro-llama-cpp-turboquant
llama.cpp fork with TurboQuant WHT-rotated KV cache & weight compression + Gemma 4 MTP speculative decoding for ~30-50% throughput gains modifications: - Add support for gemma4_mtp architecture alongside gemma4_assistant
0 commentsView comments →
No comments yet. Be the first.
Top forks by stars
No public forks with traction yet.
Comments (0)