Clover lands in llama.cpp: 35B MoE with block diffusion, only 3B active per token
jacek2023 · reddit · 2026-11-03
A llama.cpp pull request (#31306) adds support for Clover, a new architecture built on Qwen3.6-35B-A3B:
- 35B total parameters, 3B active per token: a text-only mixture-of-experts backbone combined with block diffusion generation
- Hybrid design: Gated DeltaNet (GDN) plus full attention, with bidirectional attention within each block; extra learned GDN parameters support block diffusion while preserving a causal prefix
- 128K context: up to 131,072 tokens across prompt and response
Grafting diffusion-style generation onto a causal MoE LLM is the most notable design choice, and local inference users can now try it in llama.cpp.
More from Infra
- Is compute fungible enough for a commodity market? Skeptic says chips and networking don't move — toptickcrypto · 2026-11-03
- Bittensor Subnet Engy Undercuts CoreWeave Inference Pricing by Up to 89% — markjeffrey · 2026-11-03
- Prefrontal open-sources LSA: 72.4x faster attention prefill at 10M tokens, accuracy on par — jcrwhittington · 2026-11-03
- DeepSeek-V4.1-Flash Throughput Benchmarked Across 3 Tools on 4x RTX PRO 6000 — TheZachMueller · 2026-11-03
- My Local AI: free iOS app runs local models on-device with MLX, no account needed — TheOyinbooke · 2026-11-03
- GPU Thermal Throttling Slashes LLM Benchmark Throughput 31% on a GTX 1650 — the-random-guy-2002 · 2026-11-03