» Tag
moe
16 postsQwisp: MoE expert-streaming engine runs Qwen3.6-35B-A3B on 8GB Macs
Qwisp streams MoE experts from flash to run the 35B-parameter Qwen3.6-A3B model on 8GB Macs, with bit-exact lossless decoding and raw-Metal speed.
Intel Arc B70 32GB Runs Qwen3.6-35B-A3B at 130 t/s via Vulkan
Benchmark: Intel Arc Pro B70 32GB runs Qwen3.6-35B-A3B via llama.cpp Vulkan at ~130 t/s (4-bit, 262k ctx) and ~69 t/s hybrid 8-bit offload.
Custom Vulkan Engine Beats llama.cpp by 1.44x for One Model on RDNA3
A hand-written Vulkan inference engine for one model on RDNA3 GPUs decodes 1.44x faster than llama.cpp with token-exact output parity.
CommitBrief — AI code reviews, right in your terminal
A provider-agnostic, local-first CLI that reviews your staged changes, a historic range, or a whole GitHub pull request. Zero telemetry, no server. Free and open source.
commitbrief.comantirez's DS4 engine: DeepSeek V4 Flash tops local coding on a MacBook Pro
DeepSeek V4 Flash via antirez's DS4 engine one-shots complex coding prompts on a MacBook Pro, outperforming other local models in real-world testing.
Solving Moe Load Imbalance in LLM Training via Optimal Transport
TAOT method improves MoE training speed by 43% while reducing communication costs by 74%.
SkewAdam: A New Optimizer Reducing MoE State Memory by 97%
SkewAdam reduces MoE state memory by 97%, allowing a 6.7B model to fit on a 40GB GPU.