11 projects
grouped-nf4-gemm
4-bit GPU kernels that compute directly on packed Mixture-of-Experts weights, for training and inference.
experts4bit-qlora
Fine-tune and serve large Mixture-of-Experts models on one GPU, with 4-bit experts in GPU memory, RAM or SSD.
loggetta
Plan and run Mixture-of-Experts fine-tuning on one GPU: check the machine, fit the model, and keep a report of each run.
experts4bit
Alias for experts4bit-qlora — QLoRA fine-tuning of fused 4-bit Mixture-of-Experts on a single small GPU.
experts-mxfp4
Alias for the MXFP4-family experts work — native-byte kernel + provenance in grouped-nf4-gemm, loaders/QLoRA/serving in experts4bit-qlora.
e4b-qlora
Alias for experts4bit-qlora — QLoRA fine-tuning of fused 4-bit Mixture-of-Experts on a single small GPU.
e4b
Alias for experts4bit-qlora — QLoRA fine-tuning of fused 4-bit Mixture-of-Experts on a single small GPU.
grouped-mxfp4-gemm
Alias for grouped-nf4-gemm — the MXFP4 native-byte lane (table-swap kernel, provenance, QLoRA, K3 gather) ships there.
expertsnbit
Alias for experts4bit-qlora — QLoRA fine-tuning of fused 4-bit Mixture-of-Experts on a single small GPU.
nf4gemm
Alias for grouped-nf4-gemm — fused grouped GEMM on bitsandbytes NF4-packed weights with host streaming, on consumer GPUs.
gnf4
Alias for grouped-nf4-gemm — fused grouped GEMM on bitsandbytes NF4-packed weights with host streaming, on consumer GPUs.
