mirror of
https://github.com/hanzoai/ml.git
synced 2026-11-20 22:46:50 +00:00
Three-way per file through the rename: base 39355c6c91, theirs the upstream
merge of main with 35d7ae7c (via 7ef6a694d), ours origin/main.
Lands the barrier thread pool, runtime CPU feature dispatch, aarch64 and x86
repacked quantized kernels (indexed_gemv, gemv_fused_shared_lhs), quantized
embedding on CPU/Metal/CUDA, paged flash-attn kernels, cutile fused MoE, the
in-place op aliasing API, parking_lot storage locks, Metal buffer builders and
debug labels, safetensors 0.8, hf-hub 1.0, tokenizers 0.23.
Ours keeps ROCm, Vulkan, wgpu, the IQ/MXFP4/NVFP4/TQ dtypes, and the i-quant
weight/activation block split in matmul. Our backends ride InplaceOpN and the
quantized embedding; Metal get_rows gains the IQ kernels. ug leaves, as upstream.
Co-authored-by: Hanzo Dev <[email protected]>
hanzo-ml-mobileclip
MobileCLIP is family of efficient CLIP-like models using FastViT-based image encoders.
See MobileCLIP: Fast Image-Text Models through Multi-Modal Reinforced Training
Running on an example on cpu
$ cargo run --example mobileclip --release -- --images "hanzo-ml-examples/examples/stable-diffusion/assets/stable-diffusion-xl.jpg","hanzo-ml-examples/examples/yolo-v8/assets/bike.jpg" --cpu --sequences "a cycling race","a photo of two cats","a robot holding a hanzo"
softmax_image_vec: [2.4819004e-5, 3.81081e-6, 0.9999714, 0.9999738, 2.382714e-5, 2.3317718e-6]
Results for image: hanzo-ml-examples/examples/stable-diffusion/assets/stable-diffusion-xl.jpg
Probability: 0.0025% Text: a cycling race
Probability: 0.0004% Text: a photo of two cats
Probability: 99.9971% Text: a robot holding a hanzo
Results for image: hanzo-ml-examples/examples/yolo-v8/assets/bike.jpg
Probability: 99.9974% Text: a cycling race
Probability: 0.0024% Text: a photo of two cats
Probability: 0.0002% Text: a robot holding a hanzo