mirror of
https://github.com/hanzoai/ml.git
synced 2026-11-20 23:08:29 +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-lfm2: LFM2.5 (Liquid Foundation Model 2.5)
LFM2.5 is a hybrid architecture from LiquidAI that combines attention and short convolution layers for efficient sequence processing.
Running the example
cargo run --example lfm2 --release -- --prompt "The capital of France is"
For the "thinking" model variant with chat template:
cargo run --example lfm2 --release -- \
--which lfm2.5-1.2b-thinking \
--prompt "<|im_start|>user\nWhat is 2+2?<|im_end|>\n<|im_start|>assistant\n"
On a CUDA-enabled machine with flash attention:
cargo run --example lfm2 --features cuda,flash-attn --release -- \
--use-flash-attn --prompt "The capital of France is"