Files
zeekayandHanzo Dev 6a363f7ee3 merge: upstream main ddf1b879 and the aarch64 repack branch 35d7ae7c
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]>
2026-09-15 15:52:28 -07:00
..
2026-01-16 12:30:55 -08:00

hanzo-ml-splade

SPLADE is a neural retrieval model which learns query/document sparse expansion via the BERT MLM head and sparse regularization. Sparse representations benefit from several advantages compared to dense approaches: efficient use of inverted index, explicit lexical match, interpretability... They also seem to be better at generalizing on out-of-domain data. In this example we can do the following two tasks:

  • Compute sparse embedding for a given query.
  • Compute similarities between a set of sentences using sparse embeddings.

Sparse Sentence embeddings

SPLADE is used to compute the sparse embedding for a given query. The model weights are downloaded from the hub on the first run. This makes use of the BertForMaskedLM model.

cargo run --example splade --release -- --prompt "Here is a test sentence"

> "the out there still house inside position outside stay standing hotel sitting dog animal sit bird cat statue cats"
> [0.10270107, 0.269471, 0.047469813, 0.0016636598, 0.05394874, 0.23105666, 0.037475716, 0.45949644, 0.009062732, 0.06790692, 0.0327835, 0.33122346, 0.16863061, 0.12688516, 0.340983, 0.044972017, 0.47724655, 0.01765311, 0.37331146]
cargo run --example splade --release --features

> score: 0.47 'The new movie is awesome' 'The new movie is so great'
> score: 0.43 'The cat sits outside' 'The cat plays in the garden'
> score: 0.14 'I love pasta' 'Do you like pizza?'
> score: 0.11 'A man is playing guitar' 'The cat plays in the garden'
> score: 0.05 'A man is playing guitar' 'A woman watches TV'