AI summary
Zyphra’s new attention method speeds up AI processing by using a tree structure to skip irrelevant text, maintaining high performance on complex tasks.
AI-generated summary. Check the original reporting for context and accuracy.
Zyphra AI Introduces Zyphra Hierarchical Attention (ZHA): A Path to True Subquadratic Attention
Source excerpt
This is the available feed excerpt, which may be shortened or in the publisher’s original language.
Zyphra has released Zyphra Hierarchical Attention (ZHA), a sparse attention method built around a tree-structured indexer. ZHA targets a known gap in sparse attention: the indexer that picks tokens still scans the whole context. Zyphra’s technical paper shows ZHA can be distilled into frozen pretrained models with about 4B tokens. On Qwen3.6-35B-A3B it stays close to dense attention on reasoning, agentic coding and 256K retrieval benchmarks. TL;DR Size: It is a method that adds 81,920 trainable parameters to Qwen3.6-35B-A3B (about 0.00024%). Tested up to 256K context. Runs on: Retrofits frozen pretrained models after a 4B-token KL warmup at 32K context. Speedups come from a modeled H100 BF16 roofline. No optimized kernels released. An FP4 router variant was also tested. Performance: Within a few points of dense attention on 7 reasoning, coding and retrieval benchmarks. Best: LongBench v2 hits 63.4% (deep tree) vs 58.0% dense, using about 1.1% of keys. Worst: On NoPE-based Nemotron-3.5-Lightning-30B-A3B, deep-tree RULER-13 drops 25.3 points below dense. What is Zyphra Hierarchical Attention (ZHA)? ZHA is a sparse attention method from Zyphra that finds relevant keys by searching a tree. It keeps the full KV cache. The tree only decides which cached keys each query reads. Fixed-state methods like SSMs and sliding windows saturate at long contexts. Fine-grained sparse methods like DeepSeek Sparse Attention score every past token with an indexer. The final attention is sparse, but the indexer still grows with the full context. Near 1M tokens, cited work shows indexing takes 60 to 90% of layer time. ZHA makes the search itself sparse. How does ZHA’s tree indexer work? Building the tree ZHA splits keys into contiguous chunks and compresses each chunk into a summary key. Summaries are grouped and compressed again, level by level. With chunk size 4, 6 levels cover 4,096 tokens per top node. Summaries come from a small set of learned reference queries. Each ref
About priority
An automated sorting aid based on source trust, keywords, and entities, on a scale of 1–12. High: 9–12; medium: 6–8; low: 1–5. It does not measure urgency, verify reporting, or indicate editorial endorsement.