cd /news/machine-learning/tphc-improving-hyper-connections-wit… · home › topics › machine-learning › article
[ARTICLE · art-172360] src=machinebrief.com ↗ pub= topic=machine-learning verified=true sentiment=↑ positive

TPHC: Improving Hyper-Connections with Tensor-Parallel Grouped Updates

Researchers behind the arXiv paper 2611.23026v1 introduced TPHC, a method that routes a separate mixture of residual streams to each tensor-parallel shard in a Transformer, letting each hyper-connection stream receive a different mixture of shard outputs without adding parameters or FLOPs to the attention or FFN sublayers. On a 1.1B model trained on 100B tokens, TPHC cut validation loss by 0.7-1.0% and raised average zero-shot accuracy across six commonsense reasoning benchmarks by 0.3-0.7 percentage points versus manifold-constrained hyper-connections (mHC) at matched stream counts N in {4,8,16}. TPHC cost only 9% more training time than mHC on an 8B model at four streams, and going from four to sixteen streams raised training time by just 19%; code is available at github.com/wuyou2002/TPHC.

recommendation

When training with hyper-connections under tensor parallelism, route per-shard residual mixtures and keep shard outputs separate on write-back; the source measured 0.7-1.0% lower validation loss for 9% more training time.

more in playbook: Fine-tuning and data
by read1 min views3 publishedNov 19, 2026

arXiv:2611.23026v1 Announce Type: new Abstract: Hyper-Connections (HC) expand the residual stream of a Transformer into $N$ parallel streams, and manifold-constrained HC (mHC) stabilizes them at scale. However, mHC benefits little from more than four streams, partly because each sublayer writes a single output to all streams, limiting how differently they can be updated. In this work, we propose TPHC, which integrates tensor-parallel (TP) shards with HC streams to promote greater independence across streams. Specifically, we route a separate mixture of the residual streams to each TP shard and keep the individual shard outputs separate during write-back, so each stream can receive a different mixture of those outputs. This design makes the residual-stream updates more expressive without adding parameters or FLOPs to the attention or FFN sublayers. With native support for stream sharding under TP, TPHC requires only 9% more training time than mHC on an 8B model when both use four streams. Increasing the number of streams from four to sixteen raises its training time by only 19%. On a 1.1B model trained on 100B tokens, TPHC reduces validation loss by 0.7-1.0% and raises average zero-shot accuracy across six commonsense reasoning benchmarks by 0.3-0.7 percentage points relative to mHC at matched stream counts $N\in{4,8,16}$. Our code is available at https://github.com/wuyou2002/TPHC.

── more in #machine-learning 4 stories · sorted by recency
── more on @tphc 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/tphc-improving-hyper…] indexed:0 read:1min 2026-11-19 · —