AI Coding Live News — Real-Time 𝕏/Twitter Updates & Breakthroughs

Dots Studio (Xiaohongshu) & PKU / CUHK Shenzhen@huggingface·Collected 17d ago

👀48

🔥91

Xiaohongshu, PKU, and CUHK Shenzhen Introduce LangDiT: Re-Engineering Transformers for Continuous Diffusion Language Models Slashes GenPPL to 11.33

Continuous diffusion language models (CDLMs) currently transplant architectures from causal LLMs or vision DiTs that are not optimized for parallel diffusion. Researchers from Xiaohongshu (Dots Studio), Peking University, CUHK Shenzhen, and UCLA introduced LangDiT (arXiv:2610.22330). By separating feature normalization from noise modulation, detaching shortcut gradients, and introducing confidence-gated self-conditioning, LangDiT reduces generation perplexity (GenPPL) to 21.72 on LangFlow and 11.33 on Plaid across 1024-token OpenWebText rollouts.

Key Takeaways

Xiaohongshu, PKU, and CUHK Shenzhen introduced LangDiT (arXiv:2610.22330), the first transformer backbone customized for continuous diffusion language models
Decouples representation scaling from noise modulation via Q/K RMSNorm and stabilizes training through vocabulary shortcut detachment
Drives 1024-token generation perplexity on OpenWebText down to 21.72 on LangFlow and a record 11.33 on Plaid formulations
Maintains strong quality (27.51 GenPPL) even under 128 denoising steps, isolating transformer design as the core lever for diffusion text generation

Read full story

Linked Models & Agents:

#LangDiT#Diffusion#Non-Autoregressive#Architecture#Xiaohongshu#NLP

USC / UCLA / CMU@huggingface·Collected 17d ago

👀58

🔥91

USC, UCLA, and CMU Release Grid-Bench for Multi-Step Spatial Reasoning: 8,928 Executable Instances Reveal Severe State Drift Across 64-Step Trajectories

Can LLMs reliably maintain spatial state representations across sequential operations? USC, UCLA, and CMU introduced Grid-Bench (arXiv:2610.22312, GitHub: grid-bench/Grid-Bench), an executable benchmark with 8,928 instances testing 2D/3D/4D grid state transitions across history lengths up to T=64. Evaluations across 6 frontier models show steep degradation: while GPT-5.6 Sol and Claude 5 Sonnet retain strong cell accuracy, open-weight and mid-tier models collapse to near-zero exact world reconstruction over long trajectories.

Key Takeaways

USC, UCLA, and CMU introduced Grid-Bench (arXiv:2610.22312), an executable spatial tracking benchmark across 2D, 3D, and 4D discrete manifolds
Evaluates 8,928 instances across 8 spatial transformation rule families and trajectory lengths up to T=64 steps
At intermediate length T=16, GPT-5.6 Sol (98.0% cell accuracy) and Claude 5 Sonnet (96.8%) lead while mid-tier models drop sharply
At extreme length T=64, exact world recovery drops to 0 for GLM-5 and Kimi K2.5, demonstrating that long-horizon agents must rely on external state engines

Read full story

Linked Models & Agents:

#Grid-Bench#Spatial Reasoning#USC#UCLA#Benchmark#State Tracking

@huggingface@huggingface·Collected 19d ago

👀68

🔥93

Zero-Human-Label Step Accuracy Jumps from 48.2% to 74.7%: LeanPRM Grounds Process Supervision in Lean 4, While Cornell's CPLM Matches Full Attention at w=128

Tackling step-level verification in mathematical reasoning and in-context copying bottlenecks in Transformers, NYU/UCLA/Stanford researchers introduce LeanPRM (arXiv:2610.17697)—using the Lean 4 theorem prover to automatically label 152K steps across 10,634 natural-language proofs with zero human annotation, boosting Qwen3-4B step accuracy on LeanPRM-Bench from 48.2% to 74.7% (vs. 60.3% for PRM800K). Meanwhile, Cornell University's Copy-Paste Language Models (CPLMs, arXiv:2610.19247) add a lightweight Copy-Paste head (~3% compute overhead) that relieves the LM head gradient bottleneck, boosting LAMBADA accuracy by +2.5-4.0 points and allowing a w=128 sliding-window CPLM to match full-attention perplexity.

Key Takeaways

LeanPRM uses the Lean 4 proof assistant and a syntax-error filter to generate 152K step-level labels across 10,634 proofs with zero human annotation
A Qwen3-4B verifier trained on LeanPRM reaches 74.7% step accuracy on LeanPRM-Bench (vs. 48.2% base and 60.3% with PRM800K) and improves 0.6B-8B verifiers on Hard2Verify and IMO-GradingBench
Cornell's Copy-Paste Language Models (CPLMs) add a ~3%-overhead copy head that relieves the LM head gradient bottleneck, boosting LAMBADA by +2.5-4.0 points and matching full-attention perplexity with a w=128 sliding window

Read full story

Linked Models & Agents:

#LeanPRM#Lean-4#Process-Reward-Model#Qwen3#Copy-Paste-LM#Long-Context

Hugging Face Daily Papers@HuggingFace·Collected 25d ago

👀62

🔥98

ReSPO: UCLA Introduces Reshaped Sequence Policy Optimization, Resolving Gradient Starvation in Off-Policy RLVR

Reinforcement Learning from Verifiable Rewards (RLVR) with long Chain-of-Thought (CoT) trajectories powers frontier reasoning models such as DeepSeek-R1 and OpenAI o-series. However, live on-policy sampling consumes more than 80% of cluster compute during post-training. To control training costs, practitioners frequently reuse rollout trajectories across multiple optimization steps, introducing severe off-policy distribution mismatch between the active policy and the rollout-generating policy. Researchers from UCLA unveil a fundamental failure mode in clipped policy optimization (arXiv:2609.35433): 'Sign-Dependent Gradient Starvation'. Standard clipping mechanisms in PPO and GRPO suppress under-generated positive reasoning rollouts relegated to the low-importance-weight tail, while allowing severely over-generated negative trajectories to dominate high-weight updates. This causes models to discard promising reasoning paths during early training. The authors propose ReSPO (Reshaped Sequence Policy Optimization), replacing rigid clipping with a smooth, two-branch sequence-level kernel derived from an alpha-divergence variational objective and an exponential variance-control tilt. Across dense and MoE Qwen3 architectures, ReSPO rescues long positive reasoning rollouts, accelerates early convergence, and achieves superior benchmark accuracy under rollout reuse.

Key Takeaways

UCLA identifies sign-dependent gradient starvation in off-policy RLVR, proposing ReSPO to replace clipped surrogate objectives
Protects sparse positive reasoning rollouts at low-weight tails, boosting early effective gradient throughput by 2.4x on Qwen3 models
Maintains monotonic convergence under 4x rollout reuse, elevating benchmark accuracy by 3.6 to 5.2 points while cutting sampling compute by 50%

Read full story

Linked Models & Agents:

⚡ ReSPO⚡ Qwen3-MoE⚡ DeepSeek-R1 (Open Reasoning)⚡ GRPO🤖 RLVR-Trainer 🤖 Reasoning-Optimizer 🌐 UCLA 🌐 Hugging Face

#ReSPO#RLVR#Reinforcement Learning#Gradient Starvation#UCLA#Qwen3