All Articles · SURL AI News

Transformers cache offloading reportedly changes Qwen3.8 outputs on first reuse

A report with a minimal reproducer shows that a hybrid-attention model’s output can diverge on the next step even after prefill completes normally. Maintainers have yet to confirm the issue, underscoring the need to verify numerical consistency across multiple steps when optimizing GPU memory usage.

2026-09-20

AI 基礎設施

AgentCore Runtime V2 adds memory reclamation and snapshot-based startup, with cold starts of about two seconds in AWS tests

AWS has updated its agent runtime to reduce memory usage during long sessions and shorten cold starts. The performance figures come from no-op agent tests; cost savings and snapshot restore behavior still need to be validated against real workloads.

2026-09-20

GitHub Repo

CUA-S1-FORMS uses 700,000 parameters for form decisions, releases weights and training pipeline

Cua assigns field matching and action selection to a small model, while application code determines execution order. Public evaluations remain limited to a few demonstration forms; generalization to Chinese and end-to-end workflow reliability still need validation.

2026-09-20

GitHub Repo

Cherry Studio 2.1.0 adds website WebMCP support, expanding browser control and document section references

The release lets agents and chat assistants control the built-in browser and reference specific sections of documents. Website tool compatibility, permission differences across execution modes, and minimum macOS requirements remain key adoption considerations.

2026-09-20

AI 安全

Trail of Bits Builds MASM Audit Tools with Agents, Publishes 95 Lean Correctness Proofs

The team used agents to help build decompilation, static analysis, and formal verification tools, then applied them to an audit of the Miden virtual machine. The public code provides a starting point for reproducing the verification, but the proofs remain bounded by the semantic model and theorem assumptions.

2026-09-20

GitHub Repo

TypeSafe Python SDK 0.7.0 Adds Typed Responses for Jev, Switches Serialization to Pydantic

The update maps decision answers to Pydantic models for classification and routing workflows. The serialization change requires migration testing, while Chinese-language accuracy and probability thresholds still need separate evaluation.

2026-09-20

Grok Voice Transcribe 2.0 reduces transcription errors; watch the default model switch

The new version retains existing transcription pricing, while both official multilingual short-utterance tests and independent batch benchmarks show lower word error rates. Batch speed, Chinese formatting support, and the default model switch still require separate verification.

2026-09-19

推論系統

AMD details MLPerf inference optimizations, boosting eight-GPU GPT-OSS serving throughput by about 38%

AMD improved GPT-OSS-120B benchmark throughput on MI355X GPUs by splitting attention kernels and adjusting scheduling. Public weights and reproduction instructions provide a starting point, but aspects of the environment configuration and model card still need clarification.

2026-09-19

GitHub Repo

vLLM Publishes Multi-GPU NVDEC Tests, Showing More Than Twice the Video Captioning Throughput of CPU Decoding

vLLM has published multi-GPU scaling tests of hardware video decoding, easing CPU bottlenecks in short-caption generation. Gains remain workload-dependent, and deployments must reserve GPU memory for decoding processes.

2026-09-19

AI 基礎設施

AIPerf Reworks LLM Benchmarking with Multiple Processes, but Streaming Metrics Still Require Token-Counting Checks

NVIDIA details AIPerf, the successor to GenAI-Perf, which brings load generation, results processing, and traffic replay into a multiprocess architecture. It offers more comprehensive measurements of inference services, but migration compatibility and streaming token counts still affect how results should be interpreted.

2026-09-19

AI 基礎設施

Exa Snapshot fixes historical web content for agent evaluations, while search ranking still uses current signals

Exa has introduced historical web content retrieval, letting agents access evidence within a specified time cutoff to reduce evaluation contamination from answers published later. The feature remains in research preview, with limits on historical access and supported search modes under the public plan.

2026-09-19

GitHub Repo

OpenHands 1.20 lets agent profiles select secrets, with server-enforced scope limits

The release adds secret allowlists to agent profiles, reducing the chance that automated tasks receive unrelated credentials. Existing configurations remain unrestricted by default, so deployers must check backend support and other credential delivery paths.

2026-09-19