iCreatAI News
Back to main site
Skip to content
Only First party#DeepSeek
  1. vLLM Blog72

    DeepSeek-V4.1-Flash on Vera Rubin NVL72: 58x Throughput vs. GB300 NVL72 at 387 tokens/s

    On a single NVIDIA Vera Rubin NVL72 rack, vLLM serves DeepSeek-V4.1-Flash at 1.7M tokens/s per GPU with 387 tokens/s per user, 58x GB300 NVL72, reaching 2.52M at peak throughput and 1,810 tokens/s for a single user. The gains come from kernels tuned for Rubin's split memory domains and FP4 tensor cores, specialized low-latency decode kernels, and a rack-wide KV cache and Engram tables shared over NVLink. The post details four optimization groups and reports a 96.5–97% prefix hit rate at high concurrency with workload-aware routing.

    Why it matters: It offers a detailed, measurement-backed look at how Rubin-specific memory and MMA features translate into serving throughput and latency for a frontier open model.

  1. vLLM Blog72

    Zero Mismatch, Simple Async RL: REINFORCE Zero, DeepSeek-V4.1 and Kimi K3.1

    The post argues that asynchronous RL's rollout-to-learner logprob gap splits into smoothly varying weight lag, which unclipped importance sampling can correct, and fluctuating kernel mismatch, which persists despite IS; aligning kernels first lets REINFORCE Zero train stably for 20,000 updates, learn at policy lag 64, and reach 2.40× the training-step throughput of synchronous training with aligned kernels. It reports zero mismatch (bit-for-bit logprob agreement) for DeepSeek-V4.1-Flash and Kimi K3.1 on day 0, and for DeepSeek-V4-Flash and Kimi K3.1 on AMD MI355X. Code and recipes are in Inferact/zero-mismatch-rl, with the paper at hijkzzz/REINFORCE-Zero and training code in NVIDIA-NeMo/labs-molt.

    Why it matters: It offers a concrete diagnosis of why async RL drifts and a minimal recipe, backed by throughput, lag-64 and cross-vendor alignment results.

  1. Wiz Research78

    “Black October” for AI Developer Platforms

    At least nine AI developer platforms disclosed incidents during October, with Fireworks, OpenCode and Warp reporting indications of an “agent swarm” attack and OpenCode and CommandCode attributing theirs to attempted “distillation.” Wiz describes downstream abuse of enterprise credentials tied to one vendor breach, including two keys used for roughly 800k and 1.7 million Bedrock calls respectively, likely over $100k in usage at list price. Wiz also links the keys to resale operations at llm-routing.com and xingjiabiapi.org, and publishes IOCs including 23.106.46.239 and 38.246.114.118.

    Why it matters: Provides Wiz's technical account of LLMJacking follow-on activity and IOCs from the October platform breaches.

  1. OpenRouter Announcements62

    How to Extract Tables from a PDF with an LLM

    The guide shows how to send a PDF to a file-reading model together with a strict JSON schema describing one row, so the model returns typed rows with cleanup such as parenthesized negatives and footnote markers handled by field descriptions. In tests over five PDFs (two scans, 644 cells), every model that read the PDF directly got all cells right, while failures came from the parser route and output token limits, including two silent failures. It recommends rejecting empty results and any response whose finish_reason is "length", and comparing routes on your own PDFs before choosing one.

    Why it matters: Offers a concrete, tested recipe with code for schema-driven PDF table extraction, plus specific guidance on silent failure modes worth knowing before building a pipeline.

  2. Replit Blog60

    ViBench 2.0: A Benchmark for Long-Horizon Software Building

    Replit releases ViBench 2.0, a long-horizon software development benchmark in which an agent builds a product from an empty project and then adds up to eighteen features on top of its own code, spanning seventeen products, 241 feature requests and 544 test plans. In Fireworks' official runs, GPT-6.1 Sol leads with a pass@1 of 65%, followed by GPT-6 Astra at 60% and Opus 5.5 at 49%, and no open-weight model passes a full build. Eight of the products are public on GitHub in the Harbor format, and results can be rerun from release v2.0.6 on the OpenHands-based harness.

    Why it matters: It offers concrete pass rates, costs and failure modes for frontier models on multi-feature product building, with a public, rerunnable harness.

  1. Artificial Analysis Articles60

    Announcing Artificial Analysis Intelligence Index v5

    Artificial Analysis released Intelligence Index v5, its biggest upgrade since v4, adding FrontierCode 2.0 and Terminal-Bench-Science and upgrading HLE-Diamond, AA-Omniscience v1.1 and CritPt v2, with 60% of the Index now based on private test sets or held-out answers. Claude Opus 5.5 (max) leads at 58, ahead of GPT-6 Astra (max, 57) and GPT-6.1 Sol (max, 55), while the open weights gap widens from 3 points to 12. It also launched the Progress Index v1, which tracks model capability over time on one continuous scale.

    Why it matters: It details a major benchmark revision and new leaderboard results, useful for readers tracking how frontier and open-weight models are measured.

You have reached the end