硬體

AI 晶片、GPU、資料中心、邊緣裝置與硬體安全。

Salesforce Koa Generates Training Environments from Workflow Specifications, Teaching a 120B Model to Operate CRM Tools

Using Nemotron 3 Super as its foundation, Salesforce trained Koa, its first CRM reasoning model, through synthetic enterprise scenarios, SFT, and GRPO. It delivers clear improvements in multi-turn tool use, but the paper acknowledges that its overall performance still trails the strongest frontier models, while the official claim of “three times fewer errors” comes from Salesforce’s own CRM benchmark.

2026-09-16

代理工具

FlashREINFORCE Replaces Same-Prompt Group Sampling in Agentic RL With Asynchronous Single-Trajectory Updates

NVIDIA researchers propose a critic-free, single-rollout training method that lets tool-use trajectories of varying lengths enter the update process as soon as they finish. In small-model experiments, it outperformed the reported GRPO baseline using half as many rollouts, but its stability across environments and model scales remains insufficiently validated.

2026-09-16

最新模型

Gensyn open-1b Builds a Cross-Hardware, Bit-Level Audit Trail for 80,957 Training Steps

Gensyn has released the data, checkpoints, step-by-step hashes, and replay tools for its 1.61B-parameter model, allowing third parties to spot-check whether training states match its commitments. This reproducibility comes at the cost of roughly a fivefold reduction in training performance, and matching hashes alone still cannot prove that the original cluster actually ran as claimed.

2026-09-16

最新模型

Gemini 3.8 Live Moves Reasoning into the Voice Background, So Clients Can No Longer Rely on turnComplete to Determine Idleness

Google has launched two native-audio models. Extended Thinking can perform background reasoning and asynchronous tool calls while continuously streaming audio. The new protocol improves concurrency for long-running voice agents, but also requires developers to rewrite session state machines and interruption handling.

2026-09-16

推論系統

Perplexity Brings Its Local Agent to Windows RTX, but “Data Stays on the Device” Still Has Cloud Exceptions

Portable Computer can now run models, tools, and sandboxes on Windows RTX PCs with at least 24GB of VRAM, with most local work not counting against cloud quotas. However, the system can still escalate to cloud models with user authorization, while connectors and external actions mean it should not be treated as a fully offline environment.

2026-09-15

AI infrastructure

First Rubin NVL72 AgentX Benchmarks Released, but Peak Tokens-per-Watt Gains Still Depend on the Baseline

SemiAnalysis tested the Vera Rubin NVL72 using pre-release software and reported up to roughly seven times the token throughput per megawatt of Blackwell. The results highlight the role of HBM4 and NVLink 6 in long-context MoE inference, but the headline 67-fold price-performance gain incorporates rental pricing and assumptions about specific operating points.

2026-09-15

最新模型

Atria Dawn Preview Releases 744B MoE Weights, Trained for Long-Horizon Research Agents in Verifiable Environments

The Shanghai AI Laboratory has released Atria Dawn Preview, along with FP8 weights and inference recipes, positioning it as a full-loop agent capable of searching, coding, and running experiments. The organization claims the best reported results on five of 16 benchmarks, though model scale, test-time compute, and its in-house evaluation setup still limit comparability.

2026-09-15

代理工具

Otis Unifies llama.cpp, Local Model Selection, and Agent Permissions in a Single Desktop and Terminal Interface

In a concentrated public promotion push on September 13–14, Otis recommends and downloads GGUF models based on the user’s hardware, then launches a local agent using its bundled llama.cpp. Offline mode requires no account, but search, PAIR clusters, and Fireworks inference operate across distinct data and trust boundaries.

2026-09-14

推論系統

llama.cpp Mainline Adds Maple 20B-A1B Ternary MoE, With Initial Support Targeting CPUs

Architecture support merged on September 14 allows Maple-Preview to be loaded and converted without relying on DeepGrove’s llama.cpp fork. The model activates only about 1 billion parameters per token, but GPU backends, quality evaluations, and official speed claims still require separate verification.

2026-09-14

推論系統

smolperfbenchmark Launches On-Device LLM Energy-Efficiency Tests, but Its Live Leaderboard Currently Shows Zero Results

The new open-source benchmark runs small models on Jetson devices, Macs, Raspberry Pis, and Android devices, measuring decoding speed, tokens per joule, latency, and temperature instead of focusing solely on data-center GPU throughput. Its creator says roughly 1,000 Jetson configurations have been tested, but the public leaderboard currently loads no data, leaving users to verify results through repository reports and raw artifacts.

2026-09-14

GitHub Repo

llama.cpp Daily Builds Fall Back to F32 on Older GPUs and Prevent Divide-by-Zero Crashes in Nemotron MTP Weights

The September 13–14 llama.cpp prerelease builds add a CUDA F32 path for systems where BF16 is unavailable, preventing outright failures on some older GPUs. Another fix reports an error when Nemotron-H NextN/MTP metadata is incomplete instead of triggering SIGFPE while loading the model.

2026-09-14

推論系統

Colibrì Tiers Trillion-Scale MoE Weights Across VRAM, RAM, and NVMe, but Low Memory Usage Does Not Mean Fast Inference

Colibrì, which has recently gained traction in the community, uses a pure C engine to load sparse experts on demand, enabling models with 744B to 2.8T parameters to run on non-data-center hardware. What it lowers is the resident-memory requirement; actual decoding speed remains heavily constrained by SSD bandwidth, expert hit rates, and quantization quality.

2026-09-14