NVIDIA and GitHub update AI infrastructure #121
Today's Letter
NVIDIA, Qwen3.8-Flash-Next on GB300 NVL72
- NVIDIA validated Alibaba’s Qwen3.8-Flash-Next on the GB300 NVL72 for agentic coding workloads
- The multimodal MoE model has 125B parameters, 6B activated per token, and a 262,144-token native context window
- YaRN extends the context window to 1M tokens
- Its architecture combines three Gated DeltaNet layers with one Qwen Sparse Attention layer per cycle
- Alibaba benchmarks report up to 7.6x faster prefill and 4.9x faster decoding than full attention
- NVIDIA reports more than 16K tokens per second per GPU and more than 200 tokens per second per user on GB300 NVL72
- Deployment recipes are available for SGLang, vLLM, TokenSpeed, and TensorRT-LLM, with fine-tuning support through NeMo AutoModel and NeMo RL
Source: developer.nvidia.com
GitHub, enterprise plugin marketplace auto-update
- Enterprise-managed settings can now enable automatic updates for individual plugin marketplaces
- Administrators set `autoUpdate: true` on an `extraKnownMarketplaces` entry
- Supported clients periodically check the marketplace and update installed plugins
- The marketplace must remain permitted by the effective `strictKnownMarketplaces` allowlist
- Generally available with Copilot Business and Copilot Enterprise in the GitHub Copilot app, Copilot CLI, and Visual Studio Code
- Announced in the GitHub Changelog on August 26, 2026
Source: github.blog
Jocoletter curates AI, software, and product trends for developers and builders.
#GitHub #NVIDIA