Inkling 2 Mini Runs Locally at 210 tok/s on M5 Ultra Mac Studio
Hugging Face and Thinky Machines optimized Inkling 2 Mini for local inference, reaching 210 tok/s decode on a 96GB M5 Ultra Mac Studio and 125 tok/s on a 48GB M…
Hugging Face and Thinky Machines optimized Inkling 2 Mini for local inference, reaching 210 tok/s decode on a 96GB M5 Ultra Mac Studio and 125 tok/s on a 48GB M…
LiteLLM shows how to route model calls from Claude Code, Codex, OpenCode and Pi into Lens as traces, with per-key, per-user and per-team ownership and a termina…
Hugging Face introduced a new default Home feed where trending model releases appear directly on the logged-in homepage, each as its own dedicated card.
Hugging Face runs its own CDN since this summer, shipping almost 1 exabyte per month (~420 GB/s across 80,000 requests per second) because AI downloads break th…
Thinky Machines released Inkling 2 Mini, a 26B-parameter looped transformer under Apache 2.0 that runs on a laptop and scores 20.1 on the AA Index for agentic w…
A new US House report recognizes that American researchers, startups, nonprofits, government, students and hobbyists benefit from open-source AI regardless of o…
OpenCode Console now supports keyless authentication, letting users call models from CI or any OIDC-capable environment without long-lived API tokens.
dax from OpenCode says their experimental password manager will eventually support MCP, and that a single codemode script can already pass a credit card number…
Arena's Alignment Index places Kimi K3.1 Pro tied for #2 across labs with a score of 78.2, matching Anthropic and making Moonshot the highest-ranked Chinese lab…
Hugging Face Diffusers adds an example for LoRA fine-tuning Nvidia's RFD4-Proteina protein design model, training rank-32 adapters on the denoiser with flow-mat…