Inkling Open-Source Model Released
BanghuaZ · x · 2026-07-16
SGLang shared and introduced the first open-source model release of Inkling:
- Total parameters 975B, active parameters 41B MoE
- Supports up to 1M context
- Capable of native reasoning across text, image, and audio
- Serving and RL support are simultaneously available, allowing direct execution and fine-tuning on an open-source stack
The post also highlighted its architectural and deployment optimizations:
- The new architecture includes ShortConv, attention with relative positional encoding, and shared expert sink MoE
- Inference-side optimizations include full CUDA graph and MXFP8 KV cache
- Training-side provides full-parameter and LoRA RL, ensuring training/inference consistency through a custom Megatron backend, routing replay, and cross-runtime parameter synchronization
- SGLang and Miles provide Day 0 support
More from Infra
- ClickHouse aggregates 100B rows in 0.4s with parallel replicas auto-scaling — Abhishekcur · 2026-10-18
- Auction-based LLM routing paper finds fixed token pricing wastes 10%-71% in costs — _onionesque · 2026-10-18
- Running Ling 3.1 Flash (560B MoE, 25B Active) Across 4 DGX Sparks via llama.cpp RPC — BerryBlush_Rogers691 · 2026-10-18
- SGLang Summit sponsors jump from 11 to 36 in one week — ying11231 · 2026-10-18
- Oracle CDS hits record 300bps, called first AI trade domino after Barclays sell downgrade — arthurcolle · 2026-10-18
- Agent Coding Workflow Would Cost ~$40k/Month on API Pricing, Dev Reveals — ChrisRackauckas · 2026-10-18