TensorFold 1.2.10 ships: Qwen3.8-27B decodes 57% faster on GB10
AIFlow_ML · x · 2026-11-19
TensorFold community released v1.2.10 with notable inference gains:
- Qwen3.8-27B decodes 57% faster on a GB10 at 8 streams with a shared 32K prefix, with identical outputs
- DeepSeek-V4.1 cold prompts are 5–7% faster on an M5 Ultra
- New features: background responses, tensorfold launch and tensorfold rm
- Qwen3-Embedding and Unlimited-OCR-2 now supported on Metal
Several bugs were squashed, with credits to community members who tested on their hardware.
More from Infra
- NVIDIA pitches AI factory digital twins: test designs before construction, boost tokens per watt — luisceze · 2026-11-19
- NYT: Singapore Firm Megaspeed Bought $4B+ in Nvidia Chips Within Reach of China's Military — Gregory_C_Allen · 2026-11-19
- Musk: Some of His Best Data Center Builders Came From In-N-Out, Trained in Days — yunta_tsai · 2026-11-19
- Model Factory launch echoed with Musk's quote: what matters is the machine that builds the machine — eliebakouch · 2026-11-19
- NVIDIA's Physical AI stack: Hyperion picks Aeva as reference Lidar for auto and robotics — Leather_Landscape_20 · 2026-11-19
- SpaceX said to be building AI compute faster than anyone, eyeing frontier model next year — PTrubey · 2026-11-19