TensorFold 1.0.12 runs 120B vision model on 1-2 DGX Sparks, adds Mac image input
HankYeomans · x · 2026-10-16
Open-source local inference framework TensorFold shipped v1.0.12, focused on running multimodal models on desktop hardware:
- Nemotron 3.5 Super VL 120B now serves on CUDA with one or two DGX Sparks (expert weights split across ranks; outputs verified token-identical to single-GPU), using NVIDIA's native FP4/FP8 math
- Macs gain image input for Qwen3.8-27B and GLM-5.3-Flash, plus Nemotron oQ mixed-width checkpoints
- LiquidAI's LFM2.5-24B-A2B serves at 4/6/8 bits; GLM-5.3-Flash NVFP4 previews for Sparks and Intel Arc via source builds (AMD previews too)
- New opt-in Recall lets drafts reuse previously-read text; Living Weights can undo learned changes; release gate confirmed token-identical outputs vs 1.0.11
More from Infra
- MiniCPM-V 4.5 MoE reportedly hits 13+ tokens/s on iPhone with just 3GB memory — heyshrutimishra · 2026-10-16
- Solo-built Vulkan inference engine LaForge hits ComfyUI-level SD1.5 speed: 512px in 0.8s, 402MB container — Ichidan256 · 2026-10-16
- Trading 2 RTX 5090s for 4 R9700s: does doubled VRAM pay off? — poofph · 2026-10-16
- Running PolyStrata + Cline coding agent on a $0-fancy laptop at ~90 tok/s — rpwoerk · 2026-10-16
- vLLM brings frontier model local deployment to NVIDIA RTX Spark PCs — vllm_project · 2026-10-16
- NVIDIA signals $2B in new European AI deals at GTC Berlin, totaling ~$10B in a year — CurieuxExplorer · 2026-10-16