tensorfold

Imagination Studio by keys+Wasserman powered by TensorFold: an AI film studio for Mac and NVIDIA DGX Spark. Plan a film like a crew (Slate workflow by Sam Wasserman), render Qwen-Image-2.1 / FLUX3-K pictures and MiniMax H3 / FastH3 video with sound on TensorFold.

ReAlign models instantly, on demand. Per-request alignment for verified users and channels, or Always On. Use directly or with Hermes, OpenClaw and custom harnesses.

DeepSeek-V4.1-Flash-1017 EXL3 2.9bpw on 2x NVIDIA DGX Spark (GB10) with TensorFold: 2.1-2.3x the vLLM kit's decode speed.

DeepSeek 1017 EXL3 on 3-4 DGX Sparks. TP4 4.0.2: 4,264 cold prefill tok/s; C16 code 320.1 / prose 285.4 native decode sum. Smart KV cache controller + Smart Adaptive speculation. C16/R48, 9M KV, 3.5 GiB checkpoints. TP3 1.2.8 (+58% cold prefill) preserved. Exact recipes, source and benchmarks.

GLM-5.3-Flash BIZ for DGX Spark-class GB10 systems. 3.x serves the XSFP4 checkpoint or NVIDIA's pinned NVFP4 one with TensorFold's native Zig/CUDA engine (TP=2 or TP=3, MTP drafts equal serial, tool calls, structured output, images, video). Apache-2.0 code, weights fetched separately. BIZ = business-use intent, not support or certification.

Imagination Studio by Keys (x.com/u1tra_instinct) + Sam Wasserman / Wasserman Productions (wassermanproductions.com · x.com/SamJWasserman), powered by TensorFold: a local all-in-one AI film studio for NVIDIA DGX Spark and Apple silicon, with Wasserman's Filmmaker Suite built in.

DeepSeek-V4.1-Flash-1017 on one 256 GB M5 Ultra: 1M context, 70.7-104.9 tok/s decode with DSpark drafts, 1,739-1,867 tok/s cold prompts (8k-131k) with int8 x int8 prompt experts, a 1M prompt in 740 s, images, tools, JSON. TensorFold native Metal.

MiniMax H3 audio-video model on TensorFold NVFP4 kernels: a faster ComfyUI loader for one RTX 50-series GPU (2.3x stock at 20 steps, 81 s per 5 s 1344x768 video with Turbo + sparse attention on an RTX 5070 Ti)

GLM-5.3-Flash EXL3 on 4x RTX PRO 6000 with TensorFold: one-command recipe and engine patches. By Aevonix Research in collaboration with Mia's AI Lab.

Qwen3.8-Flash-Next EXL3 3.05bpw on ONE DGX Spark (GB10): ~2.4x prefill (700->1777 t/s), prompt-lookup drafter, native vision/video, 262k ctx. Unofficial TensorFold v0.6.5 fork, 18 patches, byte-exact receipts.

DeepSeek-V4.1-Flash-1017 on two NVIDIA DGX Sparks (GB10) with TensorFold TP2: scripts, chat-template fix, speed and quality results

DeepSeek Flash on 1-2 NVIDIA DGX Sparks, measured honestly: DeepSeek-V4.1-Flash-1017 on TensorFold TP2 (91.5 tok/s C1, 98% quality, 1M context), V4-Flash-0731 certified numbers, the laws, the harness, every CSV

ThinkingCap Kearuga EXL3: TandemLLM-first DGX Spark serving, independent RTX 5090 receipts and a retained TensorFold reproduction recipe.

GLM-5.3-Flash EXL3 on 2x RTX PRO 6000 with TensorFold: one-command recipe and engine patches. By Aevonix Research in collaboration with Mia's AI Lab.

GLM-5.3 EXL3 on 4x RTX PRO 6000 with TensorFold: one-command recipe and engine patches. By Aevonix Research in collaboration with Mia's AI Lab.

Live dashboard for vLLM and TensorFold servers, with NVIDIA and AMD GPU and host metrics. One Go binary for one node or a fleet.

Kolibri-1 (Aleph Alpha) on two NVIDIA DGX Sparks with TensorFold: ~2x decode tok/s vs one Spark. Recipe, patches, measured results.

TensorFold deployment for GLM-5.3-Flash on 4x NVIDIA CMP 170HX (GA100 sm_80): bit-exact 4-rank TP, shm all-gather, W4A16 affine - complete tested solution (199.4 tok/s structured, exactness maintained)

Keep every AI model on a NAS; insert any of them on your GPU box with one click (Ollama / vLLM) — never evicts a model you didn't name. Stdlib Python, demo mode, web UI.

Qwen3.8 Flash Next on one NVIDIA Jetson AGX Thor with TensorFold: recipe, gated benchmarks, and a head-to-head with a DGX Spark