SenseNova-U1.5-8B-MoT-bf16

SenseNova-U1.5-8B-MoT-bf16

Apple Silicon (MLX) artifact of SenseNova-U1.5-8B-MoT — SenseTime's unified T2I + editing + VQA flagship (NEO-unify Mixture-of-Transformers, pixel-space rectified flow, no VAE).

This artifact is the upstream checkpoint cast to bf16 (the reference implementation's own default inference dtype; the fp32-stored generation stream is storage precision).

Runtime: sensenova-u1-swift (MLX-Swift). Weights are stored in the runtime's key layout (NHWC convs; quantized linears as weight/scales/biases) and load with zero conversion transient — peak load memory ≈ resident.

Performance (M5 Max)

peak 35.1 GB · 1024² ≈ 6.6 s (8-step, cfg 4) · 2048² ≈ 40 s (8-step, cfg 4); 50-step quality tier ≈ 5× those times

Use it for: the quality tier: 50-step T2I, instruction editing, VQA, and <think> reasoning mode. 35 GB download instead of the 50 GB original.

Quick start

git clone https://github.com/xocialize/sensenova-u1-swift && cd sensenova-u1-swift && swift build -c release
hf download mlx-community/SenseNova-U1.5-8B-MoT-bf16 --local-dir SenseNova-U1.5-8B-MoT-bf16
.build/release/sensenova-cli --weights SenseNova-U1.5-8B-MoT-bf16 \
  --prompt "A cinematic mountain lake at sunrise, realistic photography." \
  --width 1024 --height 1024 --steps 50 --cfg 4.0 --out out.npy

The runtime also does instruction editing (--edit-image), VQA (--vqa), and <think> reasoning mode (--think) — see the repository README for the full surface and the parity report (component parity < 1e-4; e2e per-pass cosine 0.999+ vs the reference PyTorch implementation).

Provenance & license

• Upstream: sensenova/SenseNova-U1.5-8B-MoT @ 07d76f6, Apache-2.0, by SenseTime / SenseNova — paper · reference implementation.
• This repository is a format conversion (bf16 cast) of the upstream checkpoint, redistributed under the same Apache-2.0 license with modifications noted here. All credit for the model to the SenseNova team.
• tokenizer.json is generated from the upstream vocab.json/merges.txt (byte-identical tokenization, verified against reference ids).