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

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 bf16 checkpoint with both transformer streams quantized to 8-bit (group 64); embeddings, lm_head, norms, vision patchify, FM embedders and the pixel head stay high-precision (<3% of bytes). Quantization: group size 64, 8-bit (mlx convention quantization block in config.json).

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 22.9 GB · 1024² ≈ 7.4 s (8-step, cfg 4)

Use it for: the memory-lean quality tier — 8-bit reproduces the bf16 image at fixed seed (cos 0.998, 33 dB): 50-step T2I, editing, VQA, think mode in under 23 GB.

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-8bit --local-dir SenseNova-U1.5-8B-MoT-8bit
.build/release/sensenova-cli --weights SenseNova-U1.5-8B-MoT-8bit \
  --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 + 8-bit group-64 quantization) 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).