avlp12 / Lance-3B-MLX-Traced

Lance-3B-MLX-Traced (image weights)

ByteDance Lance 3B image weights converted to Apple MLX , byte-clean against the original PyTorch.

The weight is not the point — the verification is

This model.safetensors is bit-identical (SHA256 5ede2f0a376096d51004dad85705938abbfe3cedb64e212631527e76b9547817 ) with RockTalk/Lance-3B-MLX . Both follow the same deterministic, bf16-preserving conversion path from bytedance-research/Lance — so the weights should match, and they do.

The differentiator of this release is not the weight. It is the verification trace : every stage of the port was cross-validated against the original PyTorch via byte-diff before the next stage started, and the full harness + lesson log is public:

👉 github.com/avlp12/lance_alis_mlx

This is complementary to the other MLX ports (RockTalk, xocialize, mlx-community) — same weights, but the verification process is shown.

Verification (image track)

Every gate uses original PyTorch direct import (not a clean re-implementation) under a shim, same PRNG (NumPy) on both sides, byte-diff at every layer. 23 lessons distilled across stages 1–9; see the repo's LEARNING_LOG/ .

(Video weights — STAGE 8 VAE + STAGE 9 t2v, all verified — are a separate release. This repo is the image track.)

Usage

Inference is pure MLX — no PyTorch needed at runtime. (PyTorch is only imported by the verification harnesses in tools/ .)

git clone https://github.com/avlp12/lance_alis_mlx
cd lance_alis_mlx
python3.12 -m venv .venv && source .venv/bin/activate
pip install mlx mlx-vlm transformers safetensors einops pillow huggingface_hub numpy

# this weight
hf download avlp12/Lance-3B-MLX-Traced --local-dir checkpoints/Lance-3B-MLX
hf download RockTalk/Wan2.2-VAE-MLX    --local-dir checkpoints/Wan2.2-VAE-MLX

# generate
PYTHONPATH=. .venv/bin/python tools/stage6_t2i_smoke.py    # text-to-image
PYTHONPATH=. .venv/bin/python tools/stage7_x2t_smoke.py    # image-to-text
PYTHONPATH=. .venv/bin/python tools/stage7_ti2i_smoke.py   # image edit

Apple Silicon required (developed on M3 Ultra). Python 3.12.

License & citation

Apache 2.0 — same as upstream ByteDance Lance and Alibaba Wan 2.2 VAE.

@misc{fu2026lanceunifiedmultimodalmodeling,
      title         = {Lance: Unified Multimodal Modeling by Multi-Task Synergy},
      author        = {Fengyi Fu and Mengqi Huang and Shaojin Wu and Yunsheng Jiang and Yufei Huo and Hao Li and Yinghang Song and Fei Ding and Jianzhu Guo and Qian He and Zheren Fu and Zhendong Mao and Yongdong Zhang},
      year          = {2026},
      eprint        = {2605.18678},
      archivePrefix = {arXiv},
      primaryClass  = {cs.CV},
      url           = {https://arxiv.org/abs/2605.18678},
}

Acknowledgments

• ByteDance Lance team — original PyTorch model and research
• RockTalk — MLX checkpoint conversion used as the parity reference
• Alibaba Wan 2.2 team — 3D Causal VAE architecture