LingBot Video Dense 1.3B SDNQ UINT4 Static
This is a complete, loadable derivative of robbyant/lingbot-video-dense-1.3b with the diffusion transformer stored using static SDNQ UINT4 weights. It is tied to source model revision f9789a7d9b4772a47aba62d4eb5282ddefd1da21, LingBot Video code a2bb04b78edd848500dc27a26e035a95442ae186, and SDNQ d841c383ff7be38728d4df829e17af4f15d4fd66 (v0.2.1-17-gd841c38).
Text encoder, tokenizer/processor, scheduler, and VAE remain at their upstream precision. Convolutions and embeddings are not quantized. The recipe is uint4-static-transformer-only-with-3d-expert-adapter: weights_dtype=uint4, auto group size (group_size=0), no dynamic quantization, SVD, Hadamard transform, convolution quantization, or embedding quantization.
Coverage
Coverage is calculated from the original parameter inventory, not from model-file sizes.
| Component | Total logical params | Quantized params | Original parameter bytes covered | Packed storage | Packed grouped experts |
|---|---|---|---|---|---|
transformer | 1,357,171,264 | 97.1278% | 94.4287% | 0.69 GiB | n/a |
This Dense checkpoint has no grouped-expert tensors. Exact module-level coverage and unquantized tensors are in quantization_manifest.json and benchmark/coverage.
Reproducible base benchmark
All five pairs use identical prompts, negative prompt, seeds 4201-4205, scheduler inputs, 832x480 dimensions, 73 frames, 24 fps, 40 steps, guidance 3.0, shift 3.0, batch_cfg=False, and null_cond_clone_zero=False. Resources were sampled every 250 ms from /proc, psutil, and nvidia-smi.
| Variant | Load (s) | Cold generation (s) | Hot mean (s) | Peak VRAM (MiB) | Peak Torch allocated (MiB) | Process RSS (GiB) | System RAM used (GiB) |
|---|---|---|---|---|---|---|---|
| Original BF16 | 40.39 | 59.35 | 58.64 | 27122 | 20698 | 2.80 | 85.31 |
| SDNQ UINT4 | 36.70 | 59.92 | 59.22 | 24674 | 18888 | 2.83 | 135.90 |
Observed base peak-VRAM reduction: 9.03%. Timing and memory are measurements on the environment recorded in benchmark/environment, not universal performance claims.
Frame-aligned aggregate quality across the five pairs: MAE 0.082366, RMSE 0.131955, PSNR 18.048 dB, SSIM 0.618208, LPIPS-Alex 0.390647.
The complete contact sheets and side-by-side MP4s are under assets/comparison/base. Quantized sample MP4s are under samples/base. Raw per-prompt CSV/JSONL, resource samples, commands, ffprobe records, output sizes, and SHA-256 values are under benchmark.
Installation and load
Use the exact pinned dependencies shipped with the repository:
git clone https://huggingface.co/WaveCut/LingBot-Video-Dense-1.3B-SDNQ-uint4-static
cd LingBot-Video-Dense-1.3B-SDNQ-uint4-static
python -m pip install -r runtime-requirements.txt
The repository includes the runtime adapter; no unmerged LingBot branch or local hidden file is needed:
import sys
from huggingface_hub import snapshot_download
root = snapshot_download("WaveCut/LingBot-Video-Dense-1.3B-SDNQ-uint4-static")
sys.path.insert(0, root)
from lingbot_sdnq_runtime import load_pipeline
pipe = load_pipeline(root, device="cuda")
# For the MoE refiner: load_pipeline(root, transformer_subfolder="refiner", device="cuda")
See prompts.json for the exact A/B inputs and benchmark/summary.json for portable metrics. Recorded consumer/offload smoke artifacts: benchmark/smokes/dense-sdnq-model.json, benchmark/smokes/dense-sdnq-model.mp4, benchmark/smokes/dense-sdnq-sequential.json, benchmark/smokes/dense-sdnq-sequential.mp4, benchmark/smokes/dense-sdnq-standard.json, benchmark/smokes/dense-sdnq-standard.mp4.
Runtime behavior and limitations
- Generic SDNQ Linear layers use eager BF16 dequantization followed by
F.linearin the tested Torch 2.8/CUDA 12.8 environment because the current SDNQ Triton quantized-matmul path is incompatible there. - Packed MoE experts are dequantized for each expert call and executed by the pinned SGLang Triton fused-MoE path. The adapter does not keep a persistent BF16 expert-weight cache.
- Static UINT4 materially changes generated pixels. Inspect the published matrices and per-prompt metrics before choosing this derivative for quality-sensitive work.
- Peak residency and speed depend strongly on resolution, frame count, attention backend, offload mode, and GPU. The numbers above describe the exact recorded B200 run only.
- The Apache-2.0 upstream license is retained. Users remain responsible for evaluating generated content for their application.
Evidence map
quantization_manifest.json: recipe, revisions, per-component and expert coverage.prompts.json: exact structured prompts, negative prompt, seeds, and generation settings.benchmark/summary.json: portable aggregate benchmark record.benchmark/base: unmodified original and SDNQ raw metrics and resource samples.benchmark/comparison: frame-aligned MAE/RMSE/PSNR/SSIM/LPIPS records.SHA256SUMS: hashes for all published files, including model shards.