4B 3D scene coder update: better one-shot results, dataset and FP16 weights coming
ConfidentDinner6648 · reddit · 2026-10-28
The author updated their 4B Qwen3.5-based 3D coding model with notably better one-shot results and visual quality. Recommended Q8 quant with temperature 0.6, top-k 20, top-p 0.95; Q4 also available but untested. Despite no vision encoder in this training run, it can generate 3D scenes from image references. Dataset and full FP16 weights to be released today; old downloads must be replaced (filenames unchanged).
Related event: Fine-Tuned Qwen3.5-4B Generates 3D Scenes with Just 4k Samples(4 posts)→
More from Models
- Researcher: GPT6 Astra brings a step change in review quality, guided prompts help further — anshulkundaje · 2026-10-29
- Gemini 4 Carbon debuts #2 on Vals Index at half the cost, tops finance and legal benchmarks — zacharynado · 2026-10-29
- Paper Instruments launches Ultramarine 1 Flash, beating frontier models at up to 60x lower cost — baseten · 2026-10-29
- Gemini 4 reportedly "proved" nearly half of the false theorems it was given — RishabJainK · 2026-10-29
- Sentdex: DeepSeek v4.1 Flash now his top local token generator ever — Sentdex · 2026-10-29
- LoopVL: 1B-parameter VLM reuses 32 layers across 128 calls, beats 4B baselines on 5/8 benchmarks — jiqizhixin · 2026-10-29