A Qwen3.5-4B fine-tune is getting better at generating 3D scenes
ConfidentDinner6648 · reddit · 2026-10-27
A Reddit user shares progress on their Qwen3.5-4B fine-tune trained for general coding, using Three.js scene generation as the main benchmark. Newer checkpoints show noticeably better scene quality and one-shot success rates.
- Updated Q4KM and Q80 quants fix a layer issue; available on Hugging Face
- Training finishes Oct 28 07:30 UTC, after which FP16 weights, all quantizations, the dataset, and a technical report will be released
More from Models
- ChatGPT Pro user: weekly allowance exhausted within 2 hours after using a limit reset — JBusu · 2026-10-27
- Kimi K3.1 Pro seen as an agent tradeoff play, not a benchmark winner — usamanoman · 2026-10-27
- Open-weight model poisoning and sleeper-agent risk gain attention — ianand · 2026-10-27
- Open-source plugin hard-blocks Qwen's hallucinated signed OSS URLs before agent tool calls — builderjaydub · 2026-10-27
- GPT-6.1 Terra matches GPT-6 Astra at 57.6% on Agents' Last Exam at 1/16.6x cost — aakashgupta · 2026-10-27
- Asking an AI About Leaks: First "Surely Fake", Then Slowly Convinced — johnowhitaker · 2026-10-27