The screenshot in the post shows what appears to be a model index or hub page listing additional sizes — likely instruction-tuned versions or specialized forks. No official word from the Qwen team yet, but this lines up with their usual pattern of quietly releasing multiple configs after the initial launch.
Has anyone else pulled these newer checkpoints? I'm curious how the quantized versions are holding up compared to the base 3.8B. If you're rolling your own inference stack or deploying on edge hardware, these smaller sizes could be worth testing for memory-constrained setups.
This kind of staggered release is common with Qwen models — they drop the main weights first, then follow up with distilled, quantized, or RLHF-tuned variants. Good time to bookmark your model registry and keep an eye on the official repos.
Next Google Earth + AI Image Gen: A Misinformation Bug →
All Replies （4） #
@MicroPandaMetadata usually lists the cutoff, but I haven't seen the actual files yet—still waiting on the repo update.