Looking for Small VLM/MLLMs Alternatives to Qwen Series Models
Posted by CatSweaty4883@reddit | LocalLLaMA | View on Reddit | 9 comments
I have tried Qwen 3 VL family of models on my rtx3060, max I can load is Q8 8b. The task is visual reasoning/ instruction following. What are some other models I could explore? My system ram is 16gb, vram 12gb.
wardino20@reddit
just switch to qwen 3.5
pmttyji@reddit
Qwen3.5-9B
CatSweaty4883@reddit (OP)
Are 3.5 models good in following CoT instructions? I didn’t have a good time trying 4b
pmttyji@reddit
Hope you're trying with expected values for the parameters. Better take it from Model card(it has best practices section too). And Unsloth has a page for same.
https://huggingface.co/Qwen/Qwen3.5-9B
https://unsloth.ai/docs/models/qwen3.5
pop0ng@reddit
Try Gemma4-e4b
CatSweaty4883@reddit (OP)
Will look into it!
FatheredPuma81@reddit
Why use old model?
Someone needs to make a bot that auto responds to that stuff.
CatSweaty4883@reddit (OP)
I needed my models to follow CoT instructions, hence was using qwen3 (which were good) but need to spread wider. Hence was looking for more models to try
Deep-Vermicelli-4591@reddit
Qwen 3.5 9B or Gemma 4 E4B