Model CoverageVision Language Models

Qwen3.6

View as Markdown

Qwen3.6 is Alibaba Cloud’s unified vision-language model generation following Qwen3.5. NeMo AutoModel supports both the dense 27B checkpoint and the 35B-A3B MoE checkpoint.

TaskImage-Text-to-Text
ArchitectureQwen3_5ForConditionalGeneration, Qwen3_5MoeForConditionalGeneration
Parameters27B dense; 35B total / 3B active MoE
HF OrgQwen

Example Models and Recipes

ModelRecipeDescription
Qwen/Qwen3.6-27Bqwen3_6_27b.yamlFull SFT on MedPix-VQA
Qwen/Qwen3.6-27Bqwen3_6_27b_lora.yamlLoRA on MedPix-VQA
Qwen/Qwen3.6-35B-A3Bqwen3_6_35b.yamlFull SFT on MedPix-VQA
Qwen/Qwen3.6-35B-A3Bqwen3_6_35b_lora.yamlLoRA on MedPix-VQA

Run a Recipe

$uv run automodel --nproc-per-node=8 examples/vlm_finetune/qwen3_5/qwen3_6_27b.yaml

Install the vlm-media extra before running these recipes. See the Installation Guide and VLM Fine-Tuning Guide.