From the model card
Fields below are copied from the tags and counters on the HuggingFace repository cyankiwi/Qwen3.5-9B-AWQ-4bit at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.
- Publisher (HF namespace)
- cyankiwi
- Pipeline tag
- image-text-to-text
- Library
- Transformers
- Weight formats
- safetensors
- License tag
apache-2.0— read the license file in the repo before relying on it- Lineage
-
- base model Qwen/Qwen3.5-9B
- quantized from Qwen/Qwen3.5-9B
- Downloads (HF counter at last fetch)
- 363,295
- Likes (HF counter at last fetch)
- 36
- Model card
- https://huggingface.co/cyankiwi/Qwen3.5-9B-AWQ-4bit
Use cases
- Multimodal image-text inference on mid-range consumer GPUs
- vLLM-hosted vision-language assistant
- Visual question answering without cloud dependency
- Comparative evaluation of dense vs MoE Qwen3.5 variants
Pros
- 9B AWQ INT4 fits in ~7 GB VRAM — broad consumer GPU compatibility
- Multimodal image-text capability in a self-hostable package
- Apache-2.0 license
- compressed-tensors for clean vLLM integration
Cons
- Community quantization — no linked accuracy regression report
- Vision tasks are more sensitive to INT4 quantization than pure text
- Dense 9B has higher active FLOP count than MoE alternatives at same param count
- No GGUF variant from this uploader for llama.cpp users
Tags
transformerssafetensorsqwen3_5image-text-to-textconversationalbase_model:Qwen/Qwen3.5-9Bbase_model:quantized:Qwen/Qwen3.5-9Blicense:apache-2.0endpoints_compatiblecompressed-tensorsregion:us