GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
HY-2012
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
HY-2012
Model Tree
agentscope-ai/QwenPaw-Flash-2B
Input Modalities
Output Modalities
Supported Functionality
qweight / qzeros / scales)model.visual.*, 24 blocks) is kept in bf16.| File | Description |
|---|---|
model.safetensors | GPTQ-Int4 language model + bf16 vision encoder |
config.json | Model config with quantization_config |
quantize_config.json / quant_log.csv | GPTQ quantization config and per-module loss log |
tokenizer.json / tokenizer_config.json / chat_template.jinja | Tokenizer |
processor_config.json | Vision processor (Qwen3VLProcessor) |
generation_config.json | Generation defaults |
With GPTQModel:
python
from gptqmodel import GPTQModel model = GPTQModel.load("HY-2012/QwenPaw-Flash-2B-GPTQ-int4")With vLLM (Qwen3.5 architecture requires a recent vLLM):
shell
vllm serve HY-2012/QwenPaw-Flash-2B-GPTQ-int4 --port 8000 \ --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_xmlfrom gptqmodel import GPTQModel model = GPTQModel.load("HY-2012/QwenPaw-Flash-2B-GPTQ-int4")vllm serve HY-2012/QwenPaw-Flash-2B-GPTQ-int4 --port 8000 \ --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_xml