Qwen3.5-4B — OpenVINO IR (Split-IR Layer-Wise)
OpenVINO split-IR layer-wise intermediate-representation export of Qwen/Qwen3.5-4B.
Converted on 2026-06-14 with dsainvg/openvino-model-conv
via qwen35/scripts/convert_to_openvino.py — layer-by-layer conversion.
Load with QwenPipeline
from qwen_npu_pipeline import QwenPipeline
pipe = QwenPipeline("path/to/model/directory")
print(pipe.generate("Hello, my name is", max_new_tokens=64))
Files
| File | Description |
|---|---|
embed_tokens.xml / .bin |
Embedding layer (FP16) |
layer_*.xml / .bin |
Stateful Transformer decoder layers (INT4) |
lm_head.xml / .bin |
LM head & normalization layer (INT4) |
tokenizer.json, tokenizer_config.json |
Tokeniser assets |
config.json, generation_config.json |
Model config |
- Downloads last month
- 19