optimum-intel-internal-testing/tiny-random-qwen3-vl-embedding

import torch
from transformers import AutoConfig, AutoModel, AutoProcessor


MODEL_NAME = "Qwen/Qwen3-VL-Embedding-2B"
path_tiny = "./qwen3_vl_embedding_tiny/"
config = AutoConfig.from_pretrained(MODEL_NAME)
config.text_config.intermediate_size = 10
config.text_config.hidden_size = 4
config.text_config.max_position_embeddings = 10
config.text_config.num_hidden_layers = 2
config.text_config.head_dim = 32
config.text_config.num_attention_heads = 2
config.text_config.num_key_value_heads = 2

config.vision_config.hidden_size = 64
config.vision_config.intermediate_size = 64
config.vision_config.num_heads = 4


hf_model = AutoModel.from_config(config)
hf_model = hf_model.to(torch.float32)
hf_model.save_pretrained(path_tiny)
processor = AutoProcessor.from_pretrained(MODEL_NAME)
processor.save_pretrained(path_tiny)