import numpy as np
import torch
from PIL import Image
from transformers import AutoModel, AutoTokenizer

model_id = "tiny-random/minicpm-v-4_5"
model = AutoModel.from_pretrained(model_id, trust_remote_code=True,
                                  attn_implementation='sdpa', torch_dtype=torch.bfloat16)
model = model.eval().cuda()
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)

image = Image.fromarray(np.random.randint(0, 255, (224, 224, 3), dtype=np.uint8), 'RGB')
question = "What is the landform in the picture?"
msgs = [{'role': 'user', 'content': [image, question]}]
answer = model.chat(
    msgs=msgs,
    image=image,
    tokenizer=tokenizer,
    max_new_tokens=32,
)
print(answer)

# Second round chat, pass history context of multi-turn conversation
msgs.append({"role": "assistant", "content": [answer]})
msgs.append({"role": "user", "content": [
            "What should I pay attention to when traveling here?"]})
answer = model.chat(
    msgs=msgs,
    image=None,
    tokenizer=tokenizer,
    max_new_tokens=32,
)
print(answer)

Codes to create this repo:

python

import json
from pathlib import Path

import accelerate
import torch
from huggingface_hub import hf_hub_download
from transformers import (
    AutoConfig,
    AutoModel,
    AutoModelForCausalLM,
    AutoProcessor,
    AutoTokenizer,
    GenerationConfig,
    set_seed,
)

source_model_id = "openbmb/MiniCPM-V-4_5"
save_folder = "/tmp/tiny-random/minicpm-v-4_5"

processor = AutoProcessor.from_pretrained(source_model_id, trust_remote_code=True)
processor.save_pretrained(save_folder)

with open(hf_hub_download(source_model_id, filename='config.json', repo_type='model',), 'r', encoding='utf-8') as f:
    config_json = json.load(f)
for k, v in config_json['auto_map'].items():
    config_json['auto_map'][k] = f'{source_model_id}--{v}'
automap = config_json['auto_map']

config_json['head_dim'] = 32
config_json["hidden_size"] = 128  # required by Sampler -- num_heads=embed_dim // 128
config_json['intermediate_size'] = 128
config_json['num_attention_heads'] = 2
config_json['num_key_value_heads'] = 1
config_json['num_hidden_layers'] = 2
config_json['tie_word_embeddings'] = True

# factor = config_json['rope_scaling']['long_factor']
# config_json['rope_scaling']['long_factor'] = factor[:16]
# config_json['rope_scaling']['short_factor'] = factor[:16]

config_json['vision_config']['intermediate_size'] = 128
config_json['vision_config']['hidden_size'] = 64
config_json['vision_config']['num_attention_heads'] = 2
config_json['vision_config']['num_hidden_layers'] = 2

with open(f"{save_folder}/config.json", "w", encoding='utf-8') as f:
    json.dump(config_json, f, indent=2)

config = AutoConfig.from_pretrained(
    save_folder,
    trust_remote_code=True,
)
print(config)
torch.set_default_dtype(torch.bfloat16)
model = AutoModel.from_config(config, trust_remote_code=True)
torch.set_default_dtype(torch.float32)
model.generation_config = GenerationConfig.from_pretrained(
    source_model_id, trust_remote_code=True,
)
set_seed(42)
num_params = sum(p.numel() for p in model.parameters())
with torch.no_grad():
    for name, p in sorted(model.named_parameters()):
        torch.nn.init.normal_(p, 0, 0.1)
        print(name, p.shape, p.dtype, p.device, f'{p.numel() / num_params * 100: .2f}%')
        pass
model.save_pretrained(save_folder)

def modify_automap(path, source_model_id):
    import json
    with open(path, 'r', encoding='utf-8') as f:
        content = json.load(f)
    automap = {}
    if content.get('auto_map', None) is not None:
        for key, value in content.get('auto_map').items():
            if isinstance(value, str):
                value = source_model_id + '--' + value.split('--')[-1]
            else:
                value = [(source_model_id + '--' + v.split('--')[-1]) for v in value]
            automap[key] = value
        with open(path, 'w', encoding='utf-8') as f:
            json.dump({**content, 'auto_map': automap}, f, indent=2)

modify_automap(f"{save_folder}/config.json", source_model_id)
modify_automap(f'{save_folder}/processor_config.json', source_model_id)
modify_automap(f'{save_folder}/preprocessor_config.json', source_model_id)
modify_automap(f'{save_folder}/tokenizer_config.json', source_model_id)
for f in Path(save_folder).glob('*.py'):
    f.unlink()