01引言
Janus-Pro是DeepSeek最新开源的多模态模型，是一种新颖的自回归框架，统一了多模态理解和生成。通过将视觉编码解耦为独立的路径，同时仍然使用单一的、统一的变压器架构进行处理，该框架解决了先前方法的局限性。这种解耦不仅缓解了视觉编码器在理解和生成中的角色冲突，还增强了框架的灵活性。Janus-Pro 超过了以前的统一模型，并且匹配或超过了特定任务模型的性能。Janus-Pro 的简洁性、高灵活性和有效性使其成为下一代统一多模态模型的强大候选者。
代码链接：
https://github.com/deepseek-ai/Janus
模型链接：
https://modelscope.cn/collections/Janus-Pro-0f5e48f6b96047
体验页面：
https://modelscope.cn/studios/AI-ModelScope/Janus-Pro-7B
Janus-Pro 是一个统一的理解和生成 MLLM，它将视觉编码解耦以支持多模态理解和生成。Janus-Pro 基于 DeepSeek-LLM-1.5b-base/DeepSeek-LLM-7b-base 构建。
对于多模态理解，它使用 SigLIP-L 作为视觉编码器，支持 384 x 384 图像输入。对于图像生成，Janus-Pro 使用来自LlamaGen的分词器，降采样率为 16。
02模型效果
图片生成
简单 Prompt
an apple |
a car |
a dog |
|
|
|
复杂 Prompt
a bear standing on a car, sunset, winter |
a boy and a girl, the boy stands at the left side, the boy wears a red t-shirt and blue pants, the girl wears a green t-shirt and pink pants. |
the apple is in the box, the box is on the chair, the chair is on the desk, the desk is in the room |
|
|
|
颜色可以做到分开控制
多风格
Chinese ink painting, a girl, long hair, colorful hair, shining eyes |
oil painting, a girl, long hair, colorful hair, shining eyes |
anime, a girl, long hair, colorful hair, shining eyes |
|
|
|
图片理解
识物
提问：
这道菜是什么？
回答：
这道菜看起来是一道中式炖菜，可能是“牛杂汤”或“牛杂火锅”。它通常由牛的各个部位的肉和内脏组成，如牛肚、牛肠、牛筋等。这些食材经过炖煮，味道浓郁，口感丰富。牛杂汤通常会加入一些香料和调味料，如八角、桂皮、花椒等，以增加风味。这道菜不仅味道鲜美，而且营养丰富，富含蛋白质和多种微量元素。
OCR&公式转换
提问：
Convert the formula into latex code.
回答：
Here is the formula converted into LaTeX code:
\[
A_n = a_0 \left[ 1 + \frac{3}{4} \sum_{k=1}^{n} \left( \frac{4}{9} \right)^k \right]
\]
03模型推理
安装依赖
pip install git+https://github.com/deepseek-ai/Janus
多模态理解：
import torch from transformers import AutoModelForCausalLM from janus.models import MultiModalityCausalLM, VLChatProcessor from janus.utils.io import load_pil_images from modelscope import snapshot_download # specify the path to the model model_path = snapshot_download("deepseek-ai/Janus-Pro-7B") vl_chat_processor: VLChatProcessor = VLChatProcessor.from_pretrained(model_path) tokenizer = vl_chat_processor.tokenizer vl_gpt: MultiModalityCausalLM = AutoModelForCausalLM.from_pretrained( model_path, trust_remote_code=True ) vl_gpt = vl_gpt.to(torch.bfloat16).cuda().eval() question = "discribe the image" image = "/mnt/workspace/Janus/images/doge.png" conversation = [ { "role": "<|User|>", "content": f"<image_placeholder>\n{question}", "images": [image], }, {"role": "<|Assistant|>", "content": ""}, ] # load images and prepare for inputs pil_images = load_pil_images(conversation) prepare_inputs = vl_chat_processor( conversations=conversation, images=pil_images, force_batchify=True ).to(vl_gpt.device) # # run image encoder to get the image embeddings inputs_embeds = vl_gpt.prepare_inputs_embeds(**prepare_inputs) # # run the model to get the response outputs = vl_gpt.language_model.generate( inputs_embeds=inputs_embeds, attention_mask=prepare_inputs.attention_mask, pad_token_id=tokenizer.eos_token_id, bos_token_id=tokenizer.bos_token_id, eos_token_id=tokenizer.eos_token_id, max_new_tokens=512, do_sample=False, use_cache=True, ) answer = tokenizer.decode(outputs[0].cpu().tolist(), skip_special_tokens=True) print(f"{prepare_inputs['sft_format'][0]}", answer)
多模态生成：
import os import PIL.Image import torch import numpy as np from transformers import AutoModelForCausalLM from janus.models import MultiModalityCausalLM, VLChatProcessor from modelscope import snapshot_download # specify the path to the model model_path = snapshot_download("deepseek-ai/Janus-Pro-7B") vl_chat_processor: VLChatProcessor = VLChatProcessor.from_pretrained(model_path) tokenizer = vl_chat_processor.tokenizer vl_gpt: MultiModalityCausalLM = AutoModelForCausalLM.from_pretrained( model_path, trust_remote_code=True ) vl_gpt = vl_gpt.to(torch.bfloat16).cuda().eval() conversation = [ { "role": "<|User|>", "content": "A stunning princess from kabul in red, white traditional clothing, blue eyes, brown hair", }, {"role": "<|Assistant|>", "content": ""}, ] sft_format = vl_chat_processor.apply_sft_template_for_multi_turn_prompts( conversations=conversation, sft_format=vl_chat_processor.sft_format, system_prompt="", ) prompt = sft_format + vl_chat_processor.image_start_tag @torch.inference_mode() def generate( mmgpt: MultiModalityCausalLM, vl_chat_processor: VLChatProcessor, prompt: str, temperature: float = 1, parallel_size: int = 16, cfg_weight: float = 5, image_token_num_per_image: int = 576, img_size: int = 384, patch_size: int = 16, ): input_ids = vl_chat_processor.tokenizer.encode(prompt) input_ids = torch.LongTensor(input_ids) tokens = torch.zeros((parallel_size*2, len(input_ids)), dtype=torch.int).cuda() for i in range(parallel_size*2): tokens[i, :] = input_ids if i % 2 != 0: tokens[i, 1:-1] = vl_chat_processor.pad_id inputs_embeds = mmgpt.language_model.get_input_embeddings()(tokens) generated_tokens = torch.zeros((parallel_size, image_token_num_per_image), dtype=torch.int).cuda() for i in range(image_token_num_per_image): outputs = mmgpt.language_model.model(inputs_embeds=inputs_embeds, use_cache=True, past_key_values=outputs.past_key_values if i != 0 else None) hidden_states = outputs.last_hidden_state logits = mmgpt.gen_head(hidden_states[:, -1, :]) logit_cond = logits[0::2, :] logit_uncond = logits[1::2, :] logits = logit_uncond + cfg_weight * (logit_cond-logit_uncond) probs = torch.softmax(logits / temperature, dim=-1) next_token = torch.multinomial(probs, num_samples=1) generated_tokens[:, i] = next_token.squeeze(dim=-1) next_token = torch.cat([next_token.unsqueeze(dim=1), next_token.unsqueeze(dim=1)], dim=1).view(-1) img_embeds = mmgpt.prepare_gen_img_embeds(next_token) inputs_embeds = img_embeds.unsqueeze(dim=1) dec = mmgpt.gen_vision_model.decode_code(generated_tokens.to(dtype=torch.int), shape=[parallel_size, 8, img_size//patch_size, img_size//patch_size]) dec = dec.to(torch.float32).cpu().numpy().transpose(0, 2, 3, 1) dec = np.clip((dec + 1) / 2 * 255, 0, 255) visual_img = np.zeros((parallel_size, img_size, img_size, 3), dtype=np.uint8) visual_img[:, :, :] = dec os.makedirs('generated_samples', exist_ok=True) for i in range(parallel_size): save_path = os.path.join('generated_samples', "img_{}.jpg".format(i)) PIL.Image.fromarray(visual_img[i]).save(save_path) generate( vl_gpt, vl_chat_processor, prompt, )
04模型微调
我们介绍使用ms-swift对deepseek-ai/Janus-Pro-7B进行微调（注意：目前只支持图像理解的训练而不支持图像生成）。这里，我们将展示可运行的微调demo，并给出自定义数据集的格式。
在开始微调之前，请确保您的环境已准备妥当。
# pip install git+https://github.com/modelscope/ms-swift.git git clone https://github.com/modelscope/ms-swift.git cd ms-swift pip install -e .
微调脚本如下：
CUDA_VISIBLE_DEVICES=0 \ swift sft \ --model deepseek-ai/Janus-Pro-7B \ --dataset AI-ModelScope/LaTeX_OCR:human_handwrite#20000 \ --train_type lora \ --torch_dtype bfloat16 \ --num_train_epochs 1 \ --per_device_train_batch_size 1 \ --per_device_eval_batch_size 1 \ --learning_rate 1e-4 \ --lora_rank 8 \ --lora_alpha 32 \ --target_modules all-linear \ --freeze_vit true \ --gradient_accumulation_steps 16 \ --eval_steps 100 \ --save_steps 100 \ --save_total_limit 2 \ --logging_steps 5 \ --max_length 2048 \ --output_dir output \ --warmup_ratio 0.05 \ --dataloader_num_workers 4 \ --dataset_num_proc 4
训练显存占用：
如果要使用自定义数据集进行训练，你可以参考以下格式，并指定`--dataset <dataset_path>`。
{"messages": [{"role": "user", "content": "浙江的省会在哪？"}, {"role": "assistant", "content": "浙江的省会在杭州。"}]} {"messages": [{"role": "user", "content": "<image><image>两张图片有什么区别"}, {"role": "assistant", "content": "前一张是小猫，后一张是小狗"}], "images": ["/xxx/x.jpg", "/xxx/x.png"]}
训练完成后，使用以下命令对训练后的权重进行推理：
提示：这里的`--adapters`需要替换成训练生成的last checkpoint文件夹。由于adapters文件夹中包含了训练的参数文件`args.json`，因此不需要额外指定`--model`，swift会自动读取这些参数。如果要关闭此行为，可以设置`--load_args false`。
CUDA_VISIBLE_DEVICES=0 \ swift infer \ --adapters output/vx-xxx/checkpoint-xxx \ --stream false \ --max_batch_size 1 \ --load_data_args true \ --max_new_tokens 2048
推送模型到ModelScope：
CUDA_VISIBLE_DEVICES=0 \ swift export \ --adapters output/vx-xxx/checkpoint-xxx \ --push_to_hub true \ --hub_model_id '<your-model-id>' \ --hub_token '<your-sdk-token>'
