Vary-toy

pip install tiktoken

model:
  ...
  seq_length: 2048  # 句子长度
  checkpoint_name_or_path: "/path/to/vary_toy.ckpt"  # 权重路径
  repetition_penalty: 1.5  # 生成重复值的惩罚项
  max_decode_length: 2048  # 最大生成的句子长度
  max_new_tokens: 1024  # 生成的新token的个数
  ...
...
processor:
  ...
  tokenizer:
    vocab_file: "/path/to/qwen.tiktoken"  # 分词器路径
  ...
...

pip install torch

python mindocr/models/llm/convert_weight.py \
    --torch_ckpt_path="/path/to/pytorch_model.bin" \
    --mindspore_ckpt_path="/path/to/vary_toy.ckpt"

# 参数说明：
# torch_ckpt_path：huggingface下载的权重路径
# mindspore_ckpt_path：导出的MindSpore权重路径

python ./tools/infer/text/predict_llm.py \
    --image_dir=/path/to/image.jpg \
    --query="Provide the ocr results of this image." \
    --config_path="/path/to/vary_toy.yaml" \
    --chat_mode=False

# 参数说明：
# image_dir：图片路径
# query：查询语句
# config_path：配置文件路径
# chat_mode：是否使用对话模式

The article discusses the analysis of traffic signals using deep learning models, specifically focusing on pedestrian crossing data. The authors propose a method to extract features from videos captured by cameras and use them to train a model for predicting pedestrian behavior. They compare their approach with other methods and show that their model outperforms others in terms of accuracy and robustness. The study also highlights the limitations of their approach, such as the need for accurate hand-crafted features and the lack of consideration for different types of vehicles. Overall, the findings suggest the potential of using machine learning models to improve traffic signal analysis and enhance safety.This article is about the use of deep learning models for predicting pedestrian behavior in traffic signals. It compares the performance of different models and highlights the limitations of these approaches.