facebookresearch/ParlAI

Model export to ONNX

Open

#2,815 创建于 2020年7月6日

在 GitHub 查看
 (3 评论) (1 反应) (0 负责人)Python (10,371 star) (2,089 fork)batch import
Help WantedMediumnever-stale

描述

How would one export the trained dialogue models to ONNX? If you can point me to the right files where I can attempt to do something of the sort of what's mentioned in https://pytorch.org/tutorials/advanced/super_resolution_with_onnxruntime.html, it'd be really awesome! Thanks!

贡献者指南

技术栈
pythonpytorch
领域
machine learningai
议题类型
feature
难度面向新贡献者的预计实现难度，1 表示很小改动，5 表示专家级工作。
4
预计时间有经验贡献者完成调查、实现、测试并准备 pull request 的粗略时间范围。
1-2 days
活动状态议题当前的可参与程度：新鲜、活跃、陈旧、阻塞或等待维护者输入。
stale
清晰度议题是否清楚说明期望改动、验收标准和下一步。
unclear
前置要求
PyTorchONNXParlAI model architecture
新手友好度1-100 的估计分数，表示该议题对首次贡献者的友好程度。
20
研究方向
The issue requests guidance on exporting trained dialogue models to ONNX. To begin, explore the ParlAI repository to understand how models are defined and trained, particularly the TransformerGeneratorAgent or other dialogue models. Identify the model's forward method and examine if the model uses standard PyTorch modules. Then, follow the PyTorch ONNX tutorial to attempt exporting a simple model. Check existing files like parlai/core/torch generator agent.py for model definitions. Note that the issue is a question and lacks a concrete implementation plan; further clarification on which specific model to export would be needed.