Torch ExpertMesh Doc

ExpertMesh Integration for deepseek torch version.

Installation

#TODO

Usage

For Demo Testing of mini-deekseek-v3

First, you should generate a mini deepseekv3 model. Use the script save_mini_model.sh in examples, where the config.json file controls the model size.

./examples/save_mini_model.sh

Second, use the script run_local.sh in directory examples to test the saved model. The input text can be writen in file local_test.txt.

Requirements

pip install -i requirements.txt

Deployment Example

To test the Demo, deploy mock server first:

./mock_server --expert-dim [dim]

Then, run the bash script:

./examples/run_local.sh

References

• DeepSeek-V3 Documentation
• gRPC Documentation
• DeepSeek-R1 Documentation