Data generation code for DLN Demo

microsoft / deep-language-networks

We view Large Language Models as stochastic language layers in a network, where the learnable parameters are the natural language prompts at each layer. We stack two such layers, feeding the output of one layer to the next. We call the stacked architecture a Deep Language Network - DLN

MIT License

87 stars 12 forks source link

Data generation code for DLN Demo #17

Closed ayushj240 closed 10 months ago

ayushj240 commented 11 months ago

Save DLN result logs to ResultLogWriter object during training of DLN.
projects/demo/result_data.json is augmented with the results from training (result_data.json supports multiple experiments)
projects/demo/demo.py can load the generated result_data.json file.
Detailed README for demo.

ayushj240 commented 11 months ago

@microsoft-github-policy-service agree