ctrl

ctrl copied to clipboard

How to finetune on TPU v3-8 nodes? It runs without error but does not seem to progress.

5

Hi! thanks for the great paper and for providing code and model. I am trying to finetune the model on a TPU v3-8 node in the Google cloud. I made...

eurka

I am following the README on how to adapt fine tuning to TPU. I can't find any configuration that works (got error with v2-8, v2-32 and v3-8, with n1-standard-64). Did...

pgrandinetti

referring to: https://www.paperspace.com/console/notebooks/create Configuration looks like: > 8 CPU Quadro M4000 30 GB RAM 250 GB SSD There's also: > 8 CPU Quadro P5000 30 GB RAM 250 GB SSD...

ckoshka

Control code is **needed** for generation. Then how does it work at testing time ? Do you give the model the _gold_ control code, and see if the generated text...

astariul

Running the model on TPUs?

2

Hi, I have the 256 and 512 models working on GCP with a Tesla V100. Text generates, but slowly, and I'm wanting to get faster generation out of the system....

vessenes

benchmarking with GPT-2

2

Any suggestion for benchmarking CTRL with GPT-2? Say, loss value, PPL, or any metric to measure text generation quality?

leejason