Closed masterzzzen closed 4 years ago
I realized that in
model = t5.models.MtfModel(
model_dir=MODEL_DIR,
tpu=TPU_ADDRESS,
tpu_topology=TPU_TOPOLOGY,
model_parallelism=model_parallelism,
batch_size=train_batch_size,
sequence_length={"inputs": 128, "targets": 32},
learning_rate_schedule=0.003,
save_checkpoints_steps=200,
keep_checkpoint_max=keep_checkpoint_max if ON_CLOUD else None,
iterations_per_loop=100,
)
save_checkpoints_steps=200
must be smaller than FINETUNE_STEPS
And that solved the problem.
Hi, I'm fine tuning the "small" model a cloud TPU for 10 steps only. When I got to the Evaluate step, I got the following error:
I haven't changed the default Evaluate code that came with the Colab notebook
Here's the link to my notebook: https://colab.research.google.com/drive/1846Xp0UpEgdNTlmKcP0mcvOtsdeqLrxa?usp=sharing
And here's a screenshot of the objects inside my models/small bucket
Is there a problem with how few steps I've fine-tuned the model?
Thank you!