Repetitive generation for simple prompt about ctrl

[kostyan0005]

I have the same the problem.

Additional info:

• I converted the code to python3 as in #6.
• The model runs on CPUs because of its size.

[keskarnitish]

Thank you for the response. My primary concern is not about the warnings but is the out put in the expected format? I thought it would generate a full sentence or a paragraph of a certain length as in the examples instead of displaying every line with a new word prediction.

Aah. I see.

You can generate just once by indenting the print statements on https://github.com/salesforce/ctrl/blob/master/generation.py#L272 and https://github.com/salesforce/ctrl/blob/master/generation.py#L273 to be outside the generation for loop

[strin]

Similar error for the unicorn example:

[keskarnitish]

This is usually symptomatic of not loading a model.
Are you sure that --model_dir points to the right location and that there entire checkpoint is available there?
Also, which model version are you using?

[pradeepthiyyagura]

I have the same problem and I am using

python 2.7.16
tensorflow 1.14.0 gpu_py27h39f1c70_0
model seqlen256_v1.ckpt

Command $python generation.py --model_dir /home/tpradeep/ctrl/seqlen256_v1.ckpt

Quite a few warning messages and then the following

2019-09-13 08:03:27.284843: W tensorflow/compiler/jit/mark_for_compilation_pass.cc:1412] (One-time warning): Not using XLA:CPU for cluster because envvar TF_XLA_FLAGS=--tf_xla_cpu_global_jit was not set. If you want XLA:CPU, either set that envvar, or use experimental_jit_scope to enable XLA:CPU. To confirm that XLA is active, pass --vmodule=xla_compilation_cache=1 (as a proper command-line flag, not via TF_XLA_FLAGS) or set the envvar XLA_FLAGS=--xla_hlo_profile.
WARNING:tensorflow:From /home/tpradeep/.local/lib/python2.7/site-packages/tensorflow/python/training/saver.py:1066: get_checkpoint_mtimes (from tensorflow.python.training.checkpoint_management) is deprecated and will be removed in a future version.
Instructions for updating:
Use standard file utilities to get mtimes.
Loading vocabulary from vocab ...
Read 6086453827 words (246531 unique) from vocabulary file.
Loading codes from codes ...
Read 200000 codes from the codes file.
ENTER PROMPT: Wikipedia Salesforce Inc. is
Wikipedia Salesforce Inc. is a

Wikipedia Salesforce Inc. is a software

Wikipedia Salesforce Inc. is a software company

Wikipedia Salesforce Inc. is a software company that

Wikipedia Salesforce Inc. is a software company that provides

Wikipedia Salesforce Inc. is a software company that provides cloud-based

^C^CContinuing
ENTER PROMPT: ^CTraceback (most recent call last):
File "generation.py", line 162, in
prompt = raw_input('ENTER PROMPT: ')
KeyboardInterrupt
^C

[keskarnitish]

@pradeepthiyyagura, your generation script seems to be fine? Is your concern about the warnings?

[pradeepthiyyagura]

Thank you for the response. My primary concern is not about the warnings but is the out put in the expected format? I thought it would generate a full sentence or a paragraph of a certain length as in the examples instead of displaying every line with a new word prediction.

[keskarnitish]

Seems like this has been fixed by specifying the right --model_dir. Closing for now, reopen as necessary.