model_name='355M',
steps=1000,
restore_from='fresh',
run_name='run1',
print_every=10,
sample_every=200,
save_every=500
)
"""After the model is trained, you can copy the checkpoint folder to Drive.
do everything with Drive, its messy otherwise
"""
gpt2.copy_checkpoint_to_gdrive(run_name='run1')
"""## Generate Text From The Trained Model
After you've trained the model, `generate` generates a single text from the loaded model.
"""
gpt2.generate(sess, run_name='run1')
"""
You can generate multiple texts at a time by specifing `nsamples`.
Unique to GPT-2, you can pass a `batch_size` to generate multiple samples in parallel, giving a massive speedup (in Colaboratory, set a maximum of 20 for `batch_size`).
* **`length`**: Number of tokens to generate (default 1023, the maximum)
* **`temperature`**: The higher the temperature, the crazier the text (default 0.7, recommended to keep between 0.7 and 1.0)
* **`top_k`**: Limits the generated guesses to the top *k* guesses (default 0 which disables the behavior; if the generated output is super crazy, you may want to set `top_k=40`)
!pip install gpt_2_simple -t . --no-deps --upgrade
!pip install toposort
import gpt_2_simple as gpt2
from google.colab import drive
!cp /content/drive/MyDrive/Echidna/gpt_2_simple/gpt_2.py /content/gpt_2_simple/gpt_2.py
model_name = "355M"
gpt2.download_gpt2(model_name=model_name)
gpt2.mount_gdrive()
root_dir = '/content/drive/MyDrive/Echidna'
sess = gpt2.start_tf_sess()
gpt2.finetune(sess,
root_dir + '/data/data.json',
steps=4000,
model_name=model_name,
print_every=10,
sample_every=200,
save_every=500,
run_name='echidna',
restore_from='fresh'
)
gpt2.copy_checkpoint_to_gdrive(run_name='echidna')
file_name = "EEG-concentrating.csv" # add your file here
sess = gpt2.start_tf_sess()
gpt2.finetune(
sess,
dataset=file_name,
steps=1000, #update steps as necessary
restore_from='latest', # to continue from previous
run_name='eeg-concentrating-1',
print_every=1,
sample_every=100,
save_every=500,
overwrite=True)
"""
#If using Colab:
gpt2.copy_checkpoint_to_gdrive(run_name='classical-piano-1')
"""
# Generation loop:
gpt2.copy_checkpoint_from_gdrive(run_name='eeg-concentrating-1')
sess = gpt2.start_tf_sess()
gpt2.load_gpt2(sess, run_name='eeg-concentrating-1')
#Examples of generation:
print("Generating without previous data as input")
for x in range(1, 10):
gpt2.generate(sess, run_name='eeg-concentrating-1', length=100)
sess = gpt2.start_tf_sess()
file_name = "rightcontent.txt"
gpt2.finetune(sess,
dataset=file_name,
model_name='124M',
steps=8000,
restore_from='fresh',
run_name='run_right_7000',
print_every=10,
sample_every=200,
save_every=500,
)
gpt2.copy_checkpoint_to_gdrive(run_name='run_right_7000')
"""You're done! Feel free to go to the **Generate Text From The Trained Model** section to generate text based on your retrained model.
## Load a Trained Model Checkpoint
Running the next cell will copy the `.rar` checkpoint file from your Google Drive into the Colaboratory VM.
"""
gpt2.copy_checkpoint_from_gdrive(run_name='run_right_7000')
sess = gpt2.start_tf_sess()
gpt2.load_gpt2(sess, run_name='run_right_7000')
"""## Generate Text From The Trained Model
# Or any other text file to be trained
sess = gpt2.start_tf_sess()
gpt2.finetune(sess,
dataset=fantasy,
model_name='124M',
steps=100,
restore_from='fresh',
run_name='fantasy3',
print_every=2,
sample_every=5,
save_every=10
)
#Stop to see if it's trained enough
gpt2.copy_checkpoint_to_gdrive(run_name='fantasy3')
input_text="I was riding a cycle when"
#this is where we enter the text to get suggestions
#Other hyperparameters
length=20
top_p=0.5
temperature=0.9
top_k=0.8
#To be used while running
gpt2.generate(sess,
prefix=input_text,
length=length,
temperature=temperature,
top_p=top_p,
top_k=top_k,