def main():
random.seed(42)
# Get command line arguments
parser = ParlaiParser()
parser.add_argument('-n', '--num-examples', default=10)
parser.set_defaults(datatype='train:ordered')
ImageLoader.add_cmdline_args(parser)
opt = parser.parse_args()
opt['no_cuda'] = False
opt['gpu'] = 0
# create repeat label agent and assign it to the specified task
agent = RepeatLabelAgent(opt)
world = create_task(opt, agent)
# Show some example dialogs.
with world:
for k in range(int(opt['num_examples'])):
world.parley()
print(world.display() + '\n~~')
if world.epoch_done():
print('EPOCH DONE')
break
def main():
random.seed(42)
# Get command line arguments
parser = ParlaiParser(True, True)
parser.add_argument('-n', '--num-examples', default=100000000)
parser.add_argument('-d', '--display-examples', type='bool', default=False)
parser.add_argument('-ltim', '--log-every-n-secs', type=float, default=2)
parser.set_defaults(datatype='valid')
opt = parser.parse_args(print_args=False)
eval_model(opt, parser)
def test_custom_special_tokens(self):
from parlai.agents.hugging_face.dict import Gpt2DictionaryAgent
from parlai.core.params import ParlaiParser
parser = ParlaiParser(False, False)
parser.set_defaults(gpt2_size="small",
add_special_tokens=True,
fp16=True)
Gpt2DictionaryAgent.add_cmdline_args(parser, partial_opt=None)
with testing_utils.tempdir() as tmpdir:
opt = parser.parse_kwargs(dict_file=os.path.join(tmpdir, 'dict'))
dict_agent = Gpt2DictionaryAgent(opt)
oldtokens = dict_agent.txt2vec("Hi VOLDEMORT")
prevlen = len(dict_agent)
dict_agent.add_additional_special_tokens(["VOLDEMORT"])
newlen = len(dict_agent)
assert newlen == prevlen + 1
tokens = dict_agent.txt2vec("Hi VOLDEMORT")
assert tokens != oldtokens
assert len(tokens) < len(oldtokens)
def setup_args(parser=None):
if parser is None:
parser = ParlaiParser(True, True, 'Evaluate a model')
parser.add_pytorch_datateacher_args()
# Get command line arguments
parser.add_argument('-ne', '--num-examples', type=int, default=-1)
parser.add_argument('-d', '--display-examples', type='bool', default=False)
parser.add_argument('-ltim', '--log-every-n-secs', type=float, default=2)
parser.add_argument('-micro', '--aggregate-micro', type='bool',
default=True,
help='If multitasking, average metrics over the '
'number of examples. If false, averages over the '
'number of tasks.')
parser.add_argument('--metrics', type=str, default='all',
help='list of metrics to show/compute, e.g. '
'ppl, f1, accuracy, hits@1.'
'If `all` is specified [default] all are shown.')
TensorboardLogger.add_cmdline_args(parser)
parser.set_defaults(datatype='valid')
return parser
def setup_args(parser=None):
if parser is None:
parser = ParlaiParser(True, True, 'Display data from a task')
#parser.add_pytorch_datateacher_args()
# Get command line arguments
parser.add_argument('--seed', type=int, default=42)
parser.add_argument('-n', '-ne', '--num-examples', type=int, default=10)
parser.add_argument('-ns', '--num-stored', type=int, default=10)
parser.add_argument('-mdl', '--max-display-len', type=int, default=1000)
parser.add_argument('--display-ignore-fields',
type=str,
default='agent_reply')
parser.set_defaults(datatype='train:stream')
parser.add_argument('-host', '--mongo-host', type=str)
parser.add_argument('-port', '--mongo-port', type=int)
parser.add_argument('-user', '--user-name', type=str)
parser.add_argument('-pw', '--password', type=str)
parser.add_argument('-col', '--collection-name', type=str)
WorldLogger.add_cmdline_args(parser)
return parser
def setup_args(parser=None):
if parser is None:
parser = ParlaiParser(True, True, 'Interactive chat with a model')
parser.add_argument('-d', '--display-examples', type='bool', default=False)
parser.add_argument(
'--display-prettify',
type='bool',
default=False,
help='Set to use a prettytable when displaying '
'examples with text candidates',
)
parser.add_argument(
'--display-ignore-fields',
type=str,
default='label_candidates,text_candidates',
help='Do not display these fields',
)
parser.set_defaults(model_file='models:convai2/kvmemnn/model')
LocalHumanAgent.add_cmdline_args(parser)
return parser
def setup_args(parser=None):
if parser is None:
parser = ParlaiParser(True, True, 'Display data from a task')
# Get command line arguments
parser.add_argument('-n', '-ne', '--num-examples', type=int, default=10)
parser.add_argument('-mdl', '--max-display-len', type=int, default=1000)
parser.add_argument('--display-ignore-fields',
type=str,
default='agent_reply')
parser.add_argument(
'-v',
'--display-verbose',
default=False,
action='store_true',
help=
'If false, simple converational view, does not show other message fields.',
)
parser.set_defaults(datatype='train:ordered')
return parser
def add_cmdline_args(argparser: ParlaiParser):
"""
Override to add init-fairseq-model arg.
"""
TransformerGeneratorAgent.add_cmdline_args(argparser)
group = argparser.add_argument_group('Bart Args')
group.add_argument(
'--init-fairseq-model',
type=str,
default=None,
help='fairseq checkpoint for bart',
)
group.add_argument(
'--output-conversion-path',
type=str,
default=None,
help='where to save fairseq conversion',
)
argparser.set_defaults(dict_tokenizer='gpt2')
argparser.set_defaults(**BART_ARGS)
def setup_args():
"""
Set up conversion args.
"""
parser = ParlaiParser()
parser.add_argument(
'-n',
'--num-episodes',
default=-1,
type=int,
help='Total number of episodes to convert, -1 to convert \
all examples',
)
parser.add_argument(
'-of',
'--outfile',
default=None,
type=str,
help='Output file where to save, by default will be \
created in /tmp',
)
parser.add_argument('-s1id',
'--speaker-0-id',
type=str,
help='Speaker id of agent who speaks first')
parser.add_argument(
'-s1id',
'--speaker-1-id',
type=str,
help='Speaker id of agent who speaks second',
)
parser.add_argument(
'--prepended-context',
type='bool',
default=False,
help='specify if the context is prepended to the first act',
)
parser.add_argument('-ltim', '--log-every-n-secs', type=float, default=10)
parser.set_defaults(datatype='train:ordered')
return parser
def setup_args(parser=None):
if parser is None:
parser = ParlaiParser(True, True, 'Evaluate a model')
# Get command line arguments
parser.add_argument('-ne', '--num-examples', type=int, default=-1)
parser.add_argument('-d', '--display-examples', type='bool', default=False)
parser.add_argument('-ltim', '--log-every-n-secs', type=float, default=2)
parser.add_argument('--metrics',
type=str,
default="all",
help="list of metrics to show/compute, e.g. "
"ppl,f1,accuracy,hits@1."
"If 'all' is specified [default] all are shown.")
parser.add_argument('-pb', '--perturb', type=str, default="None")
parser.add_argument('-sft',
'--skip_first_turn',
type='bool',
default=False)
TensorboardLogger.add_cmdline_args(parser)
parser.set_defaults(datatype='valid')
return parser
def setup_args(parser=None) -> ParlaiParser:
# Get command line arguments
if not parser:
parser = ParlaiParser()
parser.add_argument(
'-n',
'--num-examples',
default=-1,
type=int,
help='Total number of exs to convert, -1 to convert all examples',
)
parser.add_argument(
'-of',
'--outfile',
default=None,
type=str,
help='Output file where to save, by default will be created in /tmp',
)
parser.add_argument('-ltim', '--log-every-n-secs', type=float, default=2)
parser.set_defaults(datatype='train:evalmode')
return parser
def setup_args(parser=None):
if parser is None:
parser = ParlaiParser(True, True,
'compute statistics from model predictions')
parser.add_pytorch_datateacher_args()
DictionaryAgent.add_cmdline_args(parser)
# Get command line arguments
parser.add_argument('-ne', '--num-examples', type=int, default=-1)
parser.add_argument('-ltim', '--log-every-n-secs', type=float, default=2)
parser.add_argument(
'-ed',
'--external-dict',
type=str,
default=None,
help='External dictionary for stat computation',
)
parser.add_argument(
'-fb',
'--freq-bins',
type=str,
default='0,100,1000,10000',
help='Bins boundaries for rare words stat',
)
parser.add_argument(
'-dup',
'--dump-predictions-path',
type=str,
default=None,
help='Dump predictions into file',
)
parser.add_argument(
'-cun',
'--compute-unique',
type=bool,
default=True,
help='Compute %% of unique responses from the model',
)
parser.set_defaults(datatype='valid', model='repeat_label')
TensorboardLogger.add_cmdline_args(parser)
return parser
def setup_args(parser=None):
# Get command line arguments
if parser is None:
parser = ParlaiParser(True, True, 'Display data from a task')
parser.add_argument(
'-mf',
'--model-file',
type=str,
default='/private/home/bhancock/metadialog/models/unknown.mdl',
)
parser.add_argument('-dfile', '--deploy-file', type=str, default='train_c')
parser.add_argument('-sfile', '--supp-file', type=str, default='supp_c')
parser.add_argument(
'-cr',
'--conversion-rate',
type=float,
default=1.0,
help="The fraction of misses converted into new training data",
)
parser.add_argument(
'-ca',
'--conversion-acc',
type=float,
default=1.0,
help="The fraction of converted data that have a correct label",
)
parser.set_defaults(bs=1)
parser.set_defaults(ecands='inline')
parser.set_defaults(datatype='valid')
return parser
def main():
parser = ParlaiParser(True, True)
parser.set_defaults(batchsize=10,
sample=True,
max_seq_len=256,
beam_size=3,
annealing_topk=None,
annealing=0.6,
length_penalty=0.7)
ConvAIWorld.add_cmdline_args(parser)
TransformerAgent.add_cmdline_args(parser)
opt = parser.parse_args()
agent = TransformerAgent(opt)
world = ConvAIWorld(opt, [agent])
while True:
try:
world.parley()
except Exception as e:
print('Exception: {}'.format(e))
def setup_args(parser=None):
if parser is None:
parser = ParlaiParser(True, True, 'Evaluate a model')
parser.add_pytorch_datateacher_args()
# Get command line arguments
parser.add_argument(
'-rf',
'--report-filename',
type=str,
default='',
help='Saves a json file of the evaluation report either as an '
'extension to the model-file (if begins with a ".") or a whole '
'file path. Set to the empty string to not save at all.',
)
parser.add_argument('-ne', '--num-examples', type=int, default=-1)
parser.add_argument('-d', '--display-examples', type='bool', default=False)
parser.add_argument('-ltim', '--log-every-n-secs', type=float, default=2)
parser.add_argument(
'-micro',
'--aggregate-micro',
type='bool',
default=False,
help='If multitasking, average metrics over the '
'number of examples. If false, averages over the '
'number of tasks.',
)
parser.add_argument(
'-mcs',
'--metrics',
type=str,
default='default',
help='list of metrics to show/compute, e.g. all, default,'
'or give a list split by , like '
'ppl,f1,accuracy,hits@1,rouge,bleu'
'the rouge metrics will be computed as rouge-1, rouge-2 and rouge-l',
)
TensorboardLogger.add_cmdline_args(parser)
parser.set_defaults(datatype='valid')
return parser
def main():
random.seed(42)
# Get command line arguments
parser = ParlaiParser()
parser.add_argument(
'-n',
'--num-examples',
default=-1,
type=int,
help='Total number of exs to convert, -1 to convert all examples',
)
parser.add_argument(
'-of',
'--outfile',
default=None,
type=str,
help='Output file where to save, by default will be created in /tmp',
)
parser.add_argument('-ltim', '--log-every-n-secs', type=float, default=2)
parser.set_defaults(datatype='train:evalmode')
opt = parser.parse_args()
build_cands(opt)
def main():
random.seed(42)
# Get command line arguments
parser = ParlaiParser(True, True)
parser.add_argument('-n', '--num-examples', default=10)
# by default we want to display info about the validation set
parser.set_defaults(datatype='valid')
opt = parser.parse_args()
# Create model and assign it to the specified task
agent = create_agent(opt)
world = create_task(opt, agent)
# Show some example dialogs.
with world:
for k in range(int(opt['num_examples'])):
world.parley()
print(world.display() + "\n~~")
if world.epoch_done():
print("EPOCH DONE")
break
def run_actor(self):
MODEL_FILE = '/Volumes/Data/ParlAI/from_pretrained_wiki/wikiqa_tdifd'
DB_PATH = '/Volumes/Data/ParlAI/from_pretrained_wiki/wikiqa_tdifd.db'
TFIDF_PATH = '/Volumes/Data/ParlAI/from_pretrained_wiki/wikiqa_tdifd.tfidf'
# keep things quiet
logger.setLevel(ERROR)
parser = ParlaiParser(True, True)
parser.set_defaults(
model_file=MODEL_FILE,
interactive_mode=True,
)
opt = parser.parse_args([], print_args=False)
opt['interactive_mode'] = True
agent = create_agent(opt)
train_world = create_task(opt, agent)
# pass examples to dictionary
while not train_world.epoch_done():
train_world.parley()
agent.observe(obs)
reply = 'Hello'
return reply
def add_cmdline_args(cls,
parser: ParlaiParser,
partial_opt: Optional[Opt] = None) -> ParlaiParser:
agent = parser.add_argument_group('DialoGPT Args')
agent.add_argument(
'--gpt2-size',
type=str,
default='medium',
choices=['small', 'medium', 'large'],
help='Which size model to initialize.',
)
parser.set_defaults(
delimiter='<|endoftext|>',
history_add_global_end_token='<|endoftext|>',
text_truncate=768,
label_truncate=256,
dict_maxexs=0, # skip building dictionary
)
super().add_cmdline_args(parser, partial_opt=partial_opt)
warn_once(
'WARNING: this model is in beta and the API is subject to change.')
return agent
def setup_args(parser=None):
if parser is None:
parser = ParlaiParser(True, True, 'Self chat with a model')
parser.add_argument('--seed', type=int, default=42)
parser.add_argument('-d', '--display-examples', type='bool', default=True)
parser.add_argument('-n', '-ne', '--num-examples', type=int, default=10)
parser.add_argument('-ltim', '--log-every-n-secs', type=float, default=2)
parser.add_argument(
'--display-ignore-fields',
type=str,
default='label_candidates,text_candidates',
help='Do not display these fields',
)
parser.add_argument(
'-it',
'--interactive-task',
type='bool',
default=True,
help='Create interactive version of task',
)
parser.add_argument(
'--selfchat-max-turns',
type=int,
default=10,
help="The number of dialogue turns before self chat ends.",
)
parser.add_argument(
'--seed-messages-from-task',
action='store_true',
help="Automatically seed conversation with messages from task dataset.",
)
parser.add_argument('--outfile', type=str, default='/tmp/selfchat.json')
parser.add_argument('--format',
type=str,
default='json',
choices={'parlai', 'json'})
parser.set_defaults(interactive_mode=True, task='self_chat')
WorldLogger.add_cmdline_args(parser)
return parser
def main():
# Get command line arguments
parser = ParlaiParser(True, False)
parser.set_defaults(datatype='train:ordered')
opt = parser.parse_args()
bsz = opt.get('batchsize', 1)
opt['no_cuda'] = False
opt['gpu'] = 0
opt['num_epochs'] = 1
# create repeat label agent and assign it to the specified task
agent = RepeatLabelAgent(opt)
world = create_task(opt, agent)
logger = ProgressLogger(should_humanize=False)
print("Beginning image extraction...")
exs_seen = 0
total_exs = world.num_examples()
while not world.epoch_done():
world.parley()
exs_seen += bsz
logger.log(exs_seen, total_exs)
print("Finished extracting images")
def add_cmdline_args(cls,
parser: ParlaiParser,
partial_opt: Optional[Opt] = None) -> ParlaiParser:
"""
Override to add init-fairseq-model arg.
"""
super().add_cmdline_args(parser, partial_opt=partial_opt)
group = parser.add_argument_group('Bart Args')
group.add_argument(
'--init-fairseq-model',
type=str,
default=None,
help='fairseq checkpoint for bart',
)
group.add_argument(
'--output-conversion-path',
type=str,
default=None,
help='where to save fairseq conversion',
)
parser.set_defaults(dict_tokenizer='gpt2')
parser.set_defaults(**BART_ARGS)
return parser
def add_cmdline_args(cls,
parser: ParlaiParser,
partial_opt: Optional[Opt] = None) -> ParlaiParser:
agent = parser.add_argument_group("Gpt2 Args")
agent.add_argument("--model-name",
type=str,
default=None,
help="Any GPT-2 model names.")
agent.add_argument(
"--gpt2-size",
type=str,
default="small",
choices=["small", "medium", "large", "xl", "distilgpt2"],
help="Which size model to initialize.",
)
agent.add_argument(
"--add-special-tokens",
type="bool",
default=True,
help="Add special tokens (like PAD, etc.). If False, "
"Can only use with batch size 1.",
)
agent.add_argument(
"--add-start-token",
type="bool",
default=False,
help="Add start tokens when finetuning.",
)
parser.set_defaults(
text_truncate=768,
label_truncate=256,
dict_maxexs=0, # skip building dictionary
)
super().add_cmdline_args(parser, partial_opt=partial_opt)
warn_once(
"WARNING: this model is in beta and the API is subject to change.")
return agent
def add_cmdline_args(
cls, parser: ParlaiParser, partial_opt: Optional[Opt] = None
) -> ParlaiParser:
"""
Add CLI args.
"""
super().add_cmdline_args(parser, partial_opt=partial_opt)
parser = parser.add_argument_group('BERT Classifier Arguments')
parser.add_argument(
'--type-optimization',
type=str,
default='all_encoder_layers',
choices=[
'additional_layers',
'top_layer',
'top4_layers',
'all_encoder_layers',
'all',
],
help='which part of the encoders do we optimize '
'(defaults to all layers)',
)
parser.add_argument(
'--add-cls-token',
type='bool',
default=True,
help='add [CLS] token to text vec',
)
parser.add_argument(
'--sep-last-utt',
type='bool',
default=False,
help='separate the last utterance into a different'
'segment with [SEP] token in between',
)
parser.set_defaults(dict_maxexs=0) # skip building dictionary
return parser
def add_cmdline_args(cls,
parser: ParlaiParser,
partial_opt: Optional[Opt] = None) -> ParlaiParser:
"""
Add CLI args.
"""
super().add_cmdline_args(parser, partial_opt=partial_opt)
parser = parser.add_argument_group("BERT Classifier Arguments")
parser.add_argument(
"--type-optimization",
type=str,
default="all_encoder_layers",
choices=[
"additional_layers",
"top_layer",
"top4_layers",
"all_encoder_layers",
"all",
],
help="which part of the encoders do we optimize "
"(defaults to all layers)",
)
parser.add_argument(
"--add-cls-token",
type="bool",
default=True,
help="add [CLS] token to text vec",
)
parser.add_argument(
"--sep-last-utt",
type="bool",
default=False,
help="separate the last utterance into a different"
"segment with [SEP] token in between",
)
parser.set_defaults(dict_maxexs=0) # skip building dictionary
return parser
def setup_args(parser=None):
if parser is None:
parser = ParlaiParser(True, True, 'Evaluate a model')
# Get command line arguments
parser.add_argument(
'-rf',
'--report-filename',
type=str,
default='',
help='Saves a json file of the evaluation report either as an '
'extension to the model-file (if begins with a ".") or a whole '
'file path. Set to the empty string to not save at all.',
)
parser.add_argument(
'--save-world-logs',
type='bool',
default=False,
help='Saves a jsonl file containing all of the task examples and '
'model replies. Must also specify --report-filename.',
)
parser.add_argument('-ne', '--num-examples', type=int, default=-1)
parser.add_argument('-d', '--display-examples', type='bool', default=False)
parser.add_argument('-ltim', '--log-every-n-secs', type=float, default=2)
parser.add_argument(
'-mcs',
'--metrics',
type=str,
default='default',
help='list of metrics to show/compute, e.g. all, default,'
'or give a list split by , like '
'ppl,f1,accuracy,hits@1,rouge,bleu'
'the rouge metrics will be computed as rouge-1, rouge-2 and rouge-l',
)
WorldLogger.add_cmdline_args(parser)
TensorboardLogger.add_cmdline_args(parser)
parser.set_defaults(datatype='valid')
return parser
def setup_args(cls):
"""
Setup args.
"""
parser = ParlaiParser(True, True, 'Index Dense Embs')
parser.add_argument(
'--embeddings-dir', type=str, help='directory of embeddings'
)
parser.add_argument(
'--embeddings-name', type=str, default='', help='name of emb part'
)
parser.add_argument(
'--partition-index',
type='bool',
default=False,
help='specify True to partition indexing per file (useful when all files do not fit into memory)',
)
parser.add_argument(
'--save-index-dir',
type=str,
help='directory in which to save index',
default=None,
)
parser.add_argument(
'--num-shards',
type=int,
default=1,
help='how many workers to use to split up the work',
)
parser.add_argument(
'--shard-id',
type=int,
help='shard id for this worker. should be between 0 and num_shards',
)
parser = RagAgent.add_cmdline_args(parser)
parser.set_defaults(compressed_indexer_gpu_train=True)
return parser
def add_cmdline_args(cls,
parser: ParlaiParser,
partial_opt: Optional[Opt] = None) -> ParlaiParser:
add_common_args(parser)
parser = parser.add_argument_group('Bert Ranker Arguments')
parser.add_argument(
'--biencoder-model-file',
type=str,
default=None,
help='path to biencoder model. Default to model-file_bi',
)
parser.add_argument(
'--biencoder-top-n',
type=int,
default=10,
help=
'default number of elements to keep from the biencoder response',
)
parser.add_argument(
'--crossencoder-model-file',
type=str,
default=None,
help='path to crossencoder model. Default to model-file_cross',
)
parser.add_argument(
'--crossencoder-batchsize',
type=int,
default=-1,
help=
'crossencoder will be fed those many elements at train or eval time.',
)
parser.set_defaults(
encode_candidate_vecs=True,
dict_maxexs=0 # skip building dictionary
)
return parser
def main():
random.seed(42)
# Get command line arguments
parser = ParlaiParser(True, True)
parser.add_argument('-n', '--num-examples', default=100000000)
parser.add_argument('-d', '--display-examples', type='bool', default=False)
parser.set_defaults(datatype='valid')
opt = parser.parse_args()
# Create model and assign it to the specified task
agent = create_agent(opt)
world = create_task(opt, agent)
# Show some example dialogs:
for k in range(int(opt['num_examples'])):
world.parley()
print("---")
if opt['display_examples']:
print(world.display() + "\n~~")
print(world.report())
if world.epoch_done():
print("EPOCH DONE")
break
world.shutdown()
def setup_args(parser=None):
if parser is None:
parser = ParlaiParser(
True, True, 'Interactive chat with a model on the command line'
)
parser.add_argument('-d', '--display-examples', type='bool', default=False)
parser.add_argument(
'--display-prettify',
type='bool',
default=False,
help='Set to use a prettytable when displaying '
'examples with text candidates',
)
parser.add_argument(
'--display-ignore-fields',
type=str,
default='label_candidates,text_candidates',
help='Do not display these fields',
)
parser.add_argument(
'-it',
'--interactive-task',
type='bool',
default=True,
help='Create interactive version of task',
)
parser.add_argument(
'--save-world-logs',
type='bool',
default=False,
help='Saves a jsonl file containing all of the task examples and '
'model replies. Must also specify --report-filename.',
)
parser.set_defaults(interactive_mode=True, task='interactive')
LocalHumanAgent.add_cmdline_args(parser)
return parser
def add_cmdline_args(cls,
parser: ParlaiParser,
partial_opt: Optional[Opt] = None) -> ParlaiParser:
"""
Add command-line arguments specifically for this agent.
"""
AbstractReranker.add_cmdline_args(parser)
parser.set_defaults(skip_generation=False)
gen_agent = parser.add_argument_group("Generator Rerank Agent")
gen_agent.add_argument(
'--inference-strategies',
type=str,
default=None,
help='comma-separated list of inference strategies. '
'if specified, re-rank over several inference strategies',
)
gen_agent.add_argument(
'--debug-mode',
type='bool',
default=False,
help='specify to enable certain debugging procedures.',
)
return parser
def main():
random.seed(42)
# Get command line arguments
parser = ParlaiParser(True, True)
parser.add_argument('-n', '--num-examples', default=100000000)
parser.add_argument('-d', '--display-examples', type='bool', default=False)
parser.set_defaults(datatype='valid')
opt = parser.parse_args()
# Create model and assign it to the specified task
agent = create_agent(opt)
world = create_task(opt, agent)
# Show some example dialogs:
for k in range(int(opt['num_examples'])):
world.parley()
print("---")
if opt['display_examples']:
print(world.display() + "\n~~")
print(world.report())
if world.epoch_done():
print("EPOCH DONE")
break
world.shutdown()
def setup_args(parser=None):
if parser is None:
parser = ParlaiParser(True, True, 'Generate self-chats of a model')
parser.add_argument('--seed', type=int, default=42)
parser.add_argument('-d', '--display-examples', type='bool', default=True)
parser.add_argument(
'--display-add-fields',
type=str,
default='',
help='Display these fields when verbose is off (e.g., "--display-add-fields label_candidates,beam_texts")',
)
parser.add_argument(
'-st',
'--selfchat-task',
type='bool',
default=True,
help='Create a self chat version of the task',
)
parser.add_argument(
'--num-self-chats', type=int, default=1, help='Number of self chats to run'
)
parser.add_argument(
'--selfchat-max-turns',
type=int,
default=6,
help='The number of dialogue turns before self chat ends',
)
parser.add_argument(
'--seed-messages-from-task',
type='bool',
default=False,
help='Automatically seed conversation with messages from task dataset.',
)
parser.add_argument(
'--seed-messages-from-file',
default=None,
help='If specified, loads newline-separated strings from the file as conversation starters.',
)
parser.add_argument(
'--outfile', type=str, default=None, help='File to save self chat logs'
)
parser.add_argument(
'--save-format',
type=str,
default='conversations',
choices=['conversations', 'parlai'],
help='Format to save logs in. conversations is a jsonl format, parlai is a text format.',
)
parser.add_argument(
'-pmf',
'--partner-model-file',
default=None,
help='Define a different partner for self chat',
)
parser.add_argument(
'--partner-opt-file',
default=None,
help='Path to file containing opts to override for partner',
)
parser.set_defaults(interactive_mode=True, task='self_chat')
WorldLogger.add_cmdline_args(parser, partial_opt=None)
return parser
