def test_generation_from_short_input_same_as_parlai_3B(self):
FASTER_GEN_KWARGS = dict(num_beams=1,
early_stopping=True,
min_length=15,
max_length=25)
TOK_DECODE_KW = dict(skip_special_tokens=True,
clean_up_tokenization_spaces=True)
torch.cuda.empty_cache()
model = BlenderbotForConditionalGeneration.from_pretrained(
self.ckpt).half().to(torch_device)
src_text = ["Sam"]
model_inputs = self.tokenizer(src_text,
return_tensors="pt").to(torch_device)
generated_utterances = model.generate(**model_inputs,
**FASTER_GEN_KWARGS)
tgt_text = 'Sam is a great name. It means "sun" in Gaelic.'
generated_txt = self.tokenizer.batch_decode(generated_utterances,
**TOK_DECODE_KW)
assert generated_txt[0].strip() == tgt_text
src_text = "Social anxiety\nWow, I am never shy. Do you have anxiety?\nYes. I end up sweating and blushing and feel like i'm going to throw up.\nand why is that?"
model_inputs = self.tokenizer([src_text],
return_tensors="pt").to(torch_device)
generated_ids = model.generate(**model_inputs, **FASTER_GEN_KWARGS)[0]
reply = self.tokenizer.decode(generated_ids, **TOK_DECODE_KW)
assert "I think it's because we are so worried about what people think of us." == reply.strip(
)
del model
def __init__(self):
# model_name = 'facebook/blenderbot_small-90M'
model_name = 'facebook/blenderbot-400M-distill'
# model_name = 'facebook/blenderbot-1B-distill'
# model_name = 'facebook/blenderbot-3B'
print("Creating model")
self.model = BlenderbotForConditionalGeneration.from_pretrained(model_name)
print("Creating tokenizer")
self.tokenizer = BlenderbotTokenizer.from_pretrained(model_name)
self.history = []
def __init__(self, size, env, device, max_context_length):
"""
The Blender chatbot model was proposed in Recipes for building an open-domain chatbot Stephen Roller,
Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith,
Y-Lan Boureau, Jason Weston on 30 Apr 2020.
Args:
size (str): model size
env (BaseEnv): dialogue environment
device (str): device (one of ['CPU', 'CUDA', 'CUDA:N']
max_context_length (int): max history context length
(it means that length of input context tokens)
"""
assert size in ['small', 'medium', 'large', 'xlarge'], \
"model size must be one of ['small', 'medium', 'large', 'xlarge']"
if size == "small":
super().__init__("facebook/blenderbot_small-90M", env)
self.model = BlenderbotSmallForConditionalGeneration.from_pretrained(
self.name).to(device)
self.tokenizer = BlenderbotSmallTokenizer.from_pretrained(
self.name)
else:
if size == "medium":
super().__init__("facebook/blenderbot-400M-distill", env)
elif size == "large":
super().__init__("facebook/blenderbot-1B-distill", env)
elif size == "xlarge":
super().__init__("facebook/blenderbot-3B", env)
self.model = BlenderbotForConditionalGeneration.from_pretrained(
self.name).to(device)
self.tokenizer = BlenderbotTokenizer.from_pretrained(self.name)
self.size = size
self.device = device.lower()
self.max_context_length = max_context_length
self.eos = "</s> <s>"
print('Done!')
from transformers import BlenderbotTokenizer, BlenderbotForConditionalGeneration mname = 'facebook/blenderbot-400M-distill' # mname = 'facebook/blenderbot-3B' # Too big to run on laptop model = BlenderbotForConditionalGeneration.from_pretrained(mname) tokenizer = BlenderbotTokenizer.from_pretrained(mname) UTTERANCE = "My friends are cool but they eat too many carbs." inputs = tokenizer([UTTERANCE], return_tensors='pt') reply_ids = model.generate(**inputs) print(tokenizer.batch_decode(reply_ids))
tokenizer = GPT2Tokenizer.from_pretrained("microsoft/DialoGPT-small")
#------dialogpt medium------#
model = GPT2LMHeadModel.from_pretrained("microsoft/DialoGPT-medium")
tokenizer = GPT2Tokenizer.from_pretrained("microsoft/DialoGPT-medium")
print("dialogpt is done!")
elif download_type == 'gptneo':
#------gptneo small------#
model = AutoModelForCausalLM.from_pretrained("EleutherAI/gpt-neo-125M")
tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-neo-125M")
#------gptneo large------#
#model = AutoModelForCausalLM.from_pretrained("EleutherAI/gpt-neo-1.3B")
#tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-neo-1.3B")
print("gptneo is done!")
elif download_type == 'blender':
#------blender small------#
model = BlenderbotSmallForConditionalGeneration.from_pretrained(
"facebook/blenderbot_small-90M")
tokenizer = BlenderbotSmallTokenizer.from_pretrained(
"facebook/blenderbot_small-90M")
#------blender medium------#
model = BlenderbotForConditionalGeneration.from_pretrained(
"facebook/blenderbot-400M-distill")
tokenizer = BlenderbotTokenizer.from_pretrained(
"facebook/blenderbot-400M-distill")
print("blender is done!")
def load_Blenderbot(Bbot_PATH):
BbotModel = BlenderbotForConditionalGeneration.from_pretrained(Bbot_PATH)
BbotTokenizer = BlenderbotTokenizer.from_pretrained(Bbot_PATH)
return BbotModel, BbotTokenizer
def get_model():
mname = 'facebook/blenderbot-90M'
model = BlenderbotForConditionalGeneration.from_pretrained(mname)
tokenizer = BlenderbotSmallTokenizer.from_pretrained(mname)
return model, tokenizer
