Spaces:
Sleeping
Sleeping
File size: 18,900 Bytes
a1e5063 94aba93 fa94b9c 94aba93 9c225f1 fa94b9c 94aba93 5a521e1 9c225f1 71852c2 1ac4018 aa9d7ef a1e5063 aa9d7ef 57062ef aa9d7ef 57062ef aa9d7ef 1ee4b73 aa9d7ef 448dded e690b8a 448dded aa9d7ef a3a369c aa9d7ef faa25c0 621db44 aa9d7ef 5f35567 aa9d7ef 1ac4018 57062ef 1ac4018 57062ef 57ca910 1ac4018 1d40f83 1ee4b73 1ac4018 e690b8a 1ac4018 aa9d7ef 1ac4018 aa9d7ef 1ac4018 aa9d7ef a16ea65 1ac4018 621db44 1ac4018 aa9d7ef 78683da 57062ef 78683da 57062ef 78683da 128b743 1ee4b73 fd4136c 24a2b9a e690b8a 24a2b9a c27f13b 5a521e1 94aba93 ee51c1e 5a521e1 81384e9 6882935 a3a369c 9e24d40 621db44 721ea62 6882935 721ea62 6882935 94aba93 71852c2 aa9d7ef 24158f9 57062ef 24158f9 07cc877 57062ef 78683da 57062ef 78683da bebb895 aa9d7ef 1ac4018 57062ef 1ac4018 57062ef 1ac4018 57062ef 1ac4018 57ca910 1d40f83 1ac4018 aa9d7ef 57062ef aa9d7ef 57062ef aa9d7ef 57062ef aa9d7ef 5f35567 aa9d7ef 1ac4018 cab4101 94aba93 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 |
import gradio as gr
import random
import os
from huggingface_hub import login
from transformers import pipeline
from transformers import GPT2Tokenizer, GPT2LMHeadModel
login(os.environ["HF_TOKEN"])
#https://huggingface.co/facebook/opt-1.3b
#generator = pipeline('text-generation', model="microsoft/DialoGPT-medium")
tokenizer = GPT2Tokenizer.from_pretrained('microsoft/DialoGPT-medium')
original_model = GPT2LMHeadModel.from_pretrained('microsoft/DialoGPT-medium')
untethered_model = GPT2LMHeadModel.from_pretrained('zmbfeng/untethered_20240225_epochs_500')
untethered_paraphrased_model = GPT2LMHeadModel.from_pretrained('zmbfeng/untethered_20240227_epochs_350')
def create_response_untethered_paraphrased(input_str,
# num_beams,
num_return_sequences,
temperature,
repetition_penalty,
top_p,
# top_k,
do_sample):
print("input_str="+input_str)
# num_beams = int(num_beams)
# print("num_beams=" + str(num_beams))
num_return_sequences=int(num_return_sequences)
print("num_return_sequences" + str(num_return_sequences))
print("top_p" + str(top_p))
# top_k=int(top_k)
#print("top_k" + str(top_k))
print("repetition_penalty" + str(repetition_penalty))
print("temperature" + str(temperature))
print("do_sample" + str(do_sample))
if not do_sample:
num_beams = 1
print("num_beams=" + str(num_beams))
#output_raw= generator(input_str)
"""print (output_raw)"""
#output_str = output_raw[0]['generated_text']
#output_str = output_str.replace("\n", "")
#output_str = output_str.replace(input_str, "")
#output_str = tokenizer.decode(model.generate(**tokenizer("What are John West's hobbies?"+tokenizer.eos_token,return_tensors="pt",max_length=200))[0])
# output_str = tokenizer.decode(original_model.generate(**tokenizer(input_str+tokenizer.eos_token,return_tensors="pt",max_length=200),
# num_beams=num_beams,
# num_return_sequences=num_return_sequences)[0])
#input_ids = tokenizer.encode(input_str + tokenizer.eos_token, return_tensors="pt")
encoded = tokenizer.encode_plus(input_str + tokenizer.eos_token, return_tensors="pt")
input_ids = encoded["input_ids"]
attention_mask = encoded["attention_mask"]
#output_ids = fine_tuned_model.generate(input_ids,do_sample=True, max_length=100, temperature=0.2, top_p=0.9, repetition_penalty=1.5,num_return_sequences=6)
#output_ids = untethered_paraphrased_model.generate(input_ids,do_sample=do_sample, max_length=100, temperature=temperature, top_p=top_p, top_k=top_k, repetition_penalty=repetition_penalty,num_return_sequences=num_return_sequences, num_beams = num_beams)
output_ids = untethered_paraphrased_model.generate(input_ids,pad_token_id=tokenizer.eos_token_id,do_sample=True,attention_mask=attention_mask, max_length=100, temperature=temperature, top_p=top_p, repetition_penalty=repetition_penalty,num_return_sequences=num_return_sequences)
outputs = ""
for output_id in output_ids:
output = tokenizer.decode(output_id, skip_special_tokens=True)
outputs=outputs+output+"<br/>"
return outputs
def create_response_untethered(input_str,
# num_beams,
num_return_sequences,
temperature,
repetition_penalty,
top_p,
# top_k,
do_sample,
model_name):
print("input_str="+input_str)
print("model_name="+str(model_name))
# num_beams = int(num_beams)
# print("num_beams=" + str(num_beams))
num_return_sequences=int(num_return_sequences)
print("num_return_sequences" + str(num_return_sequences))
print("top_p" + str(top_p))
# top_k=int(top_k)
# print("top_k" + str(top_k))
print("repetition_penalty" + str(repetition_penalty))
print("temperature" + str(temperature))
print("do_sample" + str(do_sample))
if not do_sample:
num_beams = 1
print("num_beams=" + str(num_beams))
#output_raw= generator(input_str)
"""print (output_raw)"""
#output_str = output_raw[0]['generated_text']
#output_str = output_str.replace("\n", "")
#output_str = output_str.replace(input_str, "")
#output_str = tokenizer.decode(model.generate(**tokenizer("What are John West's hobbies?"+tokenizer.eos_token,return_tensors="pt",max_length=200))[0])
# output_str = tokenizer.decode(original_model.generate(**tokenizer(input_str+tokenizer.eos_token,return_tensors="pt",max_length=200),
# num_beams=num_beams,
# num_return_sequences=num_return_sequences)[0])
#input_ids = tokenizer.encode(input_str + tokenizer.eos_token, return_tensors="pt")
encoded = tokenizer.encode_plus(input_str + tokenizer.eos_token, return_tensors="pt")
input_ids = encoded["input_ids"]
attention_mask = encoded["attention_mask"]
#output_ids = fine_tuned_model.generate(input_ids,do_sample=True, max_length=100, temperature=0.2, top_p=0.9, repetition_penalty=1.5,num_return_sequences=6)
output_ids = untethered_model.generate(input_ids,pad_token_id=tokenizer.eos_token_id,do_sample=do_sample, attention_mask=attention_mask, max_length=100, temperature=temperature, top_p=top_p, repetition_penalty=repetition_penalty,num_return_sequences=num_return_sequences )
outputs = ""
for output_id in output_ids:
output = tokenizer.decode(output_id, skip_special_tokens=True)
outputs=outputs+output+"\\n"
return outputs
def create_response_original(input_str,
# num_beams,
num_return_sequences,
temperature,
repetition_penalty,
top_p,
# top_k,
do_sample):
print("input_str="+input_str)
# num_beams = int(num_beams)
# print("num_beams=" + str(num_beams))
num_return_sequences=int(num_return_sequences)
print("num_return_sequences" + str(num_return_sequences))
print("top_p" + str(top_p))
# top_k=int(top_k)
# print("top_k" + str(top_k))
print("repetition_penalty" + str(repetition_penalty))
print("temperature" + str(temperature))
print("do_sample" + str(do_sample))
if not do_sample:
num_beams = 1
print("num_beams=" + str(num_beams))
#output_raw= generator(input_str)
"""print (output_raw)"""
#output_str = output_raw[0]['generated_text']
#output_str = output_str.replace("\n", "")
#output_str = output_str.replace(input_str, "")
#output_str = tokenizer.decode(model.generate(**tokenizer("What are John West's hobbies?"+tokenizer.eos_token,return_tensors="pt",max_length=200))[0])
# output_str = tokenizer.decode(original_model.generate(**tokenizer(input_str+tokenizer.eos_token,return_tensors="pt",max_length=200),
# num_beams=num_beams,
# num_return_sequences=num_return_sequences)[0])
#input_ids = tokenizer.encode(input_str + tokenizer.eos_token, return_tensors="pt")
encoded = tokenizer.encode_plus(input_str + tokenizer.eos_token, return_tensors="pt")
input_ids = encoded["input_ids"]
attention_mask = encoded["attention_mask"]
#output_ids = fine_tuned_model.generate(input_ids,do_sample=True, max_length=100, temperature=0.2, top_p=0.9, repetition_penalty=1.5,num_return_sequences=6)
output_ids = original_model.generate(input_ids,pad_token_id=tokenizer.eos_token_id,do_sample=do_sample,attention_mask=attention_mask, max_length=100, temperature=temperature, top_p=top_p, repetition_penalty=repetition_penalty,num_return_sequences=num_return_sequences)
outputs = ""
for output_id in output_ids:
output = tokenizer.decode(output_id, skip_special_tokens=True)
outputs=outputs+output+"\\n"
return outputs
def create_response_fine_tuned(input_str):
#output_raw= generator(input_str)
"""print (output_raw)"""
#output_str = output_raw[0]['generated_text']
#output_str = output_str.replace("\n", "")
#output_str = output_str.replace(input_str, "")
#output_str = tokenizer.decode(model.generate(**tokenizer("What are John West's hobbies?"+tokenizer.eos_token,return_tensors="pt",max_length=200))[0])
output_str = tokenizer.decode(fine_tuned_model.generate(**tokenizer(input_str+tokenizer.eos_token,return_tensors="pt",max_length=200))[0])
return (output_str)
interface1 = gr.Interface(fn=create_response_original,
title="original",
description="original language model, no fine tuning",
examples=[
["What is death?",5,0.2,1.5,0.9,True], # The first example
["One of the best teachers in all of life turns out to be what?",5,0.2,1.5,0.9,True], # The second example
["what is your most meaningful relationship?",5,0.2,1.5,0.9,True], # The third example
["What actually gives life meaning?",5,0.2,1.5,0.9,True]
],
inputs=[
gr.Textbox(label="input text here", lines=3),
# gr.Number(label="num_beams (integer) explores the specified number of possible outputs and selects the most " +
# "likely ones (specified in num_beams)", value=7),
gr.Number(label="num_return_sequences (integer) the number of outputs selected from num_beams possible output",
value=5),
gr.Number(
label="temperature (decimal) controls the creativity or randomness of the output. A higher temperature" +
" (e.g., 0.9) results in more diverse and creative output, while a lower temperature (e.g., 0.2)" +
" makes the output more deterministic and focused",
value=0.2),
gr.Number(label="repetition_penalty (decimal) penalizes words that have already appeared in the output, " +
"making them less likely to be generated again. A higher repetition_penalty (e.g., 1.5) results" +
"in more varied and non-repetitive output.",
value=1.5),
gr.Number(label="top_p (decimal) the model will only consider the words that have a high enough probability" +
" to reach a certain threshold",
value=0.9),
# gr.Number(label="top_k (integer) The number of highest probability vocabulary word will be considered" +
# "This means that only the tokens with the highest probabilities are considered for sampling" +
# "This reduces the diversity of the generated sequences, "+
# "but also makes them more likely to be coherent and fluent.",
# value=50),
gr.Checkbox(label="do_sample. If is set to False, num_return_sequences must be 1 because the generate function will use greedy decoding, " +
"which means that it will select the word with the highest probability at each step. " +
"This results in a deterministic and fluent output, but it might also lack diversity and creativity" +
"If is set to True, the generate function will use stochastic sampling, which means that it will randomly" +
" select a word from the probability distribution at each step. This results in a more diverse and creative" +
" output, but it might also introduce errors and inconsistencies ", value=True)
], outputs=[gr.Textbox(label="output response", lines=30)])
interface2 = gr.Interface(fn=create_response_untethered,
title="untethered",
description="untethered fine tuning",
examples=[
["What is death?",5,0.2,1.5,0.9,True], # The first example
["One of the best teachers in all of life turns out to be what?",5,0.2,1.5,0.9,True], # The second example
["what is your most meaningful relationship?",5,0.2,1.5,0.9,True], # The third example
["What actually gives life meaning?",5,0.2,1.5,0.9,True]
],
inputs=[
gr.Textbox(label="input text here", lines=3),
# gr.Number(label="num_beams (integer) explores the specified number of possible outputs and selects the most " +
# "likely ones (specified in num_beams)", value=7),
gr.Number(label="num_return_sequences (integer) the number of outputs selected from num_beams possible output",
value=5),
gr.Number(
label="temperature (decimal) controls the creativity or randomness of the output. A higher temperature" +
" (e.g., 0.9) results in more diverse and creative output, while a lower temperature (e.g., 0.2)" +
" makes the output more deterministic and focused",
value=0.2),
gr.Number(label="repetition_penalty (decimal) penalizes words that have already appeared in the output, " +
"making them less likely to be generated again. A higher repetition_penalty (e.g., 1.5) results" +
"in more varied and non-repetitive output.",
value=1.5),
gr.Number(label="top_p (decimal) the model will only consider the words that have a high enough probability" +
" to reach a certain threshold",
value=0.9),
# gr.Number(label="top_k (integer) The number of highest probability vocabulary word will be considered" +
# "This means that only the tokens with the highest probabilities are considered for sampling" +
# "This reduces the diversity of the generated sequences, "+
# "but also makes them more likely to be coherent and fluent.",
# value=50),
gr.Checkbox(label="do_sample. If is set to False, num_return_sequences must be 1 because the generate function will use greedy decoding, " +
"which means that it will select the word with the highest probability at each step. " +
"This results in a deterministic and fluent output, but it might also lack diversity and creativity" +
"If is set to True, the generate function will use stochastic sampling, which means that it will randomly" +
" select a word from the probability distribution at each step. This results in a more diverse and creative" +
" output, but it might also introduce errors and inconsistencies ", value=True),
gr.Textbox(label="model", lines=3, value="untethered_model_name")
#,visible=False
], outputs=[gr.Textbox(label="output response", lines=30)])
interface3 = gr.Interface(fn=create_response_untethered_paraphrased,
title="untethered paraphrased",
description="untethered paraphrased fine tuning",
examples=[
["What is death?",5,0.2,1.5,0.9,True], # The first example
["One of the best teachers in all of life turns out to be what?",5,0.2,1.5,0.9,True], # The second example
["what is your most meaningful relationship?",5,0.2,1.5,0.9,True], # The third example
["What actually gives life meaning?",5,0.2,1.5,0.9,True]
],
inputs=[
gr.Textbox(label="input text here", lines=3),
# gr.Number(label="num_beams (integer) explores the specified number of possible outputs and selects the most " +
# "likely ones (specified in num_beams)", value=7),
gr.Number(label="num_return_sequences (integer) the number of outputs selected from num_beams possible output",
value=5),
gr.Number(
label="temperature (decimal) controls the creativity or randomness of the output. A higher temperature" +
" (e.g., 0.9) results in more diverse and creative output, while a lower temperature (e.g., 0.2)" +
" makes the output more deterministic and focused",
value=0.2),
gr.Number(label="repetition_penalty (decimal) penalizes words that have already appeared in the output, " +
"making them less likely to be generated again. A higher repetition_penalty (e.g., 1.5) results" +
"in more varied and non-repetitive output.",
value=1.5),
gr.Number(label="top_p (decimal) the model will only consider the words that have a high enough probability" +
" to reach a certain threshold",
value=0.9),
# gr.Number(label="top_k (integer) The number of highest probability vocabulary word will be considered" +
# "This means that only the tokens with the highest probabilities are considered for sampling" +
# "This reduces the diversity of the generated sequences, "+
# "but also makes them more likely to be coherent and fluent.",
# value=50),
gr.Checkbox(label="do_sample. If is set to False, num_return_sequences must be 1 because the generate function will use greedy decoding, " +
"which means that it will select the word with the highest probability at each step. " +
"This results in a deterministic and fluent output, but it might also lack diversity and creativity" +
"If is set to True, the generate function will use stochastic sampling, which means that it will randomly" +
" select a word from the probability distribution at each step. This results in a more diverse and creative" +
" output, but it might also introduce errors and inconsistencies ", value=True)
], outputs="html")
#[gr.Textbox(label="output response", lines=30)]
#interface2 = gr.Interface(fn=create_response_fine_tuned, inputs="text", outputs="text", title="Fine Tuned")
demo = gr.TabbedInterface([interface1, interface2, interface3], ["Original", "Untethered", "Untethered paraphrased"])
#demo = gr.TabbedInterface([interface1, interface3], ["Original", "Untethered paraphrased"])
# with gr.Blocks() as demo:
# with gr.Row():
#
demo.launch() |