File size: 8,033 Bytes
9e85aff
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e4bccbf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9e85aff
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e4bccbf
 
9e85aff
 
 
 
 
 
 
 
 
 
e4bccbf
 
 
 
 
9e85aff
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e4bccbf
 
 
 
9e85aff
e4bccbf
9e85aff
 
 
 
 
 
e4bccbf
9e85aff
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import gradio as gr
import inseq
import captum

import torch
import os
# import nltk
import argparse
import random
import numpy as np

from argparse import Namespace
from tqdm.notebook import tqdm
from torch.utils.data import DataLoader
from functools import partial

from transformers import AutoTokenizer, MarianTokenizer, AutoModel, AutoModelForSeq2SeqLM, MarianMTModel

model_es = "Helsinki-NLP/opus-mt-en-es"
model_fr = "Helsinki-NLP/opus-mt-en-fr"
model_zh = "Helsinki-NLP/opus-mt-en-zh"

tokenizer_es = AutoTokenizer.from_pretrained(model_es)
tokenizer_fr = AutoTokenizer.from_pretrained(model_fr)
tokenizer_zh = AutoTokenizer.from_pretrained(model_zh)

model_tr_es = MarianMTModel.from_pretrained(model_es)
model_tr_fr = MarianMTModel.from_pretrained(model_fr)
model_tr_zh = MarianMTModel.from_pretrained(model_zh)

model_es = inseq.load_model("Helsinki-NLP/opus-mt-en-es", "input_x_gradient")
model_fr = inseq.load_model("Helsinki-NLP/opus-mt-en-fr", "input_x_gradient")
model_zh = inseq.load_model("Helsinki-NLP/opus-mt-en-zh", "input_x_gradient")


dict_models = {
	'en-es': model_es,
	'en-fr': model_fr,
	'en-zh': model_zh,
}

dict_models_tr = {
	'en-es': model_tr_es,
	'en-fr': model_tr_fr,
	'en-zh': model_tr_zh,
}

dict_tokenizer_tr = {
	'en-es': tokenizer_es,
	'en-fr': tokenizer_fr,
	'en-zh': tokenizer_zh,
}

saliency_examples = [
	"Peace of Mind: Protection for consumers.",
	"The sustainable development goals report: towards a rescue plan for people and planet",
	"We will leave no stone unturned to hold those responsible to account.",
	"The clock is now ticking on our work to finalise the remaining key legislative proposals presented by this Commission to ensure that citizens and businesses can reap the benefits of our policy actions.",
	"Pumpkins, squash and gourds, fresh or chilled, excluding courgettes",
	"The labour market participation of mothers with infants has even deteriorated over the past two decades, often impacting their career and incomes for years.",
]

contrastive_examples = [
["Peace of Mind: Protection for consumers.",
"Paz mental: protección de los consumidores",
"Paz de la mente: protección de los consumidores"],
["the slaughterer has finished his work.",
"l'abatteur a terminé son travail.",
"l'abatteuse a terminé son travail."],
['A fundamental shift is needed - in commitment, solidarity, financing and action - to put the world on a better path.',
 '需要在承诺、团结、筹资和行动方面进行根本转变,使世界走上更美好的道路。',
 '我们需要从根本上转变承诺、团结、资助和行动,使世界走上更美好的道路。',]
	]


def get_k_prob_tokens(transition_scores, result, model, k_values=5):
	tokenizer_tr = dict_tokenizer_tr[model]
	gen_sequences = result.sequences[:, 1:]

	result_output = []
	# bs_alt = []
	# bs_alt_scores = []

	# First beam only... 
	bs = 0
	text = ' '
	for tok, score, i_step in zip(gen_sequences[bs], transition_scores[bs],range(len(gen_sequences[bs]))):
		# bs_alt.append([tokenizer_tr.decode(tok) for tok in result.scores[i_step][bs].topk(k_values).indices ] )
		# bs_alt_scores.append(np.exp(result.scores[i_step][bs].topk(k_values).values))

		bs_alt = [tokenizer_tr.decode(tok) for tok in result.scores[i_step][bs].topk(k_values).indices ]
		bs_alt_scores = np.exp(result.scores[i_step][bs].topk(k_values).values)
		result_output.append([np.array(result.scores[i_step][bs].topk(k_values).indices), np.array(bs_alt_scores),bs_alt])

	return result_output


def split_token_from_sequences(sequences, model) -> dict :
	n_sentences = len(sequences)

	gen_sequences_texts = []
	for bs in range(n_sentences): 
		#### decoder per token.
		gen_sequences_texts.append(dict_tokenizer_tr[model].decode(sequences[:, 1:][bs],  skip_special_tokens=True).split(' '))
	print(gen_sequences_texts)
	score = 0

	#raw dict is bos
	text = 'bos'
	new_id = text +'--1'
	dict_parent = [{'id': new_id, 'parentId': None , 'text': text, 'name': 'bos', 'prob':score }]
	id_dict_pos = {}
	step_i = 0
	cont = True
	words_by_step = [] #[['bos' for i in range(n_sentences)]]

	while cont: 
		# append to dict_parent for all beams of step_i
		cont = False
		step_words = []
		for beam in range(n_sentences):
			app_text = ''
			if step_i < len(gen_sequences_texts[beam]): 
				app_text = gen_sequences_texts[beam][step_i]
				cont = True
			step_words.append(app_text)
		words_by_step.append(step_words)
		print(words_by_step)

		for i_bs, step_w in enumerate(step_words):
			if step_w != '':
				#new id if the same word is not in another beam (?) [beam[i] was a token id]
				#parent id = previous word and previous step.

				
				# new_parent_id = "-".join([str(beam[i]) for i in range(step_i)])
				
				new_id = "-".join([str(words_by_step[i][i_bs])+ '-' + str(i) for i in range(step_i+1)])
				parent_id = "-".join([words_by_step[i][i_bs] + '-' + str(i) for i in range(step_i) ])
				
				# new_id = step_w +'-' + str(step_i)
				# parent_id = words_by_step[step_i-1][i_bs] + '-' + str(step_i -1)
				
				if step_i == 0 : 
					parent_id =  'bos--1'
				## if the dict already exists remove it, if it is not a root... 
				## root?? then next is ''
				next_word_flag = len(gen_sequences_texts[i_bs][step_i]) > step_i 
				if next_word_flag:
					if not (new_id in id_dict_pos):
						dict_parent.append({'id': new_id, 'parentId': parent_id , 'text': step_w, 'name': step_w, 'prob' : score })
						id_dict_pos[new_id] = len(dict_parent) - 1
				else: 
					dict_parent.append({'id': new_id, 'parentId': parent_id , 'text': step_w, 'name': step_w, 'prob' : score })
					id_dict_pos[new_id] = len(dict_parent) - 1

		step_i += 1
	return dict_parent





html = """
<html>
<script async src="https://cdnjs.cloudflare.com/ajax/libs/require.js/2.3.6/require.min.js"></script>
  <body>
    
    <p id="demo"></p>
    <p id="viz"></p>

    <p id="demo2"></p>
    <h4> Exploring top-k probable tokens </h4>
    <div id="d3_text_grid">... top 10 tokens generated at each step ...</div>

    <h4> Exploring the Beam Search sequence generation</h4>
    <div id="d3_beam_search">... top 4 generated sequences using Beam Search...</div>



  </body>
</html>
"""


def sentence_maker(w1, model, var2={}):
  #translate and get internal values
  # src_text = saliency_examples[0]
  inputs = dict_tokenizer_tr[model](w1, return_tensors="pt")

  num_ret_seq = 4
  translated  = dict_models_tr[model].generate(**inputs, 
                  num_beams=4,
                  num_return_sequences=num_ret_seq,
                  return_dict_in_generate=True, 
                  output_attentions =True,  
                  output_hidden_states = True, 
                  output_scores=True,)

  beam_dict = split_token_from_sequences(translated.sequences,model )

  tgt_text = dict_tokenizer_tr[model].decode(translated.sequences[0], skip_special_tokens=True)
  transition_scores = dict_models_tr[model].compute_transition_scores(
	translated.sequences, translated.scores, translated.beam_indices  , normalize_logits=True
	)
  prob_tokens = get_k_prob_tokens(transition_scores, translated, model, k_values=10)

  return [tgt_text,[beam_dict,prob_tokens]]

def sentence_maker2(w1,j2):
   print(w1,j2)
   return "in sentence22..."



with gr.Blocks(js="plotsjs.js") as demo:
	gr.Markdown(
	"""
	# MAKE NMT Workshop \t `BeamSearch` 
	""")
	in_text = gr.Textbox(label="source text")
	out_text  = gr.Textbox(label="target text")
	out_text2  = gr.Textbox(visible=False)
	var2 = gr.JSON(visible=False)
	radio_c = gr.Radio(choices=['en-zh', 'en-es', 'en-fr'], value="en-zh", label= '', container=False)
	btn = gr.Button("Translate")
	input_mic = gr.HTML(html)


	btn.click(sentence_maker, [in_text, radio_c], [out_text,var2], js="(in_text,radio_c) => testFn_out(in_text,radio_c)") #should return an output comp.
	out_text.change(sentence_maker2, [out_text, var2], out_text2, js="(out_text,var2) => testFn_out_json(var2)") #

	# run script function on load,
	# demo.load(None,None,None,js="plotsjs.js")

if __name__ == "__main__":   
    demo.launch()