app.py

import gradio as gr
from time import time

import torch
import os
# import nltk
import argparse
import random
import numpy as np
import faiss
from argparse import Namespace
from tqdm.notebook import tqdm
from torch.utils.data import DataLoader
from functools import partial
from sklearn.manifold import TSNE

from transformers import AutoTokenizer, MarianTokenizer, AutoModel, AutoModelForSeq2SeqLM, MarianMTModel
import os 
dir_path = os.path.dirname(os.path.realpath(__file__))
print(dir_path)

metadata_all = {} 
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)

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,
}

from faiss import write_index, read_index
import pickle 



def translation_model(w1,model ):
	inputs = dict_tokenizer_tr[model](w1, return_tensors="pt")
	# embeddings = get_tokens_embeddings(inputs, model)
	input_embeddings = dict_models_tr[model].get_encoder().embed_tokens(inputs.input_ids)
	# model_tr_es.get_input_embeddings()
	print(inputs)
	num_ret_seq = 1
	translated  = dict_models_tr[model].generate(**inputs, 
											  num_beams=5,
											  num_return_sequences=num_ret_seq,
											  return_dict_in_generate=True, 
											  output_attentions =False,  
											  output_hidden_states = True,
											  output_scores=True,)

	tgt_text = dict_tokenizer_tr[model].decode(translated.sequences[0], skip_special_tokens=True)
	
	target_embeddings = dict_models_tr[model].get_decoder().embed_tokens(translated.sequences)

	return tgt_text, translated, inputs.input_ids, input_embeddings, target_embeddings

def create_vocab_multiple(embeddings_list, model): 
	"""_summary_

	Args:
		embeddings_list (list): embedding array 

	Returns:
		Dict: vocabulary of tokens' embeddings
	"""
	print("START VOCAB CREATION MULTIPLE \n \n ")
	vocab = {} ## add embedds. 
	sentence_tokens_text_list = []
	for embeddings in embeddings_list: 
		tokens_id = embeddings['tokens'] # [[tokens_id]x n_sentences ]
		for sent_i, sentence in enumerate(tokens_id):
			sentence_tokens = []
			for tok_i, token in enumerate(sentence): 
				sentence_tokens.append(token)
				if not (token in vocab):
					vocab[token] = {
						'token' : token,
						'count': 1, 
						# 'text': embeddings['texts'][sent_i][tok_i],
						'text': dict_tokenizer_tr[model].decode([token]),
						# 'text': src_token_lists[sent_i][tok_i], 
						'embed': embeddings['embeddings'][sent_i][tok_i]}
				else: 
					vocab[token]['count'] = vocab[token]['count'] + 1  
		# print(vocab)
			sentence_tokens_text_list.append(sentence_tokens)
	print("END VOCAB CREATION MULTIPLE \n \n ")
	return vocab, sentence_tokens_text_list

def vocab_words_all_prefix(token_embeddings, model, sufix="@@",prefix = '▁' ):
	vocab = {} 
	# inf_model = dict_models_tr[model]
	sentence_words_text_list = []
	if prefix : 
		n_prefix = len(prefix)
		for input_sentences in token_embeddings: 
			# n_tokens_in_word 
			for sent_i, sentence in enumerate(input_sentences['tokens']):
				words_text_list = []
				# embedding = input_sentences['embed'][sent_i]
				word = '' 
				tokens_ids = []
				embeddings = []
				ids_to_tokens = dict_tokenizer_tr[model].convert_ids_to_tokens(sentence)
				# print("validate same len", len(sentence) == len(ids_to_tokens), len(sentence), len(ids_to_tokens), ids_to_tokens)

				to_save= False
				for tok_i, token_text in enumerate(ids_to_tokens): 
					token_id = sentence[tok_i]
					if token_text[:n_prefix] == prefix : 
						#first we save the previous word 
						if to_save: 
							vocab[word] = {
									'word' : word,
									'text': word,
									'count': 1, 
									'tokens_ids' : tokens_ids, 
									'embed': np.mean(np.array(embeddings), 0).tolist()
								}
							words_text_list.append(word)
						#word is starting if prefix
						tokens_ids = [token_id]
						embeddings = [input_sentences['embeddings'][sent_i][tok_i]]
						word = token_text[n_prefix:]
						## if word 
						to_save = True 
						
					else : 
						if (token_text in dict_tokenizer_tr[model].special_tokens_map.values()):
							# print('final or save', token_text, token_id, to_save, word)
							if to_save: 
								# vocab[word] = ids
								vocab[word] = {
									'word' : word,
									'text': word,
									'count': 1, 
									'tokens_ids' : tokens_ids, 
									'embed': np.mean(np.array(embeddings), 0).tolist()
								}
								words_text_list.append(word)
							#special token is one token element, no continuation 
							# vocab[token_text] = [token_id]
							tokens_ids = [token_id]
							embeddings = [input_sentences['embeddings'][sent_i][tok_i]]
							vocab[token_text] = {
									'word' : token_text,
									'count': 1, 
									'text': word,
									'tokens_ids' : tokens_ids, 
									'embed': np.mean(np.array(embeddings), 0).tolist()
								}
							words_text_list.append(token_text)
							to_save = False
						else: 
							# is a continuation; we do not know if it is final; we don't save here.
							to_save = True 
							word += token_text 
							tokens_ids.append(token_id)
							embeddings.append(input_sentences['embeddings'][sent_i][tok_i])
				if to_save: 
					# print('final save', token_text, token_id, to_save, word)
					vocab[word] = tokens_ids
					if not (word in vocab):
						vocab[word] = {
							'word' : word,
							'count': 1, 
							'text': word,
							'tokens_ids' : tokens_ids, 
							'embed': np.mean(np.array(embeddings), 0).tolist()
							}
						words_text_list.append(word)
					else: 
						vocab[word]['count'] = vocab[word]['count'] + 1 
				sentence_words_text_list.append(words_text_list)

	return vocab, sentence_words_text_list

# nb_ids.append(token_values['token']) # for x in vocab_tokens]
# nb_embds.append(token_values['embed']) # for x in vocab_tokens]

def create_index_voronoi(vocab):
	"""
	it returns an index of words and a metadata of ids. 
	"""
	d = 1024
	nb_embds = [] ##ordered embeddings list
	metadata = {}
	i_pos = 0
	for key_token, token_values in vocab.items():
		nb_embds.append(token_values['embed']) # for x in vocab_tokens]
		metadata[i_pos] = {'token': token_values['token'], 'text': token_values['text']}
		i_pos += 1
	# nb_embds = [x['embed'] for x in vocab_tokens]

	# print(len(nb_embds),len(nb_embds[0]) )
	xb = np.array(nb_embds).astype('float32') #elements to index
	# ids = np.array(nb_ids)
	d = len(xb[0]) # dimension of each element

	nlist = 5 # Nb of Voronois 
	quantizer = faiss.IndexFlatL2(d)
	index = faiss.IndexIVFFlat(quantizer, d, nlist)
	index.train(xb)
	index.add(xb)
	# index.add(xb) 
	
	return index, metadata## , nb_embds, nb_ids 

def create_index_voronoi_words(vocab):
	"""
	it returns an index of words and a metadata of ids. 
	"""
	d = 1024
	nb_embds = [] ##ordered embeddings list
	metadata = {}
	i_pos = 0
	for key_token, token_values in vocab.items():
		nb_embds.append(token_values['embed']) # for x in vocab_tokens]
		metadata[i_pos] = {'word': token_values['word'], 'tokens': token_values['tokens_ids'],'text': token_values['text']}
		i_pos += 1
	# nb_embds = [x['embed'] for x in vocab_tokens]

	# print(len(nb_embds),len(nb_embds[0]) )
	xb = np.array(nb_embds).astype('float32') #elements to index
	# ids = np.array(nb_ids)
	d = len(xb[0]) # dimension of each element

	nlist = 5 # Nb of Voronois 
	quantizer = faiss.IndexFlatL2(d)
	index = faiss.IndexIVFFlat(quantizer, d, nlist)
	index.train(xb)
	index.add(xb)
	# index.add(xb) 
	
	return index, metadata## , nb_embds, nb_ids 

def search_query_vocab(index, vocab_queries,  topk = 10, limited_search = []):
	""" the embed queries are a vocabulary of words : embds_input_voc

	Args:
		index (_type_): faiss index
		embed_queries (_type_): vocab format.
			{   'token' : token,
				'count': 1, 
				'text': src_token_lists[sent_i][tok_i], 
				'embed': embeddings[0]['embeddings'][sent_i][tok_i] }
		nb_ids (_type_): hash to find the token_id w.r.t the faiss index id. 
		topk (int, optional): nb of similar tokens. Defaults to 10.

	Returns:
		_type_: Distance matrix D, indices matrix I and tokens ids (using nb_ids)
	"""
	# nb_qi_ids = [] ##ordered ids list
	nb_q_embds = [] ##ordered embeddings list
	metadata = {}
	qi_pos = 0
	for key , token_values in vocab_queries.items():
		# nb_qi_ids.append(token_values['token']) # for x in vocab_tokens]
		metadata[qi_pos] = {'word': token_values['word'], 'tokens': token_values['tokens_ids'], 'text': token_values['text']}
		qi_pos += 1
		nb_q_embds.append(token_values['embed']) # for x in vocab_tokens]
	
	xq = np.array(nb_q_embds).astype('float32') #elements to query

	D,I = index.search(xq, topk)

	return D,I, metadata

def search_query_vocab_token(index, vocab_queries,  topk = 10, limited_search = []):
	""" the embed queries are a vocabulary of words : embds_input_vov
	Returns:
		_type_: Distance matrix D, indices matrix I and tokens ids (using nb_ids)
	"""
	# nb_qi_ids = [] ##ordered ids list
	nb_q_embds = [] ##ordered embeddings list
	metadata = {}
	qi_pos = 0
	for key , token_values in vocab_queries.items():
		# nb_qi_ids.append(token_values['token']) # for x in vocab_tokens]
		metadata[qi_pos] = {'token': token_values['token'], 'text': token_values['text']}
		qi_pos += 1
		nb_q_embds.append(token_values['embed']) # for x in vocab_tokens]
	
	xq = np.array(nb_q_embds).astype('float32') #elements to query

	D,I = index.search(xq, topk)

	return D,I, metadata		

def build_search(query_embeddings, model,type="input"):
	global metadata_all 

	# ## biuld vocab for index 
	vocab_queries, sentence_tokens_list = create_vocab_multiple(query_embeddings, model)
	words_vocab_queries, sentence_words_list = vocab_words_all_prefix(query_embeddings, model, sufix="@@",prefix="▁")
	
	index_vor_tokens = metadata_all[type]['tokens'][1]
	md_tokens = metadata_all[type]['tokens'][2]
	D, I, meta = search_query_vocab_token(index_vor_tokens, vocab_queries)

	qi_pos = 0 
	similar_tokens = {}
	# similar_tokens = []
	for dist, ind in zip(D,I):
		try: 
			# similar_tokens.append({
			similar_tokens[str(meta[qi_pos]['token'])] = {
				'token': meta[qi_pos]['token'], 
				'text': meta[qi_pos]['text'], 
				# 'text': dict_tokenizer_tr[model].decode(meta[qi_pos]['token'])
				# 'text': meta[qi_pos]['text'], 
				"similar_topk": [md_tokens[i_index]['token'] for i_index in ind if (i_index != -1) ], 
				"distance": [dist[i] for (i, i_index) in enumerate(ind) if (i_index != -1)], 
				}
			# )
		except: 
			print("\n ERROR ", qi_pos, dist, ind)
		qi_pos += 1


	index_vor_words = metadata_all[type]['words'][1]
	md_words = metadata_all[type]['words'][2]

	Dw, Iw, metaw = search_query_vocab(index_vor_words, words_vocab_queries)
	# D, I, meta, vocab_words, sentence_words_list = result_input['words']# [2] # D ; I ; meta
	qi_pos = 0 
	# similar_words = []
	similar_words = {}
	for dist, ind in zip(Dw,Iw):
		try: 
			# similar_words.append({
			similar_words[str(metaw[qi_pos]['word']) ] = {
				'word': metaw[qi_pos]['word'], 
				'text': metaw[qi_pos]['word'], 
				"similar_topk": [md_words[i_index]['word'] for i_index in ind if (i_index != -1) ], 
				"distance": [dist[i] for (i, i_index) in enumerate(ind) if (i_index != -1)], 
				}
			# )
		except: 
			print("\n ERROR ", qi_pos, dist, ind)
		qi_pos += 1


	return {'tokens': {'D': D, 'I': I, 'meta': meta, 'vocab_queries': vocab_queries, 'similar':similar_tokens, 'sentence_key_list': sentence_tokens_list}, 
			'words': {'D':Dw,'I': Iw, 'meta': metaw, 'vocab_queries':words_vocab_queries, 'sentence_key_list': sentence_words_list, 'similar': similar_words}
			}

def build_reference(all_embeddings, model):
	
	# ## biuld vocab for index 
	vocab, sentence_tokens = create_vocab_multiple(all_embeddings,model)
	words_vocab, sentences = vocab_words_all_prefix(all_embeddings, model, sufix="@@",prefix="▁")
	
	index_tokens, meta_tokens = create_index_voronoi(vocab)
	index_words, meta_words = create_index_voronoi_words(words_vocab)


	
	return {'tokens': [vocab, index_tokens, meta_tokens], 
			'words': [words_vocab, index_words, meta_words]
	 		} # , index, meta 


def embds_input_projection_vocab(vocab, key="token"): 
	t0 = time()
	
	nb_ids = [] ##ordered ids list
	nb_embds = [] ##ordered embeddings list
	nb_text = [] ##ordered embeddings list
	tnse_error = []
	for _ , token_values in vocab.items():
		tnse_error.append([0,0])
		nb_ids.append(token_values[key]) # for x in vocab_tokens]
		nb_text.append(token_values['text']) # for x in vocab_tokens]
		nb_embds.append(token_values['embed']) # for x in vocab_tokens]

	X = np.array(nb_embds).astype('float32') #elements to project 
	try:
		tsne = TSNE(random_state=0, n_iter=1000)
		tsne_results = tsne.fit_transform(X)
	
		tsne_results = np.c_[tsne_results, nb_ids, nb_text, range(len(nb_ids))] ## creates a zip array : [[TNSE[X,Y], tokenid, token_text], ...]
	except: 
		tsne_results = np.c_[tnse_error, nb_ids, nb_text, range(len(nb_ids))] ## creates a zip array : [[TNSE[X,Y], tokenid, token_text], ...]

	t1 = time()
	print("t-SNE: %.2g sec" % (t1 - t0))
	print(tsne_results)
	
	return tsne_results.tolist()

def filtered_projection(similar_key, vocab, type="input", key="word"): 
	global metadata_all
	vocab_proj = vocab.copy()
	## tnse projection Input words
	source_words_voc_similar = set()
	# for words_set in similar_key:
	for key_i in similar_key:
		words_set = similar_key[key_i]
		source_words_voc_similar.update(words_set['similar_topk'])

	print(len(source_words_voc_similar))
	# source_embeddings_filtered = {key:  metadata_all['input']['words'][0][key] for key in source_words_voc_similar}
	source_embeddings_filtered = {key_value:  metadata_all[type][key][0][key_value] for key_value in source_words_voc_similar}
	vocab_proj.update(source_embeddings_filtered)
	## 	vocab_proj add 
	try:
		result_TSNE = embds_input_projection_vocab(vocab_proj, key=key[:-1]) ## singular => without 's'
		dict_projected_embds_all = {str(embds[2]): [embds[0], embds[1], embds[2], embds[3], embds[4]] for embds in result_TSNE}
	except: 
		print('TSNE error', type, key)
		dict_projected_embds_all = {}

	

	# print(result_TSNE)
	return dict_projected_embds_all 

def first_function(w1, model):
	global metadata_all
	#translate and get internal values
	# print(w1)
	sentences = w1.split("\n")
	all_sentences = []
	translated_text = ''
	input_embeddings = []
	output_embeddings = []
	for sentence in sentences :
		# print(sentence, end=";") 
		params = translation_model(sentence, model)
		all_sentences.append(params)
		# print(len(params))
		translated_text +=  params[0] + ' \n'
		input_embeddings.append({	
			'embeddings': params[3].detach(), ## create a vocabulary with the set of embeddings 
			'tokens': params[2].tolist(), # one translation = one sentence
			# 'texts' : 	dict_tokenizer_tr[model].decode(params[2].tolist())

		}) 
		output_embeddings.append({
			'embeddings' : params[4].detach(),
			'tokens': params[1].sequences.tolist(),
			# 'texts' : 	dict_tokenizer_tr[model].decode(params[1].sequences.tolist())
		})
	# print(input_embeddings)
	# print(output_embeddings)
	
	## Build FAISS index 
	# ---> preload faiss using the respective model with a initial dataset. 
	result_input = build_reference(input_embeddings,model)
	result_output = build_reference(output_embeddings,model)
	# print(result_input, result_output)

	metadata_all = {'input': result_input, 'output': result_output}

  ### get translation

	return [translated_text, params]

def first_function_tr(w1, model, var2={}):
	global metadata_all
	#Translate and find similar tokens in token 
	print("SEARCH -- ")
	sentences = w1.split("\n")
	all_sentences = []
	translated_text = ''
	input_embeddings = []
	output_embeddings = []
	for sentence in sentences :
		# print(sentence, end=";") 
		params = translation_model(sentence, model)
		all_sentences.append(params)
		# print(len(params))
		translated_text +=  params[0] + ' \n'
		input_embeddings.append({	
			'embeddings': params[3].detach(), ## create a vocabulary with the set of embeddings 
			'tokens': params[2].tolist(), # one translation = one sentence
			# 'texts' : dict_tokenizer_tr[model].decode(params[2].tolist()[0])
		}) 
		output_embeddings.append({
			'embeddings' : params[4].detach(),
			'tokens': params[1].sequences.tolist(),
			# 'texts' : dict_tokenizer_tr[model].decode(params[1].sequences.tolist())
		})

	## Build FAISS index 
	# ---> preload faiss using the respective model with a initial dataset. 
	result_search = {}
	result_search['input'] = build_search(input_embeddings, model, type='input')
	result_search['output'] = build_search(output_embeddings, model, type='output')
	
	# D, I, meta, vocab_words, sentence_words_list = result_input['words']# [2] # D ; I ; meta
	# md = metadata_all['input']['words'][2]
	# qi_pos = 0 
	# similar_words = []
	# for dist, ind in zip(D,I):
	# 	try: 
	# 		similar_words.append({
	# 			'word': meta[qi_pos]['word'], 
	# 			"similar_topk": [md[i_index]['word'] for i_index in ind if (i_index != -1) ], 
	# 			"distance": [D[qi_pos][i] for (i, i_index) in enumerate(ind) if (i_index != -1)], 
	# 			})
	# 	except: 
	# 		print("\n ERROR ", qi_pos, dist, ind)
	# 	qi_pos += 1
	# similar_vocab_queries = similar_vocab_queries[3]
	
	# result_output = build_search(output_embeddings, model, type="output")
	## {'tokens': {'D': D, 'I': I, 'meta': meta, 'vocab_queries': vocab_queries, 'similar':similar_tokens}, 
##			'words': {'D':Dw,'I': Iw, 'meta': metaw, 'vocab_queries':words_vocab_queries, 'sentence_key_list': sentence_words_list, 'similar': similar_words}
##			}

	# print(result_input, result_output)

	
	# json_out['input']['tokens'] = {	'similar_queries' : result_input['token'][5], # similarity and distance dict. 
	# 								'tnse': dict_projected_embds_all, #projected points (all)
	# 								'key_text_list': result_input['token'][4], # current sentences keys 
	# 								}

	json_out = {'input': {'tokens': {}, 'words': {}}, 'output': {'tokens': {}, 'words': {}}}
	dict_projected = {}
	for type in ['input', 'output']:
		dict_projected[type] = {} 
		for key in ['tokens', 'words']: 
			similar_key = result_search[type][key]['similar']
			vocab = result_search[type][key]['vocab_queries']
			dict_projected[type][key] =  filtered_projection(similar_key, vocab, type=type, key=key)
			json_out[type][key]['similar_queries'] = similar_key
			json_out[type][key]['tnse'] = dict_projected[type][key]
			json_out[type][key]['key_text_list'] = result_search[type][key]['sentence_key_list']

	return [translated_text, [ json_out, json_out['output']['words'], json_out['output']['tokens']] ]


from pathlib import Path
## First create html and divs
html = """
<html>
<script async src="https://cdnjs.cloudflare.com/ajax/libs/require.js/2.3.6/require.min.js"></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/jquery/2.0.0/jquery.min"></script>
<script async data-require="d3@3.5.3" data-semver="3.5.3"
  src="//cdnjs.cloudflare.com/ajax/libs/d3/3.5.3/d3.js"></script>
<body>
  <div id="select_div">
    <select id="select_type" class="form-select" aria-label="select example" hidden>
      <option selected value="words">Words</option>
      <option value="tokens">Tokens</option>
    </select>
  </div>
  <div id="d3_embed_div">
    <div class="row">
      <div class="col-6">
        <div id="d3_embeds_input_words" class="d3_embed words"></div>
      </div>
      <div class="col-6">
        <div id="d3_embeds_output_words" class="d3_embed words"></div>
        
      </div>
      <div class="col-6">
        <div id="d3_embeds_input_tokens" class="d3_embed tokens"></div>
      </div>
      <div class="col-6">
        <div id="d3_embeds_output_tokens" class="d3_embed tokens"></div>
      </div>
    </div>
  </div>
  <div id="d3_graph_div">
    <div class="row">
      <div class="col-4">
        <div id="d3_graph_input_words" class="d3_graph words"></div>
        
      </div>
	  <div class="col-4">
	    <div id="similar_input_words" class=""></div>
    </div>
	  <div class="col-4">
        <div id="d3_graph_output_words" class="d3_graph words"></div>
        <div id="similar_output_words" class="d3_graph words"></div>
      </div>
	  </div>
	  <div class="row">
      <div class="col-6">
        <div id="d3_graph_input_tokens" class="d3_graph tokens"></div>
        <div id="similar_input_tokens" class="d3_graph tokens"></div>
      </div>
      <div class="col-6">
        <div id="d3_graph_output_tokens" class="d3_graph tokens"></div>
        <div id="similar_output_tokens" class="d3_graph tokens"></div>
      </div>
    </div>
  </div>
</body>

</html>
"""
html0 =  """
<html>
<script async src="https://cdnjs.cloudflare.com/ajax/libs/require.js/2.3.6/require.min.js"></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/jquery/2.0.0/jquery.min"></script>
<script async data-require="d3@3.5.3" data-semver="3.5.3"
  src="//cdnjs.cloudflare.com/ajax/libs/d3/3.5.3/d3.js"></script>
<body>
  <div id="select_div">
    <select id="select_type" class="form-select" aria-label="select example" hidden>
      <option selected value="words">Words</option>
      <option value="tokens">Tokens</option>
    </select>
  </div>
</body>

</html>
"""

html_col1 = """ 
      <div id="d3_graph_input_words" class="d3_graph words"></div>
      <div id="d3_graph_input_tokens" class="d3_graph tokens"></div>
	"""

html_col2 = """
 <div id="similar_input_words" class=""></div>
  <div id="similar_output_words" class=""></div>
  <div id="similar_input_tokens" class=" "></div>
<div id="similar_output_tokens" class=" "></div>

"""


html_col3 = """
<div id="d3_graph_output_words" class="d3_graph words"></div>
<div id="d3_graph_output_tokens" class="d3_graph tokens"></div>
"""


# # <div class="row"> 
#             <div class="col-6" id="d3_legend_data_source"> </div>
#             <div class="col-6" id="d3_legend_similar_source"> </div>
# </div>
def second_function(w1,j2):
	#  json_value = {'one':1}#  return f"{w1['two']} in sentence22..."
	# to transfer the data to json.
	print("second_function -- after the js", w1,j2)
	return "transition to second js function finished."	

paths = []
def save_index(model) : 
	names = []
	with open(model + '_metadata_ref.pkl', 'wb') as f:
		pickle.dump(metadata_all, f)
		names.append(model + '_metadata_ref.pkl')
	for type in ['tokens','words']:
		for kind in ['input', 'output']: 
			## save index file
			name = model + "_" + kind + "_"+ type + ".index" 
			write_index(metadata_all[kind][type][1], name)
			names.append(name)
	print("in save index done")
	return gr.File(names)


with gr.Blocks(js="plotsjs.js") as demo:
	gr.Markdown(
	"""
	# MAKE NMT Workshop \t `Embeddings representation` 
	""")
	with gr.Row():
		with gr.Column(scale=1):
			model_radio_c = gr.Radio(choices=['en-es', 'en-zh', 'en-fr'], value="en-es", label= '', container=False)

		with gr.Column(scale=2):
			gr.Markdown(
				"""
				### Reference Translation Sentences 
				Enter at least 50 sentences to be used as comparison.
				This is submitted just once. 
				""")
			in_text = gr.Textbox(lines=2, label="reference source text")
			out_text  = gr.Textbox(label="reference target text", interactive=False)
			out_text2  = gr.Textbox(visible=False)
			var2 = gr.JSON(visible=False)
			btn = gr.Button("Reference Translation")
			# save_index_btn = gr.Button("Download reference index")
			# file_obj = gr.File(label="Input File")
			# input = file_obj
			save_index_btn = gr.Button("Generate index files to download ",)
			tab2_outputs = gr.File()
			input = tab2_outputs

			# save_output = gr.Button("Download", link="/file=en-es_input_tokens.index")
			
	
		with gr.Column(scale=3):

			gr.Markdown(
				"""
				### Translation Sentences 
				Sentences to be analysed. 
				""")
			in_text_tr = gr.Textbox(lines=2, label="source text")
			out_text_tr  = gr.Textbox(label="target text", interactive=False)
			out_text2_tr  = gr.Textbox(visible=False)
			var2_tr = gr.JSON(visible=False)
			btn_faiss= gr.Button("Translation ")
			gr.Button("Download", link="/file=en-es_input_tokens.index")

	with gr.Row():
	# input_mic = gr.HTML(html)
		with gr.Column(scale=1):
			input_mic = gr.HTML(html0)
			input_html2 = gr.HTML(html_col2)

		with gr.Column(scale=2):
			input_html1 = gr.HTML(html_col1)
		# with gr.Column(scale=2):
			
		with gr.Column(scale=2):
			input_html3 = gr.HTML(html_col3)

	## first function input w1, model ; return out_text, var2; it does first function and js;
	btn.click(first_function, [in_text, model_radio_c], [out_text,var2], js="(in_text,model_radio_c) => testFn_out(in_text,model_radio_c)") #should return an output comp.
	btn_faiss.click(first_function_tr, [in_text_tr, model_radio_c], [out_text_tr,var2_tr], js="(in_text_tr,model_radio_c) => testFn_out(in_text_tr,model_radio_c)") #should return an output comp.
	## second function input out_text(returned in first_function), [json]var2(returned in first_function) ;
	## second function returns out_text2, var2; it does second function and js(with the input params);
	out_text.change(second_function, [out_text, var2], out_text2, js="(out_text,var2) => testFn_out_json(var2)") #
	out_text_tr.change(second_function, [out_text_tr, var2_tr], out_text2_tr, js="(out_text_tr,var2_tr) => testFn_out_json_tr(var2_tr)") #
	save_index_btn.click(save_index, [model_radio_c], [tab2_outputs])

	# tab2_submit_button.click(func2,
	#                     inputs=tab2_inputs,
	#                     outputs=tab2_outputs)
	
	# run script function on load,
	# demo.load(None,None,None,js="plotsjs.js")
# allowed_paths
if __name__ == "__main__":   
    demo.launch(allowed_paths=["./", ".", "/"])