ashishraics commited on
Commit
8bb7965
1 Parent(s): 356e503

optimize app

Browse files
.gitignore CHANGED
@@ -1,6 +1,6 @@
1
  venv/
2
- sentiment_model_dir/
3
- sent_mdl_dir/
4
- zs_model_dir/
5
  #sent_clf_onnx_dir/
6
  #zs_onnx_dir/
 
1
  venv/
2
+ #exclude model files as they are large
3
+ sentiment_model_dir/pytorch_model.bin
4
+ zs_model_dir/pytorch_model.bin
5
  #sent_clf_onnx_dir/
6
  #zs_onnx_dir/
app.py CHANGED
@@ -103,10 +103,12 @@ def create_model_dir(chkpt, model_dir):
103
  pass
104
 
105
 
106
- st.title("NLP use cases")
107
-
 
108
  with st.sidebar:
109
- st.title("NLP tasks")
 
110
  select_task=st.selectbox(label="Select task from drop down menu",
111
  options=['README',
112
  'Detect Sentiment','Zero Shot Classification'])
@@ -114,7 +116,7 @@ with st.sidebar:
114
  ############### Pre-Download & instantiate objects for sentiment analysis *********************** START **********************
115
 
116
  # #create model/token dir for sentiment classification for faster inference
117
- # create_model_dir(chkpt=sent_chkpt, model_dir=sent_mdl_dir)
118
 
119
 
120
  @st.cache(allow_output_mutation=True, suppress_st_warning=True, max_entries=None, ttl=None)
@@ -125,26 +127,26 @@ def sentiment_task_selected(task,
125
  sent_onnx_mdl_name=sent_onnx_mdl_name,
126
  sent_onnx_quant_mdl_name=sent_onnx_quant_mdl_name):
127
  #model & tokenizer initialization for normal sentiment classification
128
- model_sentiment=AutoModelForSequenceClassification.from_pretrained(sent_chkpt)
129
- tokenizer_sentiment=AutoTokenizer.from_pretrained(sent_chkpt)
 
130
 
131
- # create onnx model for sentiment classification
132
- create_onnx_model_sentiment(_model=model_sentiment, _tokenizer=tokenizer_sentiment)
133
 
134
  #create inference session
135
  sentiment_session = ort.InferenceSession(f"{sent_onnx_mdl_dir}/{sent_onnx_mdl_name}")
136
  # sentiment_session_quant = ort.InferenceSession(f"{sent_onnx_mdl_dir}/{sent_onnx_quant_mdl_name}")
137
 
138
- return model_sentiment,tokenizer_sentiment,sentiment_session
139
 
140
  ############## Pre-Download & instantiate objects for sentiment analysis ********************* END **********************************
141
 
142
 
143
  ############### Pre-Download & instantiate objects for Zero shot clf *********************** START **********************
144
 
145
- # # create model/token dir for zeroshot clf
146
- # create_model_dir(chkpt=zs_chkpt, model_dir=zs_mdl_dir)
147
-
148
 
149
  @st.cache(allow_output_mutation=True, suppress_st_warning=True, max_entries=None, ttl=None)
150
  def zs_task_selected(task,
@@ -157,10 +159,11 @@ def zs_task_selected(task,
157
  ##model & tokenizer initialization for normal ZS classification
158
  # model_zs=AutoModelForSequenceClassification.from_pretrained(zs_chkpt)
159
  # we just need tokenizer for inference and not model since onnx model is already saved
160
- tokenizer_zs=AutoTokenizer.from_pretrained(zs_chkpt)
 
161
 
162
- # create onnx model for zeroshot
163
- create_onnx_model_zs()
164
 
165
  #create inference session from onnx model
166
  zs_session = ort.InferenceSession(f"{zs_onnx_mdl_dir}/{zs_onnx_mdl_name}")
@@ -172,11 +175,11 @@ def zs_task_selected(task,
172
 
173
  if select_task=='README':
174
  st.header("NLP Summary")
 
175
 
176
  if select_task == 'Detect Sentiment':
177
  t1=time.time()
178
- model_sentiment,tokenizer_sentiment,\
179
- sentiment_session = sentiment_task_selected(task=select_task)
180
  t2 = time.time()
181
  st.write(f"Total time to load Model is {(t2-t1)*1000:.1f} ms")
182
 
@@ -185,28 +188,16 @@ if select_task == 'Detect Sentiment':
185
  c1,c2,_,_=st.columns(4)
186
 
187
  with c1:
188
- response1=st.button("Normal runtime")
189
- with c2:
190
- response2=st.button("ONNX runtime")
191
-
192
- if any([response1,response2]):
193
- if response1:
194
- start=time.time()
195
- sentiments = classify_sentiment(input_texts,
196
- model=model_sentiment,
197
- tokenizer=tokenizer_sentiment
198
- )
199
- end=time.time()
200
- st.write(f"Time taken for computation {(end-start)*1000:.1f} ms")
201
- elif response2:
202
- start = time.time()
203
- sentiments=classify_sentiment_onnx(input_texts,
204
- _session=sentiment_session,
205
- _tokenizer=tokenizer_sentiment)
206
- end = time.time()
207
- st.write(f"Time taken for computation {(end - start) * 1000:.1f} ms")
208
- else:
209
- pass
210
  for i,t in enumerate(input_texts.split(',')):
211
  if sentiments[i]=='Positive':
212
  response=st_text_rater(t + f"--> This statement is {sentiments[i]}",
@@ -214,6 +205,8 @@ if select_task == 'Detect Sentiment':
214
  else:
215
  response = st_text_rater(t + f"--> This statement is {sentiments[i]}",
216
  color_background='rgb(233, 116, 81)',key=t)
 
 
217
 
218
  if select_task=='Zero Shot Classification':
219
  t1=time.time()
@@ -228,7 +221,7 @@ if select_task=='Zero Shot Classification':
228
  c1,_,_,_=st.columns(4)
229
 
230
  with c1:
231
- response1=st.button("Compute with ONNX runtime")
232
 
233
  if response1:
234
  start = time.time()
 
103
  pass
104
 
105
 
106
+ #title using markdown
107
+ st.markdown("<h1 style='text-align: center; color: #3366ff;'>NLP Basic Use Cases</h1>", unsafe_allow_html=True)
108
+ st.markdown("---")
109
  with st.sidebar:
110
+ # title using markdown
111
+ st.markdown("<h1 style='text-align: left; color: ;'>NLP Tasks</h1>", unsafe_allow_html=True)
112
  select_task=st.selectbox(label="Select task from drop down menu",
113
  options=['README',
114
  'Detect Sentiment','Zero Shot Classification'])
 
116
  ############### Pre-Download & instantiate objects for sentiment analysis *********************** START **********************
117
 
118
  # #create model/token dir for sentiment classification for faster inference
119
+ create_model_dir(chkpt=sent_chkpt, model_dir=sent_mdl_dir)
120
 
121
 
122
  @st.cache(allow_output_mutation=True, suppress_st_warning=True, max_entries=None, ttl=None)
 
127
  sent_onnx_mdl_name=sent_onnx_mdl_name,
128
  sent_onnx_quant_mdl_name=sent_onnx_quant_mdl_name):
129
  #model & tokenizer initialization for normal sentiment classification
130
+ # model_sentiment=AutoModelForSequenceClassification.from_pretrained(sent_chkpt)
131
+ # tokenizer_sentiment=AutoTokenizer.from_pretrained(sent_chkpt)
132
+ tokenizer_sentiment = AutoTokenizer.from_pretrained(sent_mdl_dir)
133
 
134
+ # # create onnx model for sentiment classification but once created in your local app comment this out
135
+ # create_onnx_model_sentiment(_model=model_sentiment, _tokenizer=tokenizer_sentiment)
136
 
137
  #create inference session
138
  sentiment_session = ort.InferenceSession(f"{sent_onnx_mdl_dir}/{sent_onnx_mdl_name}")
139
  # sentiment_session_quant = ort.InferenceSession(f"{sent_onnx_mdl_dir}/{sent_onnx_quant_mdl_name}")
140
 
141
+ return tokenizer_sentiment,sentiment_session
142
 
143
  ############## Pre-Download & instantiate objects for sentiment analysis ********************* END **********************************
144
 
145
 
146
  ############### Pre-Download & instantiate objects for Zero shot clf *********************** START **********************
147
 
148
+ # create model/token dir for zeroshot clf -- already created so not required
149
+ create_model_dir(chkpt=zs_chkpt, model_dir=zs_mdl_dir)
 
150
 
151
  @st.cache(allow_output_mutation=True, suppress_st_warning=True, max_entries=None, ttl=None)
152
  def zs_task_selected(task,
 
159
  ##model & tokenizer initialization for normal ZS classification
160
  # model_zs=AutoModelForSequenceClassification.from_pretrained(zs_chkpt)
161
  # we just need tokenizer for inference and not model since onnx model is already saved
162
+ # tokenizer_zs=AutoTokenizer.from_pretrained(zs_chkpt)
163
+ tokenizer_zs = AutoTokenizer.from_pretrained(zs_mdl_dir)
164
 
165
+ # # create onnx model for zeroshot but once created locally comment it out.
166
+ # create_onnx_model_zs()
167
 
168
  #create inference session from onnx model
169
  zs_session = ort.InferenceSession(f"{zs_onnx_mdl_dir}/{zs_onnx_mdl_name}")
 
175
 
176
  if select_task=='README':
177
  st.header("NLP Summary")
178
+ # st.write()
179
 
180
  if select_task == 'Detect Sentiment':
181
  t1=time.time()
182
+ tokenizer_sentiment,sentiment_session = sentiment_task_selected(task=select_task)
 
183
  t2 = time.time()
184
  st.write(f"Total time to load Model is {(t2-t1)*1000:.1f} ms")
185
 
 
188
  c1,c2,_,_=st.columns(4)
189
 
190
  with c1:
191
+ response1=st.button("Compute (ONNX runtime)")
192
+
193
+ if response1:
194
+ start = time.time()
195
+ sentiments=classify_sentiment_onnx(input_texts,
196
+ _session=sentiment_session,
197
+ _tokenizer=tokenizer_sentiment)
198
+ end = time.time()
199
+ st.write(f"Time taken for computation {(end - start) * 1000:.1f} ms")
200
+
 
 
 
 
 
 
 
 
 
 
 
 
201
  for i,t in enumerate(input_texts.split(',')):
202
  if sentiments[i]=='Positive':
203
  response=st_text_rater(t + f"--> This statement is {sentiments[i]}",
 
205
  else:
206
  response = st_text_rater(t + f"--> This statement is {sentiments[i]}",
207
  color_background='rgb(233, 116, 81)',key=t)
208
+ else:
209
+ pass
210
 
211
  if select_task=='Zero Shot Classification':
212
  t1=time.time()
 
221
  c1,_,_,_=st.columns(4)
222
 
223
  with c1:
224
+ response1=st.button("Compute (ONNX runtime)")
225
 
226
  if response1:
227
  start = time.time()
sentiment_model_dir/config.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "distilbert-base-uncased-finetuned-sst-2-english",
3
+ "activation": "gelu",
4
+ "architectures": [
5
+ "DistilBertForSequenceClassification"
6
+ ],
7
+ "attention_dropout": 0.1,
8
+ "dim": 768,
9
+ "dropout": 0.1,
10
+ "finetuning_task": "sst-2",
11
+ "hidden_dim": 3072,
12
+ "id2label": {
13
+ "0": "NEGATIVE",
14
+ "1": "POSITIVE"
15
+ },
16
+ "initializer_range": 0.02,
17
+ "label2id": {
18
+ "NEGATIVE": 0,
19
+ "POSITIVE": 1
20
+ },
21
+ "max_position_embeddings": 512,
22
+ "model_type": "distilbert",
23
+ "n_heads": 12,
24
+ "n_layers": 6,
25
+ "output_past": true,
26
+ "pad_token_id": 0,
27
+ "qa_dropout": 0.1,
28
+ "seq_classif_dropout": 0.2,
29
+ "sinusoidal_pos_embds": false,
30
+ "tie_weights_": true,
31
+ "torch_dtype": "float32",
32
+ "transformers_version": "4.18.0",
33
+ "vocab_size": 30522
34
+ }
sentiment_model_dir/special_tokens_map.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"unk_token": "[UNK]", "sep_token": "[SEP]", "pad_token": "[PAD]", "cls_token": "[CLS]", "mask_token": "[MASK]"}
sentiment_model_dir/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
sentiment_model_dir/tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"do_lower_case": true, "unk_token": "[UNK]", "sep_token": "[SEP]", "pad_token": "[PAD]", "cls_token": "[CLS]", "mask_token": "[MASK]", "tokenize_chinese_chars": true, "strip_accents": null, "model_max_length": 512, "special_tokens_map_file": null, "name_or_path": "distilbert-base-uncased-finetuned-sst-2-english", "do_basic_tokenize": true, "never_split": null, "tokenizer_class": "DistilBertTokenizer"}
sentiment_model_dir/vocab.txt ADDED
The diff for this file is too large to render. See raw diff
 
zs_model_dir/config.json ADDED
@@ -0,0 +1,58 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "valhalla/distilbart-mnli-12-1",
3
+ "_num_labels": 3,
4
+ "activation_dropout": 0.0,
5
+ "activation_function": "gelu",
6
+ "add_bias_logits": false,
7
+ "add_final_layer_norm": false,
8
+ "architectures": [
9
+ "BartForSequenceClassification"
10
+ ],
11
+ "attention_dropout": 0.1,
12
+ "bos_token_id": 0,
13
+ "classif_dropout": 0.0,
14
+ "classifier_dropout": 0.0,
15
+ "d_model": 1024,
16
+ "decoder_attention_heads": 16,
17
+ "decoder_ffn_dim": 4096,
18
+ "decoder_layerdrop": 0.0,
19
+ "decoder_layers": 1,
20
+ "decoder_start_token_id": 2,
21
+ "dropout": 0.1,
22
+ "encoder_attention_heads": 16,
23
+ "encoder_ffn_dim": 4096,
24
+ "encoder_layerdrop": 0.0,
25
+ "encoder_layers": 12,
26
+ "eos_token_id": 2,
27
+ "extra_pos_embeddings": 2,
28
+ "finetuning_task": "mnli",
29
+ "force_bos_token_to_be_generated": false,
30
+ "forced_eos_token_id": 2,
31
+ "gradient_checkpointing": false,
32
+ "id2label": {
33
+ "0": "contradiction",
34
+ "1": "neutral",
35
+ "2": "entailment"
36
+ },
37
+ "init_std": 0.02,
38
+ "is_encoder_decoder": true,
39
+ "label2id": {
40
+ "contradiction": 0,
41
+ "entailment": 2,
42
+ "neutral": 1
43
+ },
44
+ "max_position_embeddings": 1024,
45
+ "model_type": "bart",
46
+ "normalize_before": false,
47
+ "normalize_embedding": true,
48
+ "num_hidden_layers": 12,
49
+ "output_past": false,
50
+ "pad_token_id": 1,
51
+ "scale_embedding": false,
52
+ "static_position_embeddings": false,
53
+ "torch_dtype": "float32",
54
+ "total_flos": 153130534133111808,
55
+ "transformers_version": "4.18.0",
56
+ "use_cache": true,
57
+ "vocab_size": 50265
58
+ }
zs_model_dir/merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
zs_model_dir/special_tokens_map.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"bos_token": {"content": "<s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "eos_token": {"content": "</s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "unk_token": {"content": "<unk>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "sep_token": {"content": "</s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "pad_token": {"content": "<pad>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "cls_token": {"content": "<s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "mask_token": {"content": "<mask>", "single_word": false, "lstrip": true, "rstrip": false, "normalized": true}}
zs_model_dir/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
zs_model_dir/tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"errors": "replace", "bos_token": {"content": "<s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "eos_token": {"content": "</s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "sep_token": {"content": "</s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "cls_token": {"content": "<s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "unk_token": {"content": "<unk>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "pad_token": {"content": "<pad>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "mask_token": {"content": "<mask>", "single_word": false, "lstrip": true, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "add_prefix_space": false, "trim_offsets": true, "model_max_length": 1024, "special_tokens_map_file": "/Users/ashishrai/.cache/huggingface/transformers/1897a33c6ca1e896797e7f370753103e4fb6980c6371197c5658ff3a8269dc4a.cb2244924ab24d706b02fd7fcedaea4531566537687a539ebb94db511fd122a0", "name_or_path": "valhalla/distilbart-mnli-12-1", "tokenizer_class": "BartTokenizer"}
zs_model_dir/vocab.json ADDED
The diff for this file is too large to render. See raw diff