cifkao commited on
Commit
dc1d5c3
·
1 Parent(s): 4b314cc

Implement generation mode

Browse files
app.py CHANGED
@@ -1,26 +1,33 @@
1
  from enum import Enum
2
  from pathlib import Path
 
3
 
4
  import streamlit as st
5
  import streamlit.components.v1 as components
6
  import numpy as np
7
  import torch
8
  import torch.nn.functional as F
9
- from transformers import AutoModelForCausalLM, AutoTokenizer, BatchEncoding
10
 
11
  root_dir = Path(__file__).resolve().parent
12
  highlighted_text_component = components.declare_component(
13
  "highlighted_text", path=root_dir / "highlighted_text" / "build"
14
  )
15
 
16
- def get_windows_batched(examples: BatchEncoding, window_len: int, stride: int = 1, pad_id: int = 0) -> BatchEncoding:
 
 
 
 
 
 
17
  return BatchEncoding({
18
  k: [
19
  t[i][j : j + window_len] + [
20
  pad_id if k in ["input_ids", "labels"] else 0
21
  ] * (j + window_len - len(t[i]))
22
  for i in range(len(examples["input_ids"]))
23
- for j in range(0, len(examples["input_ids"][i]) - 1, stride)
24
  ]
25
  for k, t in examples.items()
26
  })
@@ -78,13 +85,12 @@ if not compact_layout:
78
  """
79
  )
80
 
81
- generation_mode = False
82
- # st.radio("Mode", ["Standard", "Generation"], horizontal=True) == "Generation"
83
- # st.caption(
84
- # "In standard mode, we analyze the model's predictions on the input text. "
85
- # "In generation mode, we generate a continuation of the input text "
86
- # "and visualize the contributions of different contexts to each generated token."
87
- # )
88
 
89
  model_name = st.selectbox("Model", ["distilgpt2", "gpt2", "EleutherAI/gpt-neo-125m"])
90
  metric_name = st.radio(
@@ -113,6 +119,13 @@ window_len = st.select_slider(
113
  max_tokens = int(MAX_MEM / (multiplier * window_len) - window_len)
114
  max_tokens = min(max_tokens, 4096)
115
 
 
 
 
 
 
 
 
116
  DEFAULT_TEXT = """
117
  We present context length probing, a novel explanation technique for causal
118
  language models, based on tracking the predictions of a model as a function of the length of
@@ -124,22 +137,21 @@ dependencies.
124
  """.replace("\n", " ").strip()
125
 
126
  text = st.text_area(
127
- f"Input text (≤\u2009{max_tokens} tokens)",
128
  st.session_state.get("input_text", DEFAULT_TEXT),
129
  key="input_text",
130
  )
131
 
132
- if tokenizer.eos_token:
133
- text += tokenizer.eos_token
134
  inputs = tokenizer([text])
135
  [input_ids] = inputs["input_ids"]
136
- inputs["labels"] = [[*input_ids[1:], tokenizer.eos_token_id]]
137
- num_user_tokens = len(input_ids) - (1 if tokenizer.eos_token else 0)
 
138
 
139
  if num_user_tokens < 1:
140
  st.error("Please enter at least one token.", icon="🚨")
141
  st.stop()
142
- if num_user_tokens > max_tokens:
143
  st.error(
144
  f"Your input has {num_user_tokens} tokens. Please enter at most {max_tokens} tokens "
145
  f"or try reducing the window size.",
@@ -150,53 +162,109 @@ if num_user_tokens > max_tokens:
150
  with st.spinner("Loading model…"):
151
  model = st.cache_resource(AutoModelForCausalLM.from_pretrained, show_spinner=False)(model_name)
152
 
153
- window_len = min(window_len, len(input_ids))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
154
 
155
  @torch.inference_mode()
156
- def get_logprobs(_model, _inputs):
157
- return _model(**_inputs).logits.log_softmax(dim=-1).to(torch.float16)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
158
 
159
- @st.cache_data(show_spinner=False)
160
  @torch.inference_mode()
161
- def run_context_length_probing(_model, _tokenizer, _inputs, window_len, metric, cache_key):
 
 
 
 
 
 
 
 
 
162
  del cache_key
163
 
164
- inputs_sliding = get_windows_batched(
165
- _inputs,
166
- window_len=window_len,
167
- pad_id=_tokenizer.eos_token_id
168
- ).convert_to_tensors("pt")
169
 
170
- logprobs = []
171
  with st.spinner("Running model…"):
172
- batch_size = 8
173
- num_items = len(inputs_sliding["input_ids"])
174
- pbar = st.progress(0)
175
- for i in range(0, num_items, batch_size):
176
- pbar.progress(i / num_items, f"{i}/{num_items}")
177
- batch = {k: v[i:i + batch_size] for k, v in inputs_sliding.items()}
178
- batch_logprobs = get_logprobs(
179
- _model,
180
- batch,
181
- #cache_key=(model_name, batch["input_ids"].cpu().numpy().tobytes())
182
  )
183
- batch_labels = batch["labels"]
184
- if metric != "KL divergence":
185
- batch_logprobs = torch.gather(
186
- batch_logprobs, dim=-1, index=batch_labels[..., None]
187
- )
188
- logprobs.append(batch_logprobs)
189
- logprobs = torch.cat(logprobs, dim=0)
190
- pbar.empty()
 
 
 
 
 
191
 
192
  with st.spinner("Computing scores…"):
193
  logprobs = logprobs.permute(1, 0, 2)
194
  logprobs = F.pad(logprobs, (0, 0, 0, window_len, 0, 0), value=torch.nan)
195
  logprobs = logprobs.view(-1, logprobs.shape[-1])[:-window_len]
196
- logprobs = logprobs.view(window_len, len(input_ids) + window_len - 2, logprobs.shape[-1])
197
 
198
  if metric == "NLL loss":
199
- scores = nll_score(logprobs=logprobs, labels=input_ids[1:])
200
  elif metric == "KL divergence":
201
  scores = kl_div_score(logprobs)
202
  del logprobs # possibly destroyed by the score computation to save memory
@@ -206,17 +274,29 @@ def run_context_length_probing(_model, _tokenizer, _inputs, window_len, metric,
206
  scores /= scores.abs().max(dim=1, keepdim=True).values + 1e-6
207
  scores = scores.to(torch.float16)
208
 
209
- return scores
 
 
 
210
 
211
- scores = run_context_length_probing(
 
 
 
212
  _model=model,
213
  _tokenizer=tokenizer,
214
  _inputs=inputs,
215
  window_len=window_len,
216
  metric=metric_name,
 
 
217
  cache_key=(model_name, text),
218
  )
219
- tokens = ids_to_readable_tokens(tokenizer, input_ids)
220
 
221
  st.markdown('<label style="font-size: 14px;">Output</label>', unsafe_allow_html=True)
222
- highlighted_text_component(tokens=tokens, scores=scores.tolist())
 
 
 
 
 
1
  from enum import Enum
2
  from pathlib import Path
3
+ from typing import Hashable
4
 
5
  import streamlit as st
6
  import streamlit.components.v1 as components
7
  import numpy as np
8
  import torch
9
  import torch.nn.functional as F
10
+ from transformers import AutoModelForCausalLM, AutoTokenizer, BatchEncoding, GPT2LMHeadModel, PreTrainedTokenizer
11
 
12
  root_dir = Path(__file__).resolve().parent
13
  highlighted_text_component = components.declare_component(
14
  "highlighted_text", path=root_dir / "highlighted_text" / "build"
15
  )
16
 
17
+ def get_windows_batched(
18
+ examples: BatchEncoding,
19
+ window_len: int,
20
+ start: int = 0,
21
+ stride: int = 1,
22
+ pad_id: int = 0
23
+ ) -> BatchEncoding:
24
  return BatchEncoding({
25
  k: [
26
  t[i][j : j + window_len] + [
27
  pad_id if k in ["input_ids", "labels"] else 0
28
  ] * (j + window_len - len(t[i]))
29
  for i in range(len(examples["input_ids"]))
30
+ for j in range(start, len(examples["input_ids"][i]), stride)
31
  ]
32
  for k, t in examples.items()
33
  })
 
85
  """
86
  )
87
 
88
+ generation_mode = st.radio("Mode", ["Standard", "Generation"], horizontal=True) == "Generation"
89
+ st.caption(
90
+ "In standard mode, we analyze the model's predictions on the input text. "
91
+ "In generation mode, we generate a continuation of the input text (prompt) "
92
+ "and visualize the contributions of different contexts to each generated token."
93
+ )
 
94
 
95
  model_name = st.selectbox("Model", ["distilgpt2", "gpt2", "EleutherAI/gpt-neo-125m"])
96
  metric_name = st.radio(
 
119
  max_tokens = int(MAX_MEM / (multiplier * window_len) - window_len)
120
  max_tokens = min(max_tokens, 4096)
121
 
122
+ max_new_tokens = None
123
+ if generation_mode:
124
+ max_new_tokens = st.slider(
125
+ "Max. number of generated tokens",
126
+ min_value=8, max_value=min(1024, max_tokens), value=min(128, max_tokens)
127
+ )
128
+
129
  DEFAULT_TEXT = """
130
  We present context length probing, a novel explanation technique for causal
131
  language models, based on tracking the predictions of a model as a function of the length of
 
137
  """.replace("\n", " ").strip()
138
 
139
  text = st.text_area(
140
+ f"Prompt" if generation_mode else "Input text (≤\u2009{max_tokens} tokens)",
141
  st.session_state.get("input_text", DEFAULT_TEXT),
142
  key="input_text",
143
  )
144
 
 
 
145
  inputs = tokenizer([text])
146
  [input_ids] = inputs["input_ids"]
147
+ label_ids = [*input_ids[1:], tokenizer.eos_token_id]
148
+ inputs["labels"] = [label_ids]
149
+ num_user_tokens = len(input_ids)
150
 
151
  if num_user_tokens < 1:
152
  st.error("Please enter at least one token.", icon="🚨")
153
  st.stop()
154
+ if not generation_mode and num_user_tokens > max_tokens:
155
  st.error(
156
  f"Your input has {num_user_tokens} tokens. Please enter at most {max_tokens} tokens "
157
  f"or try reducing the window size.",
 
162
  with st.spinner("Loading model…"):
163
  model = st.cache_resource(AutoModelForCausalLM.from_pretrained, show_spinner=False)(model_name)
164
 
165
+ @torch.inference_mode()
166
+ def get_logprobs(model, inputs, metric):
167
+ logprobs = []
168
+ batch_size = 8
169
+ num_items = len(inputs["input_ids"])
170
+ pbar = st.progress(0)
171
+ for i in range(0, num_items, batch_size):
172
+ pbar.progress(i / num_items, f"{i}/{num_items}")
173
+ batch = {k: v[i:i + batch_size] for k, v in inputs.items()}
174
+ batch_logprobs = model(**batch).logits.log_softmax(dim=-1).to(torch.float16)
175
+ if metric != "KL divergence":
176
+ batch_logprobs = torch.gather(
177
+ batch_logprobs, dim=-1, index=batch["labels"][..., None]
178
+ )
179
+ logprobs.append(batch_logprobs)
180
+ logprobs = torch.cat(logprobs, dim=0)
181
+ pbar.empty()
182
+ return logprobs
183
 
184
  @torch.inference_mode()
185
+ def generate(model, inputs, metric, window_len, max_new_tokens):
186
+ assert metric == "NLL loss"
187
+ start = max(0, inputs["input_ids"].shape[1] - window_len + 1)
188
+ inputs_window = {k: v[:, start:] for k, v in inputs.items()}
189
+ del inputs_window["labels"]
190
+
191
+ new_ids, logprobs = [], []
192
+ eos_idx = None
193
+ pbar = st.progress(0)
194
+ max_steps = max_new_tokens + window_len - 1
195
+ for i in range(max_steps):
196
+ pbar.progress(i / max_steps, f"{i}/{max_steps}")
197
+ inputs_window["attention_mask"] = torch.ones_like(inputs_window["input_ids"], dtype=torch.long)
198
+ logprobs_window = model(**inputs_window).logits.log_softmax(dim=-1).squeeze(0)
199
+ if eos_idx is None:
200
+ next_token = torch.multinomial(logprobs_window[-1].exp(), num_samples=1).item()
201
+ if next_token == tokenizer.eos_token_id or i >= max_new_tokens - 1:
202
+ eos_idx = i
203
+ else:
204
+ next_token = tokenizer.eos_token_id
205
+ new_ids.append(next_token)
206
+
207
+ inputs_window["input_ids"] = torch.cat([inputs_window["input_ids"], torch.tensor([[next_token]])], dim=1)
208
+ if inputs_window["input_ids"].shape[1] > window_len:
209
+ inputs_window["input_ids"] = inputs_window["input_ids"][:, 1:]
210
+ if logprobs_window.shape[0] == window_len:
211
+ logprobs.append(
212
+ logprobs_window[torch.arange(window_len), inputs_window["input_ids"].squeeze(0)]
213
+ )
214
+
215
+ if eos_idx is not None and i - eos_idx >= window_len - 1:
216
+ break
217
+ pbar.empty()
218
+
219
+ return torch.as_tensor(new_ids[:eos_idx + 1]), torch.stack(logprobs)[:, :, None]
220
 
 
221
  @torch.inference_mode()
222
+ def run_context_length_probing(
223
+ _model: GPT2LMHeadModel,
224
+ _tokenizer: PreTrainedTokenizer,
225
+ _inputs: dict[str, torch.Tensor],
226
+ window_len: int,
227
+ metric: str,
228
+ generation_mode: bool,
229
+ max_new_tokens: int,
230
+ cache_key: Hashable
231
+ ):
232
  del cache_key
233
 
234
+ [input_ids] = _inputs["input_ids"]
235
+ [label_ids] = _inputs["labels"]
 
 
 
236
 
 
237
  with st.spinner("Running model…"):
238
+ if generation_mode:
239
+ new_ids, logprobs = generate(
240
+ model=_model,
241
+ inputs=_inputs.convert_to_tensors("pt"),
242
+ metric=metric,
243
+ window_len=window_len,
244
+ max_new_tokens=max_new_tokens
 
 
 
245
  )
246
+ output_ids = [*input_ids, *new_ids]
247
+ window_len = logprobs.shape[1]
248
+ else:
249
+ window_len = min(window_len, len(input_ids))
250
+ inputs_sliding = get_windows_batched(
251
+ _inputs,
252
+ window_len=window_len,
253
+ start=0,
254
+ pad_id=_tokenizer.eos_token_id
255
+ ).convert_to_tensors("pt")
256
+ logprobs = get_logprobs(model=model, inputs=inputs_sliding, metric=metric)
257
+ output_ids = [*input_ids, label_ids[-1]]
258
+ num_tgt_tokens = logprobs.shape[0]
259
 
260
  with st.spinner("Computing scores…"):
261
  logprobs = logprobs.permute(1, 0, 2)
262
  logprobs = F.pad(logprobs, (0, 0, 0, window_len, 0, 0), value=torch.nan)
263
  logprobs = logprobs.view(-1, logprobs.shape[-1])[:-window_len]
264
+ logprobs = logprobs.view(window_len, num_tgt_tokens + window_len - 1, logprobs.shape[-1])
265
 
266
  if metric == "NLL loss":
267
+ scores = nll_score(logprobs=logprobs, labels=label_ids)
268
  elif metric == "KL divergence":
269
  scores = kl_div_score(logprobs)
270
  del logprobs # possibly destroyed by the score computation to save memory
 
274
  scores /= scores.abs().max(dim=1, keepdim=True).values + 1e-6
275
  scores = scores.to(torch.float16)
276
 
277
+ if generation_mode:
278
+ scores = F.pad(scores, (0, 0, max(0, len(input_ids) - window_len + 1), 0), value=0.)
279
+
280
+ return output_ids, scores
281
 
282
+ if not generation_mode:
283
+ run_context_length_probing = st.cache_data(run_context_length_probing, show_spinner=False)
284
+
285
+ output_ids, scores = run_context_length_probing(
286
  _model=model,
287
  _tokenizer=tokenizer,
288
  _inputs=inputs,
289
  window_len=window_len,
290
  metric=metric_name,
291
+ generation_mode=generation_mode,
292
+ max_new_tokens=max_new_tokens,
293
  cache_key=(model_name, text),
294
  )
295
+ tokens = ids_to_readable_tokens(tokenizer, output_ids)
296
 
297
  st.markdown('<label style="font-size: 14px;">Output</label>', unsafe_allow_html=True)
298
+ highlighted_text_component(
299
+ tokens=tokens,
300
+ scores=scores.tolist(),
301
+ prefix_len=len(input_ids) if generation_mode else 0
302
+ )
highlighted_text/build/asset-manifest.json CHANGED
@@ -1,20 +1,20 @@
1
  {
2
  "files": {
3
- "main.css": "./static/css/main.b2f89d68.chunk.css",
4
- "main.js": "./static/js/main.d475cbb2.chunk.js",
5
- "main.js.map": "./static/js/main.d475cbb2.chunk.js.map",
6
  "runtime-main.js": "./static/js/runtime-main.e6b0ae4c.js",
7
  "runtime-main.js.map": "./static/js/runtime-main.e6b0ae4c.js.map",
8
  "static/js/2.ce130e37.chunk.js": "./static/js/2.ce130e37.chunk.js",
9
  "static/js/2.ce130e37.chunk.js.map": "./static/js/2.ce130e37.chunk.js.map",
10
  "index.html": "./index.html",
11
- "static/css/main.b2f89d68.chunk.css.map": "./static/css/main.b2f89d68.chunk.css.map",
12
  "static/js/2.ce130e37.chunk.js.LICENSE.txt": "./static/js/2.ce130e37.chunk.js.LICENSE.txt"
13
  },
14
  "entrypoints": [
15
  "static/js/runtime-main.e6b0ae4c.js",
16
  "static/js/2.ce130e37.chunk.js",
17
- "static/css/main.b2f89d68.chunk.css",
18
- "static/js/main.d475cbb2.chunk.js"
19
  ]
20
  }
 
1
  {
2
  "files": {
3
+ "main.css": "./static/css/main.a84f16ea.chunk.css",
4
+ "main.js": "./static/js/main.5f2b5265.chunk.js",
5
+ "main.js.map": "./static/js/main.5f2b5265.chunk.js.map",
6
  "runtime-main.js": "./static/js/runtime-main.e6b0ae4c.js",
7
  "runtime-main.js.map": "./static/js/runtime-main.e6b0ae4c.js.map",
8
  "static/js/2.ce130e37.chunk.js": "./static/js/2.ce130e37.chunk.js",
9
  "static/js/2.ce130e37.chunk.js.map": "./static/js/2.ce130e37.chunk.js.map",
10
  "index.html": "./index.html",
11
+ "static/css/main.a84f16ea.chunk.css.map": "./static/css/main.a84f16ea.chunk.css.map",
12
  "static/js/2.ce130e37.chunk.js.LICENSE.txt": "./static/js/2.ce130e37.chunk.js.LICENSE.txt"
13
  },
14
  "entrypoints": [
15
  "static/js/runtime-main.e6b0ae4c.js",
16
  "static/js/2.ce130e37.chunk.js",
17
+ "static/css/main.a84f16ea.chunk.css",
18
+ "static/js/main.5f2b5265.chunk.js"
19
  ]
20
  }
highlighted_text/build/index.html CHANGED
@@ -1 +1 @@
1
- <!doctype html><html lang="en"><head><title>Highlighted text component</title><meta charset="UTF-8"/><meta name="viewport" content="width=device-width,initial-scale=1"/><meta name="description" content="Highlighted text"/><link href="./static/css/main.b2f89d68.chunk.css" rel="stylesheet"></head><body><div id="root"></div><script>!function(e){function t(t){for(var n,l,a=t[0],p=t[1],i=t[2],c=0,s=[];c<a.length;c++)l=a[c],Object.prototype.hasOwnProperty.call(o,l)&&o[l]&&s.push(o[l][0]),o[l]=0;for(n in p)Object.prototype.hasOwnProperty.call(p,n)&&(e[n]=p[n]);for(f&&f(t);s.length;)s.shift()();return u.push.apply(u,i||[]),r()}function r(){for(var e,t=0;t<u.length;t++){for(var r=u[t],n=!0,a=1;a<r.length;a++){var p=r[a];0!==o[p]&&(n=!1)}n&&(u.splice(t--,1),e=l(l.s=r[0]))}return e}var n={},o={1:0},u=[];function l(t){if(n[t])return n[t].exports;var r=n[t]={i:t,l:!1,exports:{}};return e[t].call(r.exports,r,r.exports,l),r.l=!0,r.exports}l.m=e,l.c=n,l.d=function(e,t,r){l.o(e,t)||Object.defineProperty(e,t,{enumerable:!0,get:r})},l.r=function(e){"undefined"!=typeof Symbol&&Symbol.toStringTag&&Object.defineProperty(e,Symbol.toStringTag,{value:"Module"}),Object.defineProperty(e,"__esModule",{value:!0})},l.t=function(e,t){if(1&t&&(e=l(e)),8&t)return e;if(4&t&&"object"==typeof e&&e&&e.__esModule)return e;var r=Object.create(null);if(l.r(r),Object.defineProperty(r,"default",{enumerable:!0,value:e}),2&t&&"string"!=typeof e)for(var n in e)l.d(r,n,function(t){return e[t]}.bind(null,n));return r},l.n=function(e){var t=e&&e.__esModule?function(){return e.default}:function(){return e};return l.d(t,"a",t),t},l.o=function(e,t){return Object.prototype.hasOwnProperty.call(e,t)},l.p="./";var a=this.webpackJsonpstreamlit_component_template=this.webpackJsonpstreamlit_component_template||[],p=a.push.bind(a);a.push=t,a=a.slice();for(var i=0;i<a.length;i++)t(a[i]);var f=p;r()}([])</script><script src="./static/js/2.ce130e37.chunk.js"></script><script src="./static/js/main.d475cbb2.chunk.js"></script></body></html>
 
1
+ <!doctype html><html lang="en"><head><title>Highlighted text component</title><meta charset="UTF-8"/><meta name="viewport" content="width=device-width,initial-scale=1"/><meta name="description" content="Highlighted text"/><link href="./static/css/main.a84f16ea.chunk.css" rel="stylesheet"></head><body><div id="root"></div><script>!function(e){function t(t){for(var n,l,a=t[0],p=t[1],i=t[2],c=0,s=[];c<a.length;c++)l=a[c],Object.prototype.hasOwnProperty.call(o,l)&&o[l]&&s.push(o[l][0]),o[l]=0;for(n in p)Object.prototype.hasOwnProperty.call(p,n)&&(e[n]=p[n]);for(f&&f(t);s.length;)s.shift()();return u.push.apply(u,i||[]),r()}function r(){for(var e,t=0;t<u.length;t++){for(var r=u[t],n=!0,a=1;a<r.length;a++){var p=r[a];0!==o[p]&&(n=!1)}n&&(u.splice(t--,1),e=l(l.s=r[0]))}return e}var n={},o={1:0},u=[];function l(t){if(n[t])return n[t].exports;var r=n[t]={i:t,l:!1,exports:{}};return e[t].call(r.exports,r,r.exports,l),r.l=!0,r.exports}l.m=e,l.c=n,l.d=function(e,t,r){l.o(e,t)||Object.defineProperty(e,t,{enumerable:!0,get:r})},l.r=function(e){"undefined"!=typeof Symbol&&Symbol.toStringTag&&Object.defineProperty(e,Symbol.toStringTag,{value:"Module"}),Object.defineProperty(e,"__esModule",{value:!0})},l.t=function(e,t){if(1&t&&(e=l(e)),8&t)return e;if(4&t&&"object"==typeof e&&e&&e.__esModule)return e;var r=Object.create(null);if(l.r(r),Object.defineProperty(r,"default",{enumerable:!0,value:e}),2&t&&"string"!=typeof e)for(var n in e)l.d(r,n,function(t){return e[t]}.bind(null,n));return r},l.n=function(e){var t=e&&e.__esModule?function(){return e.default}:function(){return e};return l.d(t,"a",t),t},l.o=function(e,t){return Object.prototype.hasOwnProperty.call(e,t)},l.p="./";var a=this.webpackJsonpstreamlit_component_template=this.webpackJsonpstreamlit_component_template||[],p=a.push.bind(a);a.push=t,a=a.slice();for(var i=0;i<a.length;i++)t(a[i]);var f=p;r()}([])</script><script src="./static/js/2.ce130e37.chunk.js"></script><script src="./static/js/main.5f2b5265.chunk.js"></script></body></html>
highlighted_text/build/static/css/{main.b2f89d68.chunk.css → main.a84f16ea.chunk.css} RENAMED
@@ -1,2 +1,2 @@
1
- body{padding:0;margin:0;font-family:"Source Sans Pro",sans-serif;font-size:1rem;line-height:1.4}.highlighted-text,.status-bar{color:#000;background-color:#fff}.highlighted-text{border:1px solid #d2d2d2;border-radius:0 0 5px 5px;padding:4px;cursor:pointer}.highlighted-text .token.active{outline:1px solid #444}.highlighted-text.frozen .token{opacity:.75}.highlighted-text.frozen .token.active,.highlighted-text.frozen .token.context{opacity:1}.highlighted-text.frozen .token.context{-webkit-text-decoration:#999 underline;text-decoration:#999 underline}.status-bar{min-height:1.4em;border:1px solid #d2d2d2;border-bottom:none;border-radius:5px 5px 0 0;padding:3px 4px}.status-bar .token{border:1px solid #aaa;border-radius:2px;background:#f5f5f5;padding:1px;margin:1px 2px 1px 1px}
2
- /*# sourceMappingURL=main.b2f89d68.chunk.css.map */
 
1
+ body{padding:0;margin:0;font-family:"Source Sans Pro",sans-serif;font-size:1rem;line-height:1.4}.highlighted-text,.status-bar{color:#000;background-color:#fff}.highlighted-text{border:1px solid #d2d2d2;border-radius:0 0 5px 5px;padding:4px;cursor:pointer}.highlighted-text .token.prefix~.token:not(.prefix){color:#2563eb}.highlighted-text .token.active{outline:1px solid #444}.status-bar{min-height:1.4em;border:1px solid #d2d2d2;border-bottom:none;border-radius:5px 5px 0 0;padding:3px 4px}.status-bar .token{border:1px solid #aaa;border-radius:2px;background:#f5f5f5;padding:1px;margin:1px 2px 1px 1px}
2
+ /*# sourceMappingURL=main.a84f16ea.chunk.css.map */
highlighted_text/build/static/css/main.a84f16ea.chunk.css.map ADDED
@@ -0,0 +1 @@
 
 
1
+ {"version":3,"sources":["webpack://src/index.scss"],"names":[],"mappings":"AAAA,KACI,SAAA,CACA,QAAA,CACA,wCAAA,CACA,cAAA,CACA,eAAA,CAGJ,8BACI,UAAA,CACA,qBAAA,CAGJ,kBACI,wBAAA,CACA,yBAAA,CACA,WAAA,CACA,cAAA,CAEA,oDACI,aAAA,CAGJ,gCACI,sBAAA,CAIR,YACI,gBAAA,CAEA,wBAAA,CAAA,kBAAA,CACA,yBAAA,CACA,eAAA,CAEA,mBACI,qBAAA,CACA,iBAAA,CACA,kBAAA,CACA,WAAA,CAEA,sBAAA","file":"main.a84f16ea.chunk.css","sourcesContent":["body {\n padding: 0;\n margin: 0;\n font-family: \"Source Sans Pro\", sans-serif;\n font-size: 1rem;\n line-height: 1.4;\n}\n\n.highlighted-text, .status-bar {\n color: black;\n background-color: white;\n}\n\n.highlighted-text {\n border: 1px solid #d2d2d2;\n border-radius: 0 0 5px 5px;\n padding: 4px;\n cursor: pointer;\n\n .token.prefix ~ .token:not(.prefix) {\n color: #2563eb;\n }\n\n .token.active {\n outline: 1px solid #444;\n }\n}\n\n.status-bar {\n min-height: 1.4em;\n border: 1px solid #d2d2d2;\n border-bottom: none;\n border-radius: 5px 5px 0 0;\n padding: 3px 4px;\n\n .token {\n border: 1px solid #aaa;\n border-radius: 2px;\n background: #f5f5f5;\n padding: 1px;\n margin: 1px;\n margin-right: 2px;\n }\n}\n"]}
highlighted_text/build/static/css/main.b2f89d68.chunk.css.map DELETED
@@ -1 +0,0 @@
1
- {"version":3,"sources":["webpack://src/index.scss"],"names":[],"mappings":"AAAA,KACI,SAAA,CACA,QAAA,CACA,wCAAA,CACA,cAAA,CACA,eAAA,CAGJ,8BACI,UAAA,CACA,qBAAA,CAGJ,kBACI,wBAAA,CACA,yBAAA,CACA,WAAA,CACA,cAAA,CAEA,gCACI,sBAAA,CAIA,gCACI,WAAA,CAEA,+EACI,SAAA,CAGJ,wCACI,sCAAA,CAAA,8BAAA,CAMhB,YACI,gBAAA,CAEA,wBAAA,CAAA,kBAAA,CACA,yBAAA,CACA,eAAA,CAEA,mBACI,qBAAA,CACA,iBAAA,CACA,kBAAA,CACA,WAAA,CAEA,sBAAA","file":"main.b2f89d68.chunk.css","sourcesContent":["body {\n padding: 0;\n margin: 0;\n font-family: \"Source Sans Pro\", sans-serif;\n font-size: 1rem;\n line-height: 1.4;\n}\n\n.highlighted-text, .status-bar {\n color: black;\n background-color: white;\n}\n\n.highlighted-text {\n border: 1px solid #d2d2d2;\n border-radius: 0 0 5px 5px;\n padding: 4px;\n cursor: pointer;\n\n .token.active {\n outline: 1px solid #444;\n }\n\n &.frozen {\n .token {\n opacity: 0.75;\n\n &.context, &.active {\n opacity: 1;\n }\n\n &.context {\n text-decoration: #999 underline;\n }\n }\n }\n}\n\n.status-bar {\n min-height: 1.4em;\n border: 1px solid #d2d2d2;\n border-bottom: none;\n border-radius: 5px 5px 0 0;\n padding: 3px 4px;\n\n .token {\n border: 1px solid #aaa;\n border-radius: 2px;\n background: #f5f5f5;\n padding: 1px;\n margin: 1px;\n margin-right: 2px;\n }\n}\n"]}
 
 
highlighted_text/build/static/js/main.5f2b5265.chunk.js ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ (this.webpackJsonpstreamlit_component_template=this.webpackJsonpstreamlit_component_template||[]).push([[0],{27:function(t,e,n){},28:function(t,e,n){"use strict";n.r(e);var a=n(7),s=n.n(a),r=n(18),c=n.n(r),i=n(4),o=n(0),l=n(1),h=n(2),d=n(3),j=n(16),u=n(6),x=function(t){Object(h.a)(n,t);var e=Object(d.a)(n);function n(){var t;Object(o.a)(this,n);for(var a=arguments.length,s=new Array(a),r=0;r<a;r++)s[r]=arguments[r];return(t=e.call.apply(e,[this].concat(s))).state={activeIndex:null,hoverIndex:null,isFrozen:!1},t}return Object(l.a)(n,[{key:"render",value:function(){var t=this,e=this.props.args.tokens,n=this.getScores(),a=this.props.args.prefix_len,s="highlighted-text";this.state.isFrozen&&(s+=" frozen");var r=function(){t.setState({isFrozen:!1})};return Object(u.jsxs)(u.Fragment,{children:[Object(u.jsxs)("div",{className:"status-bar",children:[Object(u.jsxs)("span",{className:this.state.isFrozen?"":" d-none",children:[Object(u.jsx)("i",{className:"fa fa-lock"})," "]},"lock-icon"),null!=this.state.activeIndex?Object(u.jsxs)(u.Fragment,{children:[Object(u.jsx)("strong",{children:"index:"},"index-label")," ",Object(u.jsxs)("span",{children:[this.state.activeIndex," "]},"index")]}):Object(u.jsx)(u.Fragment,{})]},"status-bar"),Object(u.jsx)("div",{className:s,onClick:r,children:e.map((function(e,s){var c="token";t.state&&t.state.activeIndex==s&&(c+=" active"),s<a&&(c+=" prefix");var i={backgroundColor:n[s]>0?"rgba(32, 255, 32, ".concat(n[s],")"):"rgba(255, 32, 32, ".concat(-n[s],")")};return Object(u.jsx)("span",{className:c,style:i,onMouseOver:function(){t.state.isFrozen||t.setState({activeIndex:s}),t.setState({hoverIndex:s})},onClick:r,children:e},s)}))},"text")]})}},{key:"getScores",value:function(){var t=this.props.args.tokens;if(!this.state||null==this.state.activeIndex||this.state.activeIndex<1)return t.map((function(){return 0}));var e=this.props.args.scores,n=this.state.activeIndex-1,a=Math.min(Math.max(0,n),e[n].length),s=e[n].slice(0,a);s.reverse();var r=[].concat(Object(i.a)(Array(Math.max(0,n-s.length)).fill(0)),Object(i.a)(s.map((function(t){return void 0==t||isNaN(t)?0:t}))));return r=[].concat(Object(i.a)(r),Object(i.a)(Array(t.length-r.length).fill(0)))}}]),n}(j.a),b=Object(j.b)(x);n(27);c.a.render(Object(u.jsx)(s.a.StrictMode,{children:Object(u.jsx)(b,{})}),document.getElementById("root"))}},[[28,1,2]]]);
2
+ //# sourceMappingURL=main.5f2b5265.chunk.js.map
highlighted_text/build/static/js/main.5f2b5265.chunk.js.map ADDED
@@ -0,0 +1 @@
 
 
1
+ {"version":3,"sources":["HighlightedText.tsx","index.tsx"],"names":["HighlightedText","_StreamlitComponentBa","_inherits","_super","_createSuper","_this","_classCallCheck","_len","arguments","length","args","Array","_key","call","apply","concat","state","activeIndex","hoverIndex","isFrozen","_createClass","key","value","_this2","tokens","this","props","scores","getScores","prefixLength","className","onClick","setState","_jsxs","_Fragment","children","_jsx","map","t","i","style","backgroundColor","onMouseOver","allScores","hi","Math","min","max","row","slice","reverse","result","_toConsumableArray","fill","x","undefined","isNaN","StreamlitComponentBase","withStreamlitConnection","ReactDOM","render","React","StrictMode","document","getElementById"],"mappings":"gQAeMA,EAAe,SAAAC,GAAAC,YAAAF,EAAAC,GAAA,IAAAE,EAAAC,YAAAJ,GAAA,SAAAA,IAAA,IAAAK,EAAAC,YAAA,KAAAN,GAAA,QAAAO,EAAAC,UAAAC,OAAAC,EAAA,IAAAC,MAAAJ,GAAAK,EAAA,EAAAA,EAAAL,EAAAK,IAAAF,EAAAE,GAAAJ,UAAAI,GACqD,OADrDP,EAAAF,EAAAU,KAAAC,MAAAX,EAAA,OAAAY,OAAAL,KACVM,MAAQ,CAACC,YAAa,KAAMC,WAAY,KAAMC,UAAU,GAAMd,CAAC,CA6ErE,OA7EoEe,YAAApB,EAAA,EAAAqB,IAAA,SAAAC,MAErE,WAAU,IAADC,EAAA,KACCC,EAAmBC,KAAKC,MAAMhB,KAAa,OAC3CiB,EAAmBF,KAAKG,YACxBC,EAAeJ,KAAKC,MAAMhB,KAAiB,WAE7CoB,EAAY,mBACZL,KAAKT,MAAMG,WACXW,GAAa,WAGjB,IAAMC,EAAU,WACZR,EAAKS,SAAS,CAAEb,UAAU,GAC9B,EAEA,OAAOc,eAAAC,WAAA,CAAAC,SAAA,CACHF,eAAA,OAAKH,UAAU,aAAYK,SAAA,CACvBF,eAAA,QAAMH,UAAWL,KAAKT,MAAMG,SAAW,GAAK,UAAUgB,SAAA,CAAiBC,cAAA,KAAGN,UAAU,eAAiB,MAA1C,aAE7B,MAA1BL,KAAKT,MAAMC,YACXgB,eAAAC,WAAA,CAAAC,SAAA,CACIC,cAAA,UAAAD,SAA0B,UAAd,eAA6B,IAACF,eAAA,QAAAE,SAAA,CAAmBV,KAAKT,MAAMC,YAAY,MAAhC,YAEtDmB,cAAAF,WAAA,MAPsB,cAUhCE,cAAA,OAAKN,UAAWA,EAAWC,QAASA,EAAQI,SAEpCX,EAAOa,KAAI,SAACC,EAAWC,GACnB,IAAIT,EAAY,QACZP,EAAKP,OACDO,EAAKP,MAAMC,aAAesB,IAC1BT,GAAa,WAGjBS,EAAIV,IACJC,GAAa,WAEjB,IAAMU,EAAQ,CACVC,gBACId,EAAOY,GAAK,EAAC,qBAAAxB,OACcY,EAAOY,GAAE,0BAAAxB,QACRY,EAAOY,GAAE,MAS7C,OAAOH,cAAA,QAAcN,UAAWA,EAAWU,MAAOA,EAC9CE,YAPgB,WACXnB,EAAKP,MAAMG,UACZI,EAAKS,SAAS,CAAEf,YAAasB,IAEjChB,EAAKS,SAAS,CAAEd,WAAYqB,GAChC,EAE8BR,QAASA,EAAQI,SAAEG,GAD/BC,EAEtB,KA3ByC,UA+BzD,GAAC,CAAAlB,IAAA,YAAAC,MAED,WACI,IAAME,EAASC,KAAKC,MAAMhB,KAAa,OACvC,IAAKe,KAAKT,OAAmC,MAA1BS,KAAKT,MAAMC,aAAuBQ,KAAKT,MAAMC,YAAc,EAC1E,OAAOO,EAAOa,KAAI,kBAAM,CAAC,IAE7B,IAAMM,EAAwBlB,KAAKC,MAAMhB,KAAa,OAEhD6B,EAAId,KAAKT,MAAMC,YAAc,EAC7B2B,EAAKC,KAAKC,IAAID,KAAKE,IAAI,EAAGR,GAAII,EAAUJ,GAAG9B,QAC3CuC,EAAML,EAAUJ,GAAGU,MAAM,EAAGL,GAClCI,EAAIE,UACJ,IAAIC,EAAM,GAAApC,OAAAqC,YACHzC,MAAMkC,KAAKE,IAAI,EAAGR,EAAIS,EAAIvC,SAAS4C,KAAK,IAAED,YAC1CJ,EAAIX,KAAI,SAACiB,GAAC,YAAUC,GAALD,GAAkBE,MAAMF,GAAK,EAAIA,CAAC,MAGxD,OADAH,EAAM,GAAApC,OAAAqC,YAAOD,GAAMC,YAAKzC,MAAMa,EAAOf,OAAS0C,EAAO1C,QAAQ4C,KAAK,IAEtE,KAACrD,CAAA,CA9EgB,CAASyD,KAiFfC,cAAwB1D,G,MC3FvC2D,IAASC,OACPxB,cAACyB,IAAMC,WAAU,CAAA3B,SACfC,cAACpC,EAAe,MAElB+D,SAASC,eAAe,Q","file":"static/js/main.5f2b5265.chunk.js","sourcesContent":["import {\n StreamlitComponentBase,\n withStreamlitConnection,\n} from \"streamlit-component-lib\";\n\ntype HighlightedTextState = {\n activeIndex: number | null,\n hoverIndex: number | null,\n isFrozen: boolean\n};\n\n/**\n * This is a React-based component template. The `render()` function is called\n * automatically when your component should be re-rendered.\n */\nclass HighlightedText extends StreamlitComponentBase<HighlightedTextState> {\n public state = {activeIndex: null, hoverIndex: null, isFrozen: false};\n\n render() {\n const tokens: string[] = this.props.args[\"tokens\"];\n const scores: number[] = this.getScores();\n const prefixLength = this.props.args[\"prefix_len\"];\n\n let className = \"highlighted-text\";\n if (this.state.isFrozen) {\n className += \" frozen\";\n }\n\n const onClick = () => {\n this.setState({ isFrozen: false });\n };\n\n return <>\n <div className=\"status-bar\" key=\"status-bar\">\n <span className={this.state.isFrozen ? \"\" : \" d-none\"} key=\"lock-icon\"><i className=\"fa fa-lock\"></i> </span>\n {\n this.state.activeIndex != null ?\n <>\n <strong key=\"index-label\">index:</strong> <span key=\"index\">{this.state.activeIndex} </span>\n </>\n : <></>\n }\n </div>\n <div className={className} onClick={onClick} key=\"text\">\n {\n tokens.map((t: string, i: number) => {\n let className = \"token\";\n if (this.state) {\n if (this.state.activeIndex == i) {\n className += \" active\";\n }\n }\n if (i < prefixLength) {\n className += \" prefix\";\n }\n const style = {\n backgroundColor:\n scores[i] > 0\n ? `rgba(32, 255, 32, ${scores[i]})`\n : `rgba(255, 32, 32, ${-scores[i]})`\n };\n\n const onMouseOver = () => {\n if (!this.state.isFrozen) {\n this.setState({ activeIndex: i });\n }\n this.setState({ hoverIndex: i });\n };\n return <span key={i} className={className} style={style}\n onMouseOver={onMouseOver} onClick={onClick}>{t}</span>;\n })\n }\n </div>\n </>;\n }\n\n private getScores() {\n const tokens = this.props.args[\"tokens\"];\n if (!this.state || this.state.activeIndex == null || this.state.activeIndex < 1) {\n return tokens.map(() => 0);\n }\n const allScores: number[][] = this.props.args[\"scores\"];\n\n const i = this.state.activeIndex - 1;\n const hi = Math.min(Math.max(0, i), allScores[i].length);\n const row = allScores[i].slice(0, hi);\n row.reverse();\n let result = [\n ...Array(Math.max(0, i - row.length)).fill(0),\n ...row.map((x) => x == undefined || isNaN(x) ? 0 : x)\n ];\n result = [...result, ...Array(tokens.length - result.length).fill(0)];\n return result;\n }\n}\n\nexport default withStreamlitConnection(HighlightedText);\n","import React from \"react\";\nimport ReactDOM from \"react-dom\";\nimport HighlightedText from \"./HighlightedText\";\nimport \"./index.scss\";\n\nReactDOM.render(\n <React.StrictMode>\n <HighlightedText />\n </React.StrictMode>,\n document.getElementById(\"root\")\n)\n"],"sourceRoot":""}
highlighted_text/build/static/js/main.d475cbb2.chunk.js DELETED
@@ -1,2 +0,0 @@
1
- (this.webpackJsonpstreamlit_component_template=this.webpackJsonpstreamlit_component_template||[]).push([[0],{27:function(t,e,n){},28:function(t,e,n){"use strict";n.r(e);var a=n(7),s=n.n(a),c=n(18),r=n.n(c),i=n(4),o=n(0),l=n(1),h=n(2),d=n(3),j=n(16),u=n(6),b=function(t){Object(h.a)(n,t);var e=Object(d.a)(n);function n(){var t;Object(o.a)(this,n);for(var a=arguments.length,s=new Array(a),c=0;c<a;c++)s[c]=arguments[c];return(t=e.call.apply(e,[this].concat(s))).state={activeIndex:null,hoverIndex:null,isFrozen:!1},t}return Object(l.a)(n,[{key:"render",value:function(){var t=this,e=this.props.args.tokens,n=this.getScores(),a="highlighted-text";this.state.isFrozen&&(a+=" frozen");var s=function(){t.setState({isFrozen:!1})};return Object(u.jsxs)(u.Fragment,{children:[Object(u.jsxs)("div",{className:"status-bar",children:[Object(u.jsxs)("span",{className:this.state.isFrozen?"":" d-none",children:[Object(u.jsx)("i",{className:"fa fa-lock"})," "]},"lock-icon"),null!=this.state.activeIndex?Object(u.jsxs)(u.Fragment,{children:[Object(u.jsx)("strong",{children:"index:"},"index-label")," ",Object(u.jsxs)("span",{children:[this.state.activeIndex," "]},"index")]}):Object(u.jsx)(u.Fragment,{})]},"status-bar"),Object(u.jsx)("div",{className:a,onClick:s,children:e.map((function(e,a){var c="token";t.state&&t.state.activeIndex==a&&(c+=" active");var r={backgroundColor:n[a]>0?"rgba(32, 255, 32, ".concat(n[a],")"):"rgba(255, 32, 32, ".concat(-n[a],")")};return Object(u.jsx)("span",{className:c,style:r,onMouseOver:function(){t.state.isFrozen||t.setState({activeIndex:a}),t.setState({hoverIndex:a})},onClick:s,children:e},a)}))},"text")]})}},{key:"getScores",value:function(){var t=this.props.args.tokens;if(!this.state||null==this.state.activeIndex||this.state.activeIndex<1)return t.map((function(){return 0}));var e=this.props.args.scores,n=this.state.activeIndex-1,a=Math.min(Math.max(0,n),e[n].length),s=e[n].slice(0,a);s.reverse();var c=[].concat(Object(i.a)(Array(Math.max(0,n-s.length)).fill(0)),Object(i.a)(s.map((function(t){return void 0==t||isNaN(t)?0:t}))));return c=[].concat(Object(i.a)(c),Object(i.a)(Array(t.length-c.length).fill(0)))}}]),n}(j.a),v=Object(j.b)(b);n(27);r.a.render(Object(u.jsx)(s.a.StrictMode,{children:Object(u.jsx)(v,{})}),document.getElementById("root"))}},[[28,1,2]]]);
2
- //# sourceMappingURL=main.d475cbb2.chunk.js.map
 
 
 
highlighted_text/build/static/js/main.d475cbb2.chunk.js.map DELETED
@@ -1 +0,0 @@
1
- {"version":3,"sources":["HighlightedText.tsx","index.tsx"],"names":["HighlightedText","_StreamlitComponentBa","_inherits","_super","_createSuper","_this","_classCallCheck","_len","arguments","length","args","Array","_key","call","apply","concat","state","activeIndex","hoverIndex","isFrozen","_createClass","key","value","_this2","tokens","this","props","scores","getScores","className","onClick","setState","_jsxs","_Fragment","children","_jsx","map","t","i","style","backgroundColor","onMouseOver","allScores","hi","Math","min","max","row","slice","reverse","result","_toConsumableArray","fill","x","undefined","isNaN","StreamlitComponentBase","withStreamlitConnection","ReactDOM","render","React","StrictMode","document","getElementById"],"mappings":"gQAeMA,EAAe,SAAAC,GAAAC,YAAAF,EAAAC,GAAA,IAAAE,EAAAC,YAAAJ,GAAA,SAAAA,IAAA,IAAAK,EAAAC,YAAA,KAAAN,GAAA,QAAAO,EAAAC,UAAAC,OAAAC,EAAA,IAAAC,MAAAJ,GAAAK,EAAA,EAAAA,EAAAL,EAAAK,IAAAF,EAAAE,GAAAJ,UAAAI,GACqD,OADrDP,EAAAF,EAAAU,KAAAC,MAAAX,EAAA,OAAAY,OAAAL,KACVM,MAAQ,CAACC,YAAa,KAAMC,WAAY,KAAMC,UAAU,GAAMd,CAAC,CAyErE,OAzEoEe,YAAApB,EAAA,EAAAqB,IAAA,SAAAC,MAErE,WAAU,IAADC,EAAA,KACCC,EAAmBC,KAAKC,MAAMhB,KAAa,OAC3CiB,EAAmBF,KAAKG,YAE1BC,EAAY,mBACZJ,KAAKT,MAAMG,WACXU,GAAa,WAGjB,IAAMC,EAAU,WACZP,EAAKQ,SAAS,CAAEZ,UAAU,GAC9B,EAEA,OAAOa,eAAAC,WAAA,CAAAC,SAAA,CACHF,eAAA,OAAKH,UAAU,aAAYK,SAAA,CACvBF,eAAA,QAAMH,UAAWJ,KAAKT,MAAMG,SAAW,GAAK,UAAUe,SAAA,CAAiBC,cAAA,KAAGN,UAAU,eAAiB,MAA1C,aAE7B,MAA1BJ,KAAKT,MAAMC,YACXe,eAAAC,WAAA,CAAAC,SAAA,CACIC,cAAA,UAAAD,SAA0B,UAAd,eAA6B,IAACF,eAAA,QAAAE,SAAA,CAAmBT,KAAKT,MAAMC,YAAY,MAAhC,YAEtDkB,cAAAF,WAAA,MAPsB,cAUhCE,cAAA,OAAKN,UAAWA,EAAWC,QAASA,EAAQI,SAEpCV,EAAOY,KAAI,SAACC,EAAWC,GACnB,IAAIT,EAAY,QACZN,EAAKP,OACDO,EAAKP,MAAMC,aAAeqB,IAC1BT,GAAa,WAGrB,IAAMU,EAAQ,CACVC,gBACIb,EAAOW,GAAK,EAAC,qBAAAvB,OACcY,EAAOW,GAAE,0BAAAvB,QACRY,EAAOW,GAAE,MAS7C,OAAOH,cAAA,QAAcN,UAAWA,EAAWU,MAAOA,EAC9CE,YAPgB,WACXlB,EAAKP,MAAMG,UACZI,EAAKQ,SAAS,CAAEd,YAAaqB,IAEjCf,EAAKQ,SAAS,CAAEb,WAAYoB,GAChC,EAE8BR,QAASA,EAAQI,SAAEG,GAD/BC,EAEtB,KAxByC,UA4BzD,GAAC,CAAAjB,IAAA,YAAAC,MAED,WACI,IAAME,EAASC,KAAKC,MAAMhB,KAAa,OACvC,IAAKe,KAAKT,OAAmC,MAA1BS,KAAKT,MAAMC,aAAuBQ,KAAKT,MAAMC,YAAc,EAC1E,OAAOO,EAAOY,KAAI,kBAAM,CAAC,IAE7B,IAAMM,EAAwBjB,KAAKC,MAAMhB,KAAa,OAEhD4B,EAAIb,KAAKT,MAAMC,YAAc,EAC7B0B,EAAKC,KAAKC,IAAID,KAAKE,IAAI,EAAGR,GAAII,EAAUJ,GAAG7B,QAC3CsC,EAAML,EAAUJ,GAAGU,MAAM,EAAGL,GAClCI,EAAIE,UACJ,IAAIC,EAAM,GAAAnC,OAAAoC,YACHxC,MAAMiC,KAAKE,IAAI,EAAGR,EAAIS,EAAItC,SAAS2C,KAAK,IAAED,YAC1CJ,EAAIX,KAAI,SAACiB,GAAC,YAAUC,GAALD,GAAkBE,MAAMF,GAAK,EAAIA,CAAC,MAGxD,OADAH,EAAM,GAAAnC,OAAAoC,YAAOD,GAAMC,YAAKxC,MAAMa,EAAOf,OAASyC,EAAOzC,QAAQ2C,KAAK,IAEtE,KAACpD,CAAA,CA1EgB,CAASwD,KA6EfC,cAAwBzD,G,MCvFvC0D,IAASC,OACPxB,cAACyB,IAAMC,WAAU,CAAA3B,SACfC,cAACnC,EAAe,MAElB8D,SAASC,eAAe,Q","file":"static/js/main.d475cbb2.chunk.js","sourcesContent":["import {\n StreamlitComponentBase,\n withStreamlitConnection,\n} from \"streamlit-component-lib\";\n\ntype HighlightedTextState = {\n activeIndex: number | null,\n hoverIndex: number | null,\n isFrozen: boolean\n};\n\n/**\n * This is a React-based component template. The `render()` function is called\n * automatically when your component should be re-rendered.\n */\nclass HighlightedText extends StreamlitComponentBase<HighlightedTextState> {\n public state = {activeIndex: null, hoverIndex: null, isFrozen: false};\n\n render() {\n const tokens: string[] = this.props.args[\"tokens\"];\n const scores: number[] = this.getScores();\n\n let className = \"highlighted-text\";\n if (this.state.isFrozen) {\n className += \" frozen\";\n }\n\n const onClick = () => {\n this.setState({ isFrozen: false });\n };\n\n return <>\n <div className=\"status-bar\" key=\"status-bar\">\n <span className={this.state.isFrozen ? \"\" : \" d-none\"} key=\"lock-icon\"><i className=\"fa fa-lock\"></i> </span>\n {\n this.state.activeIndex != null ?\n <>\n <strong key=\"index-label\">index:</strong> <span key=\"index\">{this.state.activeIndex} </span>\n </>\n : <></>\n }\n </div>\n <div className={className} onClick={onClick} key=\"text\">\n {\n tokens.map((t: string, i: number) => {\n let className = \"token\";\n if (this.state) {\n if (this.state.activeIndex == i) {\n className += \" active\";\n }\n }\n const style = {\n backgroundColor:\n scores[i] > 0\n ? `rgba(32, 255, 32, ${scores[i]})`\n : `rgba(255, 32, 32, ${-scores[i]})`\n };\n\n const onMouseOver = () => {\n if (!this.state.isFrozen) {\n this.setState({ activeIndex: i });\n }\n this.setState({ hoverIndex: i });\n };\n return <span key={i} className={className} style={style}\n onMouseOver={onMouseOver} onClick={onClick}>{t}</span>;\n })\n }\n </div>\n </>;\n }\n\n private getScores() {\n const tokens = this.props.args[\"tokens\"];\n if (!this.state || this.state.activeIndex == null || this.state.activeIndex < 1) {\n return tokens.map(() => 0);\n }\n const allScores: number[][] = this.props.args[\"scores\"];\n\n const i = this.state.activeIndex - 1;\n const hi = Math.min(Math.max(0, i), allScores[i].length);\n const row = allScores[i].slice(0, hi);\n row.reverse();\n let result = [\n ...Array(Math.max(0, i - row.length)).fill(0), \n ...row.map((x) => x == undefined || isNaN(x) ? 0 : x)\n ];\n result = [...result, ...Array(tokens.length - result.length).fill(0)];\n return result;\n }\n}\n\nexport default withStreamlitConnection(HighlightedText);\n","import React from \"react\";\nimport ReactDOM from \"react-dom\";\nimport HighlightedText from \"./HighlightedText\";\nimport \"./index.scss\";\n\nReactDOM.render(\n <React.StrictMode>\n <HighlightedText />\n </React.StrictMode>,\n document.getElementById(\"root\")\n)\n"],"sourceRoot":""}
 
 
highlighted_text/src/HighlightedText.tsx CHANGED
@@ -19,6 +19,7 @@ class HighlightedText extends StreamlitComponentBase<HighlightedTextState> {
19
  render() {
20
  const tokens: string[] = this.props.args["tokens"];
21
  const scores: number[] = this.getScores();
 
22
 
23
  let className = "highlighted-text";
24
  if (this.state.isFrozen) {
@@ -49,6 +50,9 @@ class HighlightedText extends StreamlitComponentBase<HighlightedTextState> {
49
  className += " active";
50
  }
51
  }
 
 
 
52
  const style = {
53
  backgroundColor:
54
  scores[i] > 0
 
19
  render() {
20
  const tokens: string[] = this.props.args["tokens"];
21
  const scores: number[] = this.getScores();
22
+ const prefixLength: number = this.props.args["prefix_len"];
23
 
24
  let className = "highlighted-text";
25
  if (this.state.isFrozen) {
 
50
  className += " active";
51
  }
52
  }
53
+ if (i < prefixLength) {
54
+ className += " prefix";
55
+ }
56
  const style = {
57
  backgroundColor:
58
  scores[i] > 0
highlighted_text/src/index.scss CHANGED
@@ -17,22 +17,12 @@ body {
17
  padding: 4px;
18
  cursor: pointer;
19
 
20
- .token.active {
21
- outline: 1px solid #444;
22
  }
23
 
24
- &.frozen {
25
- .token {
26
- opacity: 0.75;
27
-
28
- &.context, &.active {
29
- opacity: 1;
30
- }
31
-
32
- &.context {
33
- text-decoration: #999 underline;
34
- }
35
- }
36
  }
37
  }
38
 
 
17
  padding: 4px;
18
  cursor: pointer;
19
 
20
+ .token.prefix ~ .token:not(.prefix) {
21
+ color: #2563eb;
22
  }
23
 
24
+ .token.active {
25
+ outline: 1px solid #444;
 
 
 
 
 
 
 
 
 
 
26
  }
27
  }
28