Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -9,6 +9,8 @@ import ast
|
|
9 |
import logging
|
10 |
import numpy as np
|
11 |
import plotly.graph_objects as go
|
|
|
|
|
12 |
|
13 |
# Set up logging
|
14 |
logging.basicConfig(level=logging.DEBUG)
|
@@ -24,24 +26,7 @@ def parse_input(json_input):
|
|
24 |
return data
|
25 |
except json.JSONDecodeError as e:
|
26 |
logger.error("JSON parsing failed: %s (Input: %s)", str(e), json_input[:100] + "..." if len(json_input) > 100 else json_input)
|
27 |
-
|
28 |
-
# If JSON fails, try to parse as Python literal (e.g., with single quotes), but only for JSON-like strings
|
29 |
-
data = ast.literal_eval(json_input)
|
30 |
-
logger.debug("Successfully parsed as Python literal")
|
31 |
-
# Convert Python dictionary to JSON-compatible format (replace single quotes with double quotes)
|
32 |
-
def dict_to_json(obj):
|
33 |
-
if isinstance(obj, dict):
|
34 |
-
return {str(k): dict_to_json(v) for k, v in obj.items()}
|
35 |
-
elif isinstance(obj, list):
|
36 |
-
return [dict_to_json(item) for item in obj]
|
37 |
-
else:
|
38 |
-
return obj
|
39 |
-
converted_data = dict_to_json(data)
|
40 |
-
logger.debug("Converted to JSON-compatible format")
|
41 |
-
return converted_data
|
42 |
-
except (SyntaxError, ValueError) as e:
|
43 |
-
logger.error("Python literal parsing failed: %s (Input: %s)", str(e), json_input[:100] + "..." if len(json_input) > 100 else json_input)
|
44 |
-
raise ValueError(f"Malformed input: {str(e)}. Ensure property names are in double quotes (e.g., \"content\") and the format matches JSON (e.g., {{\"content\": [...]}}).")
|
45 |
|
46 |
# Function to ensure a value is a float, converting from string if necessary
|
47 |
def ensure_float(value):
|
@@ -69,10 +54,59 @@ def get_token(entry):
|
|
69 |
def create_empty_figure(title):
|
70 |
return go.Figure().update_layout(title=title, xaxis_title="", yaxis_title="", showlegend=False)
|
71 |
|
72 |
-
#
|
73 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
74 |
try:
|
75 |
-
# Parse the input (handles JSON only
|
76 |
data = parse_input(json_input)
|
77 |
|
78 |
# Ensure data is a dictionary with 'content' key containing a list
|
@@ -94,14 +128,13 @@ def visualize_logprobs(json_input):
|
|
94 |
logger.warning("Skipping non-dictionary entry: %s", entry)
|
95 |
continue
|
96 |
logprob = ensure_float(entry.get("logprob", None))
|
97 |
-
if logprob >= -100000: # Include all entries with default 0.0
|
98 |
-
|
99 |
-
tokens.append(token)
|
100 |
logprobs.append(logprob)
|
101 |
# Get top_logprobs, default to empty dict if None
|
102 |
top_probs = entry.get("top_logprobs", {})
|
103 |
if top_probs is None:
|
104 |
-
logger.debug("top_logprobs is None for token: %s, using empty dict",
|
105 |
top_probs = {} # Default to empty dict for None
|
106 |
# Ensure all values in top_logprobs are floats and create a list of tuples
|
107 |
finite_top_probs = []
|
@@ -115,53 +148,61 @@ def visualize_logprobs(json_input):
|
|
115 |
else:
|
116 |
logger.debug("Skipping entry with logprob: %s (type: %s)", entry.get("logprob"), type(entry.get("logprob", None)))
|
117 |
|
118 |
-
# Check if there's valid data after filtering
|
119 |
if not logprobs or not tokens:
|
120 |
-
return (create_empty_figure("Log Probabilities of Generated Tokens"), None, "No tokens to display.", create_empty_figure("Top Token Log Probabilities"), create_empty_figure("Significant Probability Drops"))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
121 |
|
122 |
# 1. Main Log Probability Plot (Interactive Plotly)
|
123 |
main_fig = go.Figure()
|
124 |
-
main_fig.add_trace(go.Scatter(x=list(range(len(
|
125 |
main_fig.update_layout(
|
126 |
-
title="Log Probabilities of Generated Tokens",
|
127 |
-
xaxis_title="Token Position",
|
128 |
yaxis_title="Log Probability",
|
129 |
hovermode="closest",
|
130 |
clickmode='event+select'
|
131 |
)
|
132 |
main_fig.update_traces(
|
133 |
-
customdata=[f"Token: {tok}, Log Prob: {prob:.4f}, Position: {i}" for i, (tok, prob) in enumerate(zip(
|
134 |
hovertemplate='<b>%{customdata}</b><extra></extra>'
|
135 |
)
|
136 |
|
137 |
# 2. Probability Drop Analysis (Interactive Plotly)
|
138 |
-
if len(
|
139 |
-
drops_fig = create_empty_figure("Significant Probability Drops")
|
140 |
else:
|
141 |
-
drops = [
|
142 |
drops_fig = go.Figure()
|
143 |
drops_fig.add_trace(go.Bar(x=list(range(len(drops))), y=drops, name='Drop', marker_color='red'))
|
144 |
drops_fig.update_layout(
|
145 |
-
title="Significant Probability Drops",
|
146 |
-
xaxis_title="Token Position",
|
147 |
yaxis_title="Log Probability Drop",
|
148 |
hovermode="closest",
|
149 |
clickmode='event+select'
|
150 |
)
|
151 |
drops_fig.update_traces(
|
152 |
-
customdata=[f"Drop: {drop:.4f}, From: {
|
153 |
hovertemplate='<b>%{customdata}</b><extra></extra>'
|
154 |
)
|
155 |
|
156 |
# Create DataFrame for the table with dynamic top_logprobs
|
157 |
table_data = []
|
158 |
-
max_alternatives = max(len(alts) for alts in
|
159 |
-
for i, entry in enumerate(content):
|
160 |
if not isinstance(entry, dict):
|
161 |
continue
|
162 |
logprob = ensure_float(entry.get("logprob", None))
|
163 |
if logprob >= -100000 and "top_logprobs" in entry: # Include all entries with default 0.0
|
164 |
-
token = get_token(entry)
|
165 |
top_logprobs = entry.get("top_logprobs", {})
|
166 |
if top_logprobs is None:
|
167 |
logger.debug("top_logprobs is None for token: %s, using empty dict", token)
|
@@ -191,38 +232,38 @@ def visualize_logprobs(json_input):
|
|
191 |
else None
|
192 |
)
|
193 |
|
194 |
-
# Generate colored text
|
195 |
-
if
|
196 |
-
min_logprob = min(
|
197 |
-
max_logprob = max(
|
198 |
if max_logprob == min_logprob:
|
199 |
-
normalized_probs = [0.5] * len(
|
200 |
else:
|
201 |
normalized_probs = [
|
202 |
-
(lp - min_logprob) / (max_logprob - min_logprob) for lp in
|
203 |
]
|
204 |
|
205 |
colored_text = ""
|
206 |
-
for i, (token, norm_prob) in enumerate(zip(
|
207 |
r = int(255 * (1 - norm_prob)) # Red for low confidence
|
208 |
g = int(255 * norm_prob) # Green for high confidence
|
209 |
b = 0
|
210 |
color = f"rgb({r}, {g}, {b})"
|
211 |
colored_text += f'<span style="color: {color}; font-weight: bold;">{token}</span>'
|
212 |
-
if i < len(
|
213 |
colored_text += " "
|
214 |
colored_text_html = f"<p>{colored_text}</p>"
|
215 |
else:
|
216 |
-
colored_text_html = "No tokens to display."
|
217 |
|
218 |
-
# Top Token Log Probabilities (Interactive Plotly, dynamic length)
|
219 |
-
alt_viz_fig = create_empty_figure("Top Token Log Probabilities") if not
|
220 |
-
if
|
221 |
-
for i, (token, probs) in enumerate(zip(
|
222 |
for j, (alt_tok, prob) in enumerate(probs):
|
223 |
-
alt_viz_fig.add_trace(go.Bar(x=[f"{token} (Pos {i})"], y=[prob], name=f"{alt_tok}", marker_color=['blue', 'green', 'red', 'purple', 'orange'][:len(probs)]))
|
224 |
alt_viz_fig.update_layout(
|
225 |
-
title="Top Token Log Probabilities",
|
226 |
xaxis_title="Token (Position)",
|
227 |
yaxis_title="Log Probability",
|
228 |
barmode='stack',
|
@@ -230,21 +271,21 @@ def visualize_logprobs(json_input):
|
|
230 |
clickmode='event+select'
|
231 |
)
|
232 |
alt_viz_fig.update_traces(
|
233 |
-
customdata=[f"Token: {tok}, Alt: {alt}, Log Prob: {prob:.4f}, Position: {i}" for i, (tok, alts) in enumerate(zip(
|
234 |
hovertemplate='<b>%{customdata}</b><extra></extra>'
|
235 |
)
|
236 |
|
237 |
-
return (main_fig, df, colored_text_html, alt_viz_fig, drops_fig)
|
238 |
|
239 |
except Exception as e:
|
240 |
logger.error("Visualization failed: %s (Input: %s)", str(e), json_input[:100] + "..." if len(json_input) > 100 else json_input)
|
241 |
-
return (create_empty_figure("Log Probabilities of Generated Tokens"), None, "No finite log probabilities to display.", create_empty_figure("Top Token Log Probabilities"), create_empty_figure("Significant Probability Drops"))
|
242 |
|
243 |
-
# Gradio interface with
|
244 |
with gr.Blocks(title="Log Probability Visualizer") as app:
|
245 |
gr.Markdown("# Log Probability Visualizer")
|
246 |
gr.Markdown(
|
247 |
-
"Paste your JSON log prob data below to visualize
|
248 |
)
|
249 |
|
250 |
with gr.Row():
|
@@ -253,6 +294,7 @@ with gr.Blocks(title="Log Probability Visualizer") as app:
|
|
253 |
lines=10,
|
254 |
placeholder="Paste your JSON (e.g., {\"content\": [{\"bytes\": [44], \"logprob\": 0.0, \"token\": \",\", \"top_logprobs\": {\" so\": -13.8046875, \".\": -13.8046875, \",\": -13.640625}}]}).",
|
255 |
)
|
|
|
256 |
|
257 |
with gr.Row():
|
258 |
plot_output = gr.Plot(label="Log Probability Plot (Click for Tokens)")
|
@@ -265,11 +307,67 @@ with gr.Blocks(title="Log Probability Visualizer") as app:
|
|
265 |
with gr.Row():
|
266 |
text_output = gr.HTML(label="Colored Text (Confidence Visualization)")
|
267 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
268 |
btn = gr.Button("Visualize")
|
269 |
btn.click(
|
270 |
fn=visualize_logprobs,
|
271 |
-
inputs=[json_input],
|
272 |
-
outputs=[plot_output, table_output, text_output, alt_viz_output, drops_output],
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
273 |
)
|
274 |
|
275 |
app.launch()
|
|
|
9 |
import logging
|
10 |
import numpy as np
|
11 |
import plotly.graph_objects as go
|
12 |
+
import asyncio
|
13 |
+
import anyio
|
14 |
|
15 |
# Set up logging
|
16 |
logging.basicConfig(level=logging.DEBUG)
|
|
|
26 |
return data
|
27 |
except json.JSONDecodeError as e:
|
28 |
logger.error("JSON parsing failed: %s (Input: %s)", str(e), json_input[:100] + "..." if len(json_input) > 100 else json_input)
|
29 |
+
raise ValueError(f"Malformed input: {str(e)}. Ensure property names are in double quotes (e.g., \"content\") and the format matches JSON (e.g., {{\"content\": [...]}}).")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
|
31 |
# Function to ensure a value is a float, converting from string if necessary
|
32 |
def ensure_float(value):
|
|
|
54 |
def create_empty_figure(title):
|
55 |
return go.Figure().update_layout(title=title, xaxis_title="", yaxis_title="", showlegend=False)
|
56 |
|
57 |
+
# Precompute the next chunk asynchronously
|
58 |
+
async def precompute_chunk(json_input, chunk_size, current_chunk):
|
59 |
+
try:
|
60 |
+
data = parse_input(json_input)
|
61 |
+
content = data.get("content", []) if isinstance(data, dict) else data
|
62 |
+
if not isinstance(content, list):
|
63 |
+
raise ValueError("Content must be a list of entries")
|
64 |
+
|
65 |
+
tokens = []
|
66 |
+
logprobs = []
|
67 |
+
top_alternatives = []
|
68 |
+
for entry in content:
|
69 |
+
if not isinstance(entry, dict):
|
70 |
+
logger.warning("Skipping non-dictionary entry: %s", entry)
|
71 |
+
continue
|
72 |
+
logprob = ensure_float(entry.get("logprob", None))
|
73 |
+
if logprob >= -100000: # Include all entries with default 0.0
|
74 |
+
tokens.append(get_token(entry))
|
75 |
+
logprobs.append(logprob)
|
76 |
+
top_probs = entry.get("top_logprobs", {})
|
77 |
+
if top_probs is None:
|
78 |
+
logger.debug("top_logprobs is None for token: %s, using empty dict", get_token(entry))
|
79 |
+
top_probs = {}
|
80 |
+
finite_top_probs = []
|
81 |
+
for key, value in top_probs.items():
|
82 |
+
float_value = ensure_float(value)
|
83 |
+
if float_value is not None and math.isfinite(float_value):
|
84 |
+
finite_top_probs.append((key, float_value))
|
85 |
+
sorted_probs = sorted(finite_top_probs, key=lambda x: x[1], reverse=True)
|
86 |
+
top_alternatives.append(sorted_probs)
|
87 |
+
|
88 |
+
if not tokens or not logprobs:
|
89 |
+
return None, None, None
|
90 |
+
|
91 |
+
next_chunk = current_chunk + 1
|
92 |
+
start_idx = next_chunk * chunk_size
|
93 |
+
end_idx = min((next_chunk + 1) * chunk_size, len(tokens))
|
94 |
+
if start_idx >= len(tokens):
|
95 |
+
return None, None, None
|
96 |
+
|
97 |
+
paginated_tokens = tokens[start_idx:end_idx]
|
98 |
+
paginated_logprobs = logprobs[start_idx:end_idx]
|
99 |
+
paginated_alternatives = top_alternatives[start_idx:end_idx]
|
100 |
+
|
101 |
+
return paginated_tokens, paginated_logprobs, paginated_alternatives
|
102 |
+
except Exception as e:
|
103 |
+
logger.error("Precomputation failed for chunk %d: %s", current_chunk + 1, str(e))
|
104 |
+
return None, None, None
|
105 |
+
|
106 |
+
# Function to process and visualize a chunk of log probs with dynamic top_logprobs
|
107 |
+
def visualize_logprobs(json_input, chunk=0, chunk_size=1000):
|
108 |
try:
|
109 |
+
# Parse the input (handles JSON only)
|
110 |
data = parse_input(json_input)
|
111 |
|
112 |
# Ensure data is a dictionary with 'content' key containing a list
|
|
|
128 |
logger.warning("Skipping non-dictionary entry: %s", entry)
|
129 |
continue
|
130 |
logprob = ensure_float(entry.get("logprob", None))
|
131 |
+
if logprob >= -100000: # Include all entries with default 0.0
|
132 |
+
tokens.append(get_token(entry))
|
|
|
133 |
logprobs.append(logprob)
|
134 |
# Get top_logprobs, default to empty dict if None
|
135 |
top_probs = entry.get("top_logprobs", {})
|
136 |
if top_probs is None:
|
137 |
+
logger.debug("top_logprobs is None for token: %s, using empty dict", get_token(entry))
|
138 |
top_probs = {} # Default to empty dict for None
|
139 |
# Ensure all values in top_logprobs are floats and create a list of tuples
|
140 |
finite_top_probs = []
|
|
|
148 |
else:
|
149 |
logger.debug("Skipping entry with logprob: %s (type: %s)", entry.get("logprob"), type(entry.get("logprob", None)))
|
150 |
|
151 |
+
# Check if there's valid data after filtering
|
152 |
if not logprobs or not tokens:
|
153 |
+
return (create_empty_figure("Log Probabilities of Generated Tokens"), None, "No tokens to display.", create_empty_figure("Top Token Log Probabilities"), create_empty_figure("Significant Probability Drops"), 1, 0)
|
154 |
+
|
155 |
+
# Paginate data for chunks of 1,000 tokens
|
156 |
+
total_chunks = max(1, (len(logprobs) + chunk_size - 1) // chunk_size)
|
157 |
+
start_idx = chunk * chunk_size
|
158 |
+
end_idx = min((chunk + 1) * chunk_size, len(logprobs))
|
159 |
+
paginated_tokens = tokens[start_idx:end_idx]
|
160 |
+
paginated_logprobs = logprobs[start_idx:end_idx]
|
161 |
+
paginated_alternatives = top_alternatives[start_idx:end_idx] if top_alternatives else []
|
162 |
|
163 |
# 1. Main Log Probability Plot (Interactive Plotly)
|
164 |
main_fig = go.Figure()
|
165 |
+
main_fig.add_trace(go.Scatter(x=list(range(len(paginated_logprobs))), y=paginated_logprobs, mode='markers+lines', name='Log Prob', marker=dict(color='blue')))
|
166 |
main_fig.update_layout(
|
167 |
+
title="Log Probabilities of Generated Tokens (Chunk %d)" % (chunk + 1),
|
168 |
+
xaxis_title="Token Position (within chunk)",
|
169 |
yaxis_title="Log Probability",
|
170 |
hovermode="closest",
|
171 |
clickmode='event+select'
|
172 |
)
|
173 |
main_fig.update_traces(
|
174 |
+
customdata=[f"Token: {tok}, Log Prob: {prob:.4f}, Position: {i+start_idx}" for i, (tok, prob) in enumerate(zip(paginated_tokens, paginated_logprobs))],
|
175 |
hovertemplate='<b>%{customdata}</b><extra></extra>'
|
176 |
)
|
177 |
|
178 |
# 2. Probability Drop Analysis (Interactive Plotly)
|
179 |
+
if len(paginated_logprobs) < 2:
|
180 |
+
drops_fig = create_empty_figure("Significant Probability Drops (Chunk %d)" % (chunk + 1))
|
181 |
else:
|
182 |
+
drops = [paginated_logprobs[i+1] - paginated_logprobs[i] for i in range(len(paginated_logprobs)-1)]
|
183 |
drops_fig = go.Figure()
|
184 |
drops_fig.add_trace(go.Bar(x=list(range(len(drops))), y=drops, name='Drop', marker_color='red'))
|
185 |
drops_fig.update_layout(
|
186 |
+
title="Significant Probability Drops (Chunk %d)" % (chunk + 1),
|
187 |
+
xaxis_title="Token Position (within chunk)",
|
188 |
yaxis_title="Log Probability Drop",
|
189 |
hovermode="closest",
|
190 |
clickmode='event+select'
|
191 |
)
|
192 |
drops_fig.update_traces(
|
193 |
+
customdata=[f"Drop: {drop:.4f}, From: {paginated_tokens[i]} to {paginated_tokens[i+1]}, Position: {i+start_idx}" for i, drop in enumerate(drops)],
|
194 |
hovertemplate='<b>%{customdata}</b><extra></extra>'
|
195 |
)
|
196 |
|
197 |
# Create DataFrame for the table with dynamic top_logprobs
|
198 |
table_data = []
|
199 |
+
max_alternatives = max(len(alts) for alts in paginated_alternatives) if paginated_alternatives else 0
|
200 |
+
for i, entry in enumerate(content[start_idx:end_idx]):
|
201 |
if not isinstance(entry, dict):
|
202 |
continue
|
203 |
logprob = ensure_float(entry.get("logprob", None))
|
204 |
if logprob >= -100000 and "top_logprobs" in entry: # Include all entries with default 0.0
|
205 |
+
token = get_token(entry)
|
206 |
top_logprobs = entry.get("top_logprobs", {})
|
207 |
if top_logprobs is None:
|
208 |
logger.debug("top_logprobs is None for token: %s, using empty dict", token)
|
|
|
232 |
else None
|
233 |
)
|
234 |
|
235 |
+
# Generate colored text (for the current chunk)
|
236 |
+
if paginated_logprobs:
|
237 |
+
min_logprob = min(paginated_logprobs)
|
238 |
+
max_logprob = max(paginated_logprobs)
|
239 |
if max_logprob == min_logprob:
|
240 |
+
normalized_probs = [0.5] * len(paginated_logprobs)
|
241 |
else:
|
242 |
normalized_probs = [
|
243 |
+
(lp - min_logprob) / (max_logprob - min_logprob) for lp in paginated_logprobs
|
244 |
]
|
245 |
|
246 |
colored_text = ""
|
247 |
+
for i, (token, norm_prob) in enumerate(zip(paginated_tokens, normalized_probs)):
|
248 |
r = int(255 * (1 - norm_prob)) # Red for low confidence
|
249 |
g = int(255 * norm_prob) # Green for high confidence
|
250 |
b = 0
|
251 |
color = f"rgb({r}, {g}, {b})"
|
252 |
colored_text += f'<span style="color: {color}; font-weight: bold;">{token}</span>'
|
253 |
+
if i < len(paginated_tokens) - 1:
|
254 |
colored_text += " "
|
255 |
colored_text_html = f"<p>{colored_text}</p>"
|
256 |
else:
|
257 |
+
colored_text_html = "No tokens to display in this chunk."
|
258 |
|
259 |
+
# Top Token Log Probabilities (Interactive Plotly, dynamic length, for the current chunk)
|
260 |
+
alt_viz_fig = create_empty_figure("Top Token Log Probabilities (Chunk %d)" % (chunk + 1)) if not paginated_logprobs or not paginated_alternatives else go.Figure()
|
261 |
+
if paginated_logprobs and paginated_alternatives:
|
262 |
+
for i, (token, probs) in enumerate(zip(paginated_tokens, paginated_alternatives)):
|
263 |
for j, (alt_tok, prob) in enumerate(probs):
|
264 |
+
alt_viz_fig.add_trace(go.Bar(x=[f"{token} (Pos {i+start_idx})"], y=[prob], name=f"{alt_tok}", marker_color=['blue', 'green', 'red', 'purple', 'orange'][:len(probs)]))
|
265 |
alt_viz_fig.update_layout(
|
266 |
+
title="Top Token Log Probabilities (Chunk %d)" % (chunk + 1),
|
267 |
xaxis_title="Token (Position)",
|
268 |
yaxis_title="Log Probability",
|
269 |
barmode='stack',
|
|
|
271 |
clickmode='event+select'
|
272 |
)
|
273 |
alt_viz_fig.update_traces(
|
274 |
+
customdata=[f"Token: {tok}, Alt: {alt}, Log Prob: {prob:.4f}, Position: {i+start_idx}" for i, (tok, alts) in enumerate(zip(paginated_tokens, paginated_alternatives)) for alt, prob in alts],
|
275 |
hovertemplate='<b>%{customdata}</b><extra></extra>'
|
276 |
)
|
277 |
|
278 |
+
return (main_fig, df, colored_text_html, alt_viz_fig, drops_fig, total_chunks, chunk)
|
279 |
|
280 |
except Exception as e:
|
281 |
logger.error("Visualization failed: %s (Input: %s)", str(e), json_input[:100] + "..." if len(json_input) > 100 else json_input)
|
282 |
+
return (create_empty_figure("Log Probabilities of Generated Tokens"), None, "No finite log probabilities to display.", create_empty_figure("Top Token Log Probabilities"), create_empty_figure("Significant Probability Drops"), 1, 0)
|
283 |
|
284 |
+
# Gradio interface with chunked visualization and proactive precomputation
|
285 |
with gr.Blocks(title="Log Probability Visualizer") as app:
|
286 |
gr.Markdown("# Log Probability Visualizer")
|
287 |
gr.Markdown(
|
288 |
+
"Paste your JSON log prob data below to visualize tokens in chunks of 1,000. Fixed filter ≥ -100000, dynamic number of top_logprobs, handles missing or null fields. Next chunk is precomputed proactively."
|
289 |
)
|
290 |
|
291 |
with gr.Row():
|
|
|
294 |
lines=10,
|
295 |
placeholder="Paste your JSON (e.g., {\"content\": [{\"bytes\": [44], \"logprob\": 0.0, \"token\": \",\", \"top_logprobs\": {\" so\": -13.8046875, \".\": -13.8046875, \",\": -13.640625}}]}).",
|
296 |
)
|
297 |
+
chunk = gr.Number(value=0, label="Current Chunk", precision=0, minimum=0)
|
298 |
|
299 |
with gr.Row():
|
300 |
plot_output = gr.Plot(label="Log Probability Plot (Click for Tokens)")
|
|
|
307 |
with gr.Row():
|
308 |
text_output = gr.HTML(label="Colored Text (Confidence Visualization)")
|
309 |
|
310 |
+
with gr.Row():
|
311 |
+
prev_btn = gr.Button("Previous Chunk")
|
312 |
+
next_btn = gr.Button("Next Chunk")
|
313 |
+
total_chunks_output = gr.Number(label="Total Chunks", interactive=False)
|
314 |
+
|
315 |
+
# Precomputed next chunk state (hidden)
|
316 |
+
precomputed_next = gr.State(value=None)
|
317 |
+
|
318 |
btn = gr.Button("Visualize")
|
319 |
btn.click(
|
320 |
fn=visualize_logprobs,
|
321 |
+
inputs=[json_input, chunk],
|
322 |
+
outputs=[plot_output, table_output, text_output, alt_viz_output, drops_output, total_chunks_output, chunk],
|
323 |
+
)
|
324 |
+
|
325 |
+
# Precompute next chunk proactively when on current chunk
|
326 |
+
async def precompute_next_chunk(json_input, current_chunk, precomputed_next):
|
327 |
+
if precomputed_next is not None:
|
328 |
+
return precomputed_next # Use cached precomputed chunk if available
|
329 |
+
next_tokens, next_logprobs, next_alternatives = await precompute_chunk(json_input, 1000, current_chunk)
|
330 |
+
if next_tokens is None or next_logprobs is None or next_alternatives is None:
|
331 |
+
return None
|
332 |
+
return (next_tokens, next_logprobs, next_alternatives)
|
333 |
+
|
334 |
+
# Update chunk on button clicks
|
335 |
+
def update_chunk(json_input, current_chunk, action, precomputed_next=None):
|
336 |
+
total_chunks = visualize_logprobs(json_input, 0)[5] # Get total chunks
|
337 |
+
if action == "prev" and current_chunk > 0:
|
338 |
+
current_chunk -= 1
|
339 |
+
elif action == "next" and current_chunk < total_chunks - 1:
|
340 |
+
current_chunk += 1
|
341 |
+
# If precomputed next chunk exists, use it; otherwise, compute it
|
342 |
+
if precomputed_next:
|
343 |
+
next_tokens, next_logprobs, next_alternatives = precomputed_next
|
344 |
+
if next_tokens and next_logprobs and next_alternatives:
|
345 |
+
logger.debug("Using precomputed next chunk for chunk %d", current_chunk)
|
346 |
+
return visualize_logprobs(json_input, current_chunk)
|
347 |
+
return visualize_logprobs(json_input, current_chunk)
|
348 |
+
|
349 |
+
prev_btn.click(
|
350 |
+
fn=update_chunk,
|
351 |
+
inputs=[json_input, chunk, gr.State(value="prev"), precomputed_next],
|
352 |
+
outputs=[plot_output, table_output, text_output, alt_viz_output, drops_output, total_chunks_output, chunk],
|
353 |
+
)
|
354 |
+
|
355 |
+
next_btn.click(
|
356 |
+
fn=update_chunk,
|
357 |
+
inputs=[json_input, chunk, gr.State(value="next"), precomputed_next],
|
358 |
+
outputs=[plot_output, table_output, text_output, alt_viz_output, drops_output, total_chunks_output, chunk],
|
359 |
+
)
|
360 |
+
|
361 |
+
# Trigger precomputation when chunk changes (via button clicks or initial load)
|
362 |
+
def trigger_precomputation(json_input, current_chunk):
|
363 |
+
asyncio.create_task(precompute_next_chunk(json_input, current_chunk, None))
|
364 |
+
return gr.update(value=current_chunk)
|
365 |
+
|
366 |
+
# Use a dummy event to trigger precomputation on chunk change (simplified for Gradio)
|
367 |
+
chunk.change(
|
368 |
+
fn=trigger_precomputation,
|
369 |
+
inputs=[json_input, chunk],
|
370 |
+
outputs=[chunk],
|
371 |
)
|
372 |
|
373 |
app.launch()
|