Spaces:
Running
Running
jedick
commited on
Commit
·
8d79272
1
Parent(s):
3c02ce2
Make parent span for app runs
Browse files
app.py
CHANGED
|
@@ -10,20 +10,37 @@ from wiki_data_fetcher import (
|
|
| 10 |
from models import classifier, judge
|
| 11 |
import logfire
|
| 12 |
from dotenv import load_dotenv
|
|
|
|
| 13 |
|
| 14 |
# Load API keys
|
| 15 |
load_dotenv()
|
| 16 |
# Setup logging with Logfire
|
| 17 |
logfire.configure()
|
| 18 |
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
|
| 24 |
|
| 25 |
@logfire.instrument("Step 1: Fetch current revision")
|
| 26 |
-
def
|
| 27 |
"""
|
| 28 |
Fetch current revision of a Wikipedia article and return its introduction.
|
| 29 |
|
|
@@ -69,15 +86,22 @@ def fetch_current_revision(title: str):
|
|
| 69 |
return None, None
|
| 70 |
|
| 71 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 72 |
@logfire.instrument("Step 2: Fetch previous revision")
|
| 73 |
-
def
|
| 74 |
"""
|
| 75 |
Fetch previous revision of a Wikipedia article and return its introduction.
|
| 76 |
|
| 77 |
Args:
|
| 78 |
title: Wikipedia article title
|
| 79 |
-
unit: "revisions" or "days"
|
| 80 |
number: Number of revisions or days behind
|
|
|
|
| 81 |
|
| 82 |
Returns:
|
| 83 |
Tuple of (introduction, timestamp)
|
|
@@ -89,15 +113,15 @@ def fetch_previous_revision(title: str, unit: str, number: int, new_revision: st
|
|
| 89 |
return None, None
|
| 90 |
|
| 91 |
try:
|
| 92 |
-
# Get previous revision based on
|
| 93 |
-
if
|
| 94 |
json_data = get_previous_revisions(title, revisions=number)
|
| 95 |
revision_info = extract_revision_info(json_data, revnum=number)
|
| 96 |
-
else: #
|
| 97 |
revision_info = get_revision_from_age(title, age_days=number)
|
| 98 |
|
| 99 |
if not revision_info.get("revid"):
|
| 100 |
-
error_msg = f"Error: Could not find revision {number} {'revisions' if
|
| 101 |
raise gr.Error(error_msg, print_exception=False)
|
| 102 |
return None, None
|
| 103 |
|
|
@@ -111,7 +135,7 @@ def fetch_previous_revision(title: str, unit: str, number: int, new_revision: st
|
|
| 111 |
introduction = f"Error: Could not retrieve introduction for previous revision (revid: {revid})"
|
| 112 |
|
| 113 |
# Get revisions_behind
|
| 114 |
-
if
|
| 115 |
revisions_behind = revision_info["revnum"]
|
| 116 |
else:
|
| 117 |
revisions_behind = get_revisions_behind(title, revid)
|
|
@@ -170,13 +194,23 @@ def run_classifier(old_revision: str, new_revision: str, prompt_style: str):
|
|
| 170 |
return noteworthy, rationale
|
| 171 |
|
| 172 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 173 |
@logfire.instrument("Step 3a: Run heuristic classifier")
|
| 174 |
-
def
|
| 175 |
return run_classifier(old_revision, new_revision, prompt_style="heuristic")
|
| 176 |
|
| 177 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 178 |
@logfire.instrument("Step 3b: Run few-shot classifier")
|
| 179 |
-
def
|
| 180 |
return run_classifier(old_revision, new_revision, prompt_style="few-shot")
|
| 181 |
|
| 182 |
|
|
@@ -205,7 +239,6 @@ def compute_confidence(
|
|
| 205 |
return "Questionable"
|
| 206 |
|
| 207 |
|
| 208 |
-
@logfire.instrument("Step 4: Run judge")
|
| 209 |
def run_judge(
|
| 210 |
old_revision: str,
|
| 211 |
new_revision: str,
|
|
@@ -214,6 +247,29 @@ def run_judge(
|
|
| 214 |
heuristic_rationale: str,
|
| 215 |
fewshot_rationale: str,
|
| 216 |
judge_mode: str,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 217 |
):
|
| 218 |
"""
|
| 219 |
Run judge on the revisions and classifiers' rationales.
|
|
@@ -295,7 +351,7 @@ with gr.Blocks(title="Noteworthy Differences") as demo:
|
|
| 295 |
label="Wikipedia Page Title", placeholder="e.g., Albert Einstein", value=""
|
| 296 |
)
|
| 297 |
number_input = gr.Number(label="Number", value=50, minimum=0, precision=0)
|
| 298 |
-
|
| 299 |
choices=["revisions", "days"], value="revisions", label="Unit"
|
| 300 |
)
|
| 301 |
judge_mode_dropdown = gr.Dropdown(
|
|
@@ -367,6 +423,8 @@ with gr.Blocks(title="Noteworthy Differences") as demo:
|
|
| 367 |
heuristic_noteworthy = gr.State()
|
| 368 |
fewshot_noteworthy = gr.State()
|
| 369 |
judge_noteworthy = gr.State()
|
|
|
|
|
|
|
| 370 |
|
| 371 |
random_btn.click(
|
| 372 |
fn=get_random_wikipedia_title,
|
|
@@ -383,24 +441,29 @@ with gr.Blocks(title="Noteworthy Differences") as demo:
|
|
| 383 |
inputs=None,
|
| 384 |
outputs=[new_revision, new_timestamp],
|
| 385 |
api_name=False,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 386 |
).then(
|
| 387 |
fn=fetch_current_revision,
|
| 388 |
-
inputs=[title_input],
|
| 389 |
outputs=[new_revision, new_timestamp],
|
| 390 |
api_name=False,
|
| 391 |
).then(
|
| 392 |
fn=fetch_previous_revision,
|
| 393 |
-
inputs=[title_input,
|
| 394 |
outputs=[old_revision, old_timestamp],
|
| 395 |
api_name=False,
|
| 396 |
).then(
|
| 397 |
fn=run_heuristic_classifier,
|
| 398 |
-
inputs=[old_revision, new_revision],
|
| 399 |
outputs=[heuristic_noteworthy, heuristic_rationale],
|
| 400 |
api_name=False,
|
| 401 |
).then(
|
| 402 |
fn=run_fewshot_classifier,
|
| 403 |
-
inputs=[old_revision, new_revision],
|
| 404 |
outputs=[fewshot_noteworthy, fewshot_rationale],
|
| 405 |
api_name=False,
|
| 406 |
).then(
|
|
@@ -413,6 +476,7 @@ with gr.Blocks(title="Noteworthy Differences") as demo:
|
|
| 413 |
heuristic_rationale,
|
| 414 |
fewshot_rationale,
|
| 415 |
judge_mode_dropdown,
|
|
|
|
| 416 |
],
|
| 417 |
outputs=[judge_noteworthy, noteworthy_text, judge_reasoning, confidence],
|
| 418 |
api_name=False,
|
|
@@ -422,12 +486,12 @@ with gr.Blocks(title="Noteworthy Differences") as demo:
|
|
| 422 |
gr.on(
|
| 423 |
triggers=[rerun_btn.click],
|
| 424 |
fn=run_heuristic_classifier,
|
| 425 |
-
inputs=[old_revision, new_revision],
|
| 426 |
outputs=[heuristic_noteworthy, heuristic_rationale],
|
| 427 |
api_name=False,
|
| 428 |
).then(
|
| 429 |
fn=run_fewshot_classifier,
|
| 430 |
-
inputs=[old_revision, new_revision],
|
| 431 |
outputs=[fewshot_noteworthy, fewshot_rationale],
|
| 432 |
api_name=False,
|
| 433 |
).then(
|
|
@@ -440,6 +504,7 @@ with gr.Blocks(title="Noteworthy Differences") as demo:
|
|
| 440 |
heuristic_rationale,
|
| 441 |
fewshot_rationale,
|
| 442 |
judge_mode_dropdown,
|
|
|
|
| 443 |
],
|
| 444 |
outputs=[judge_noteworthy, noteworthy_text, judge_reasoning, confidence],
|
| 445 |
api_name=False,
|
|
|
|
| 10 |
from models import classifier, judge
|
| 11 |
import logfire
|
| 12 |
from dotenv import load_dotenv
|
| 13 |
+
from contextlib import nullcontext
|
| 14 |
|
| 15 |
# Load API keys
|
| 16 |
load_dotenv()
|
| 17 |
# Setup logging with Logfire
|
| 18 |
logfire.configure()
|
| 19 |
|
| 20 |
+
|
| 21 |
+
def start_parent_span(title: str, number: int, units: str):
|
| 22 |
+
"""
|
| 23 |
+
Start a parent span and return the context for propagation to children.
|
| 24 |
+
See https://logfire.pydantic.dev/docs/how-to-guides/distributed-tracing/#manual-context-propagation
|
| 25 |
+
"""
|
| 26 |
+
span_name = f"{title} - {number} {units}"
|
| 27 |
+
with logfire.span(span_name) as span:
|
| 28 |
+
span.__enter__()
|
| 29 |
+
context = logfire.get_context()
|
| 30 |
+
return context
|
| 31 |
+
|
| 32 |
+
|
| 33 |
+
def fetch_current_revision(title: str, context=None):
|
| 34 |
+
"""
|
| 35 |
+
Wrapper to run _fetch_current_revision in provided Logfire context.
|
| 36 |
+
We use this to minimize indentation in the wrapped function.
|
| 37 |
+
"""
|
| 38 |
+
with logfire.attach_context(context) if context else nullcontext():
|
| 39 |
+
return _fetch_current_revision(title)
|
| 40 |
|
| 41 |
|
| 42 |
@logfire.instrument("Step 1: Fetch current revision")
|
| 43 |
+
def _fetch_current_revision(title: str):
|
| 44 |
"""
|
| 45 |
Fetch current revision of a Wikipedia article and return its introduction.
|
| 46 |
|
|
|
|
| 86 |
return None, None
|
| 87 |
|
| 88 |
|
| 89 |
+
def fetch_previous_revision(
|
| 90 |
+
title: str, number: int, units: str, new_revision: str, context=None
|
| 91 |
+
):
|
| 92 |
+
with logfire.attach_context(context) if context else nullcontext():
|
| 93 |
+
return _fetch_previous_revision(title, number, units, new_revision)
|
| 94 |
+
|
| 95 |
+
|
| 96 |
@logfire.instrument("Step 2: Fetch previous revision")
|
| 97 |
+
def _fetch_previous_revision(title: str, number: int, units: str, new_revision: str):
|
| 98 |
"""
|
| 99 |
Fetch previous revision of a Wikipedia article and return its introduction.
|
| 100 |
|
| 101 |
Args:
|
| 102 |
title: Wikipedia article title
|
|
|
|
| 103 |
number: Number of revisions or days behind
|
| 104 |
+
units: "revisions" or "days"
|
| 105 |
|
| 106 |
Returns:
|
| 107 |
Tuple of (introduction, timestamp)
|
|
|
|
| 113 |
return None, None
|
| 114 |
|
| 115 |
try:
|
| 116 |
+
# Get previous revision based on units
|
| 117 |
+
if units == "revisions":
|
| 118 |
json_data = get_previous_revisions(title, revisions=number)
|
| 119 |
revision_info = extract_revision_info(json_data, revnum=number)
|
| 120 |
+
else: # units == "days"
|
| 121 |
revision_info = get_revision_from_age(title, age_days=number)
|
| 122 |
|
| 123 |
if not revision_info.get("revid"):
|
| 124 |
+
error_msg = f"Error: Could not find revision {number} {'revisions' if units == 'revisions' else 'days'} behind for '{title}'."
|
| 125 |
raise gr.Error(error_msg, print_exception=False)
|
| 126 |
return None, None
|
| 127 |
|
|
|
|
| 135 |
introduction = f"Error: Could not retrieve introduction for previous revision (revid: {revid})"
|
| 136 |
|
| 137 |
# Get revisions_behind
|
| 138 |
+
if units == "revisions":
|
| 139 |
revisions_behind = revision_info["revnum"]
|
| 140 |
else:
|
| 141 |
revisions_behind = get_revisions_behind(title, revid)
|
|
|
|
| 194 |
return noteworthy, rationale
|
| 195 |
|
| 196 |
|
| 197 |
+
def run_heuristic_classifier(old_revision: str, new_revision: str, context=None):
|
| 198 |
+
with logfire.attach_context(context) if context else nullcontext():
|
| 199 |
+
return _run_heuristic_classifier(old_revision, new_revision)
|
| 200 |
+
|
| 201 |
+
|
| 202 |
@logfire.instrument("Step 3a: Run heuristic classifier")
|
| 203 |
+
def _run_heuristic_classifier(old_revision: str, new_revision: str):
|
| 204 |
return run_classifier(old_revision, new_revision, prompt_style="heuristic")
|
| 205 |
|
| 206 |
|
| 207 |
+
def run_fewshot_classifier(old_revision: str, new_revision: str, context=None):
|
| 208 |
+
with logfire.attach_context(context) if context else nullcontext():
|
| 209 |
+
return _run_fewshot_classifier(old_revision, new_revision)
|
| 210 |
+
|
| 211 |
+
|
| 212 |
@logfire.instrument("Step 3b: Run few-shot classifier")
|
| 213 |
+
def _run_fewshot_classifier(old_revision: str, new_revision: str):
|
| 214 |
return run_classifier(old_revision, new_revision, prompt_style="few-shot")
|
| 215 |
|
| 216 |
|
|
|
|
| 239 |
return "Questionable"
|
| 240 |
|
| 241 |
|
|
|
|
| 242 |
def run_judge(
|
| 243 |
old_revision: str,
|
| 244 |
new_revision: str,
|
|
|
|
| 247 |
heuristic_rationale: str,
|
| 248 |
fewshot_rationale: str,
|
| 249 |
judge_mode: str,
|
| 250 |
+
context=None,
|
| 251 |
+
):
|
| 252 |
+
with logfire.attach_context(context) if context else nullcontext():
|
| 253 |
+
return _run_judge(
|
| 254 |
+
old_revision,
|
| 255 |
+
new_revision,
|
| 256 |
+
heuristic_noteworthy,
|
| 257 |
+
heuristic_noteworthy,
|
| 258 |
+
heuristic_rationale,
|
| 259 |
+
fewshot_rationale,
|
| 260 |
+
judge_mode,
|
| 261 |
+
)
|
| 262 |
+
|
| 263 |
+
|
| 264 |
+
@logfire.instrument("Step 4: Run judge")
|
| 265 |
+
def _run_judge(
|
| 266 |
+
old_revision: str,
|
| 267 |
+
new_revision: str,
|
| 268 |
+
heuristic_noteworthy: bool,
|
| 269 |
+
fewshot_noteworthy: bool,
|
| 270 |
+
heuristic_rationale: str,
|
| 271 |
+
fewshot_rationale: str,
|
| 272 |
+
judge_mode: str,
|
| 273 |
):
|
| 274 |
"""
|
| 275 |
Run judge on the revisions and classifiers' rationales.
|
|
|
|
| 351 |
label="Wikipedia Page Title", placeholder="e.g., Albert Einstein", value=""
|
| 352 |
)
|
| 353 |
number_input = gr.Number(label="Number", value=50, minimum=0, precision=0)
|
| 354 |
+
units_dropdown = gr.Dropdown(
|
| 355 |
choices=["revisions", "days"], value="revisions", label="Unit"
|
| 356 |
)
|
| 357 |
judge_mode_dropdown = gr.Dropdown(
|
|
|
|
| 423 |
heuristic_noteworthy = gr.State()
|
| 424 |
fewshot_noteworthy = gr.State()
|
| 425 |
judge_noteworthy = gr.State()
|
| 426 |
+
# State to store Logfire context
|
| 427 |
+
context = gr.State()
|
| 428 |
|
| 429 |
random_btn.click(
|
| 430 |
fn=get_random_wikipedia_title,
|
|
|
|
| 441 |
inputs=None,
|
| 442 |
outputs=[new_revision, new_timestamp],
|
| 443 |
api_name=False,
|
| 444 |
+
).then(
|
| 445 |
+
# Initialize Logfire context
|
| 446 |
+
fn=start_parent_span,
|
| 447 |
+
inputs=[title_input, number_input, units_dropdown],
|
| 448 |
+
outputs=context,
|
| 449 |
).then(
|
| 450 |
fn=fetch_current_revision,
|
| 451 |
+
inputs=[title_input, context],
|
| 452 |
outputs=[new_revision, new_timestamp],
|
| 453 |
api_name=False,
|
| 454 |
).then(
|
| 455 |
fn=fetch_previous_revision,
|
| 456 |
+
inputs=[title_input, number_input, units_dropdown, new_revision, context],
|
| 457 |
outputs=[old_revision, old_timestamp],
|
| 458 |
api_name=False,
|
| 459 |
).then(
|
| 460 |
fn=run_heuristic_classifier,
|
| 461 |
+
inputs=[old_revision, new_revision, context],
|
| 462 |
outputs=[heuristic_noteworthy, heuristic_rationale],
|
| 463 |
api_name=False,
|
| 464 |
).then(
|
| 465 |
fn=run_fewshot_classifier,
|
| 466 |
+
inputs=[old_revision, new_revision, context],
|
| 467 |
outputs=[fewshot_noteworthy, fewshot_rationale],
|
| 468 |
api_name=False,
|
| 469 |
).then(
|
|
|
|
| 476 |
heuristic_rationale,
|
| 477 |
fewshot_rationale,
|
| 478 |
judge_mode_dropdown,
|
| 479 |
+
context,
|
| 480 |
],
|
| 481 |
outputs=[judge_noteworthy, noteworthy_text, judge_reasoning, confidence],
|
| 482 |
api_name=False,
|
|
|
|
| 486 |
gr.on(
|
| 487 |
triggers=[rerun_btn.click],
|
| 488 |
fn=run_heuristic_classifier,
|
| 489 |
+
inputs=[old_revision, new_revision, context],
|
| 490 |
outputs=[heuristic_noteworthy, heuristic_rationale],
|
| 491 |
api_name=False,
|
| 492 |
).then(
|
| 493 |
fn=run_fewshot_classifier,
|
| 494 |
+
inputs=[old_revision, new_revision, context],
|
| 495 |
outputs=[fewshot_noteworthy, fewshot_rationale],
|
| 496 |
api_name=False,
|
| 497 |
).then(
|
|
|
|
| 504 |
heuristic_rationale,
|
| 505 |
fewshot_rationale,
|
| 506 |
judge_mode_dropdown,
|
| 507 |
+
context,
|
| 508 |
],
|
| 509 |
outputs=[judge_noteworthy, noteworthy_text, judge_reasoning, confidence],
|
| 510 |
api_name=False,
|