Spaces:
Sleeping
Sleeping
File size: 39,048 Bytes
17df322 eb5ef66 17df322 eb5ef66 9cbcd4d eb5ef66 9cbcd4d eb5ef66 9cbcd4d eb5ef66 9cbcd4d eb5ef66 9cbcd4d eb5ef66 9cbcd4d eb5ef66 9cbcd4d eb5ef66 9cbcd4d eb5ef66 9cbcd4d eb5ef66 9cbcd4d eb5ef66 9cbcd4d eb5ef66 9cbcd4d eb5ef66 17df322 eb5ef66 17df322 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 |
import gradio as gr
from huggingface_hub import InferenceClient
#!/usr/bin/env python3
# import gradio as gr
import json
import logging
import os
import traceback
from pathlib import Path
from urllib.parse import urlparse
from typing import Dict, Any, List, Set
from git import Repo
import io
import torch
import numpy as np
import faiss
from transformers import AutoTokenizer, AutoModelForCausalLM
from sentence_transformers import SentenceTransformer, util
from huggingface_hub import snapshot_download
import os
from openai import AzureOpenAI
import requests
import re
import matplotlib.pyplot as plt
from sklearn.manifold import TSNE
from sklearn.cluster import KMeans
import plotly.graph_objects as go
import plotly.express as px
import random
from sklearn.cluster import AgglomerativeClustering
def load_env():
from dotenv import load_dotenv
env_path = Path(__file__).parent.parent / '.env'
load_dotenv(dotenv_path=env_path)
load_env()
# Centralized env parameters
HUGGINGFACE_HUB_TOKEN = os.getenv("HUGGINGFACE_HUB_TOKEN")
GITHUB_TOKEN = os.getenv("GITHUB_TOKEN")
AZURE_OPENAI_ENDPOINT = os.getenv("AZURE_OPENAI_ENDPOINT")
AZURE_OPENAI_API_KEY = os.getenv("AZURE_OPENAI_API_KEY")
MODEL_NAME = "gpt-4o-mini"
DEPLOYMENT = "gpt-4o-mini"
API_VERSION = "2024-12-01-preview"
FILE_REGEX = re.compile(r"^diff --git a/(.+?) b/(.+)")
LINE_HUNK = re.compile(r"@@ -(?P<old_start>\d+),(?P<old_len>\d+) \+(?P<new_start>\d+),(?P<new_len>\d+) @@")
# Configure logging to capture all output
log_stream = io.StringIO()
log_handler = logging.StreamHandler(log_stream)
log_handler.setLevel(logging.INFO)
log_formatter = logging.Formatter("%(asctime)s %(levelname)s %(message)s")
log_handler.setFormatter(log_formatter)
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s %(levelname)s %(message)s",
handlers=[log_handler, logging.StreamHandler()]
)
logger = logging.getLogger(__name__)
class InferenceContext:
def __init__(self, repo_url: str):
self.repo_url = repo_url
owner, name = self._parse_owner_repo(repo_url)
self.repo_id = f"{owner}/{name}"
self.repo_dir = f"{owner}-{name}"
self.hf_repo_id = "kotlarmilos/repository-learning"
# Local paths for downloaded models
self.base = Path("artifacts") / self.repo_dir
self.model_dirs = {
'fine_tune': self.base / 'fine_tune',
'contrastive': self.base / 'contrastive',
'index': self.base / 'index'
}
self.code_dir = self.base / 'code'
# Create directories
for d in (*self.model_dirs.values(), self.code_dir):
d.mkdir(parents=True, exist_ok=True)
@staticmethod
def _parse_owner_repo(url: str) -> tuple[str, str]:
parts = urlparse(url).path.strip("/").split("/")
if len(parts) < 2:
raise ValueError(f"Invalid GitHub URL: {url}")
return parts[-2], parts[-1]
class InferencePipeline:
def __init__(self, ctx: InferenceContext):
self.ctx = ctx
self.tokenizer = None
self.llm = None
self.embedder = None
self.faiss_index = None
self.faiss_metadata = None
self.download_artifacts()
self.load_models()
def download_artifacts(self):
"""Download models and index from Hugging Face if they don't exist locally."""
self.repo_files = self._clone_or_pull()
snapshot_download(
repo_id=self.ctx.hf_repo_id,
allow_patterns=f"{self.ctx.repo_dir}/**",
local_dir=str(self.ctx.base.parent),
local_dir_use_symlinks=False,
token=HUGGINGFACE_HUB_TOKEN
)
logger.info("All artifacts download complete.")
def _clone_or_pull(self) -> bool:
dest = self.ctx.code_dir
git_dir = dest / ".git"
if git_dir.exists():
Repo(dest).remotes.origin.pull()
logger.info("Pulled latest code into %s", dest)
else:
Repo.clone_from(self.ctx.repo_url, dest)
logger.info("Cloned repo %s into %s", self.ctx.repo_url, dest)
return [str(f.relative_to(dest)) for f in dest.rglob("*") if f.is_file()]
def load_models(self):
"""Load the fine-tuned LLM model."""
self.tokenizer = AutoTokenizer.from_pretrained(self.ctx.model_dirs['fine_tune'])
self.local_llm = AutoModelForCausalLM.from_pretrained(
self.ctx.model_dirs['fine_tune'],
device_map="auto",
torch_dtype=torch.bfloat16
)
self.enterprise_llm = AzureOpenAI(
api_version=API_VERSION,
azure_endpoint=AZURE_OPENAI_ENDPOINT,
api_key=AZURE_OPENAI_API_KEY,
)
self.embedder = SentenceTransformer(str(self.ctx.model_dirs['contrastive']))
self.faiss_index = faiss.read_index(str(self.ctx.model_dirs['index'] / "index.faiss"))
self.faiss_metadata = json.loads((self.ctx.model_dirs['index'] / "metadata.json").read_text())
logger.info("FAISS index loaded successfully")
def _extract_pr_data(self, pr_url: str) -> dict:
"""
Collect PR data using GitHub API.
"""
match = re.search(r'/pull/(\d+)', pr_url)
pr_number = int(match.group(1))
pr_url = f"https://api.github.com/repos/{self.ctx.repo_id}/pulls/{pr_number}"
comments_url = f"https://api.github.com/repos/{self.ctx.repo_id}/pulls/{pr_number}/comments"
headers = {}
headers["Authorization"] = f"token {GITHUB_TOKEN}"
headers["Accept"] = "application/vnd.github.v3+json"
try:
logger.info(f"Fetching PR #{pr_number} details...")
pr_response = requests.get(pr_url, headers=headers)
pr_response.raise_for_status()
pr_data = pr_response.json()
logger.info(f"Fetching PR #{pr_number} review comments...")
comments_response = requests.get(comments_url, headers=headers)
comments_response.raise_for_status()
comments_data = comments_response.json()
grouped = {}
for comment in comments_data:
hunk = comment.get("diff_hunk", "")
grouped.setdefault(hunk, []).append(comment.get("body", ""))
review_comments = [
{"diff_hunk": hunk, "comments": comments}
for hunk, comments in grouped.items()
]
logger.info(f"Fetching PR #{pr_number} diff...")
diff_headers = headers.copy()
diff_headers["Accept"] = "application/vnd.github.v3.diff"
diff_response = requests.get(pr_url, headers=diff_headers)
diff_response.raise_for_status()
parsed_diff = self.parse_diff_with_lines(diff_response.text)
result = {
"title": pr_data.get("title", ""),
"body": pr_data.get("body", ""),
"review_comments": review_comments,
"diff": diff_response.text,
"changed_files": list(parsed_diff['changed_files']),
"diff_hunks": parsed_diff['diff_hunks']
}
logger.info(f"Successfully collected PR #{pr_number} data")
return result
except Exception as e:
logger.error(f"Error processing PR #{pr_number} data: {e}")
raise
def parse_diff_with_lines(self, diff_text: str) -> Dict[str, Any]:
lines = diff_text.splitlines()
result = {
'changed_files': set(),
'diff_hunks': {}
}
current_file = None
current_hunk_content = []
current_line_range = None
file_header_lines = []
for line in lines:
# Check if this is a new file header
file_match = FILE_REGEX.match(line)
if file_match:
# Save previous file data if exists
if current_file and current_hunk_content and current_line_range:
if current_file not in result['diff_hunks']:
result['diff_hunks'][current_file] = []
result['diff_hunks'][current_file].append({
'line_range': current_line_range,
'content': '\n'.join(current_hunk_content)
})
# Start new file
current_file = file_match.group(2) # Use the 'b/' file path (new file)
result['changed_files'].add(current_file)
file_header_lines = [line]
current_hunk_content = []
current_line_range = None
elif current_file: # Only process if we're inside a file
# Check for hunk headers to extract line ranges
hunk_match = LINE_HUNK.match(line)
if hunk_match:
# Save previous hunk if exists
if current_hunk_content and current_line_range:
if current_file not in result['diff_hunks']:
result['diff_hunks'][current_file] = []
result['diff_hunks'][current_file].append({
'line_range': current_line_range,
'content': '\n'.join(current_hunk_content)
})
# Start new hunk
old_start = int(hunk_match.group('old_start'))
old_len = int(hunk_match.group('old_len'))
new_start = int(hunk_match.group('new_start'))
new_len = int(hunk_match.group('new_len'))
# Calculate the range of changed lines
if new_len > 0:
line_start = new_start
line_end = new_start + new_len - 1
current_line_range = (line_start, line_end)
else:
current_line_range = (new_start, new_start)
# Start fresh hunk content with file headers and current hunk header
current_hunk_content = file_header_lines + [line]
else:
# Add content line to current hunk
if current_hunk_content is not None:
current_hunk_content.append(line)
# Save the last hunk data
if current_file and current_hunk_content and current_line_range:
if current_file not in result['diff_hunks']:
result['diff_hunks'][current_file] = []
result['diff_hunks'][current_file].append({
'line_range': current_line_range,
'content': '\n'.join(current_hunk_content)
})
return result
def analyze_file_similarity(self, changed_files: List[str]) -> Dict[str, Any]:
result = {
'similar_file_groups': [],
'anomalous_files': [],
'analysis_summary': {
'total_files': len(changed_files),
'num_groups': 0,
'num_anomalies': 0,
'avg_group_size': 0
}
}
# Handle edge cases
if len(changed_files) == 0:
logger.info("No changed files to analyze")
return result
if len(changed_files) == 1:
logger.info(f"Only one file changed: {changed_files[0]} - no similarity analysis needed")
result['analysis_summary']['num_anomalies'] = 1
result['anomalous_files'].append({
'file': changed_files[0],
'reason': 'single_file',
'max_similarity_to_others': 0.0,
'most_similar_file': None,
'is_anomaly': False
})
return result
# Encode all changed files
file_embeddings = self.embedder.encode(changed_files, convert_to_tensor=True)
similarity_matrix = util.pytorch_cos_sim(file_embeddings, file_embeddings)
# Convert similarity matrix to distance matrix for clustering
distance_matrix = 1 - similarity_matrix.cpu().numpy()
# Perform hierarchical clustering
clustering = AgglomerativeClustering(
n_clusters=None,
distance_threshold=0.3, # 1 - 0.7 = 0.3 (similarity threshold of 0.7)
metric='precomputed',
linkage='average'
)
cluster_labels = clustering.fit_predict(distance_matrix)
# Group files by cluster
clusters = {}
for i, label in enumerate(cluster_labels):
if label not in clusters:
clusters[label] = []
clusters[label].append((changed_files[i], i)) # Store file and its index
# Process clusters to identify groups and anomalies
for cluster_id, files_with_indices in clusters.items():
files_in_cluster = [f[0] for f in files_with_indices]
if len(files_in_cluster) > 1:
# This is a group of similar files
group_similarities = []
pairwise_similarities = []
for i in range(len(files_with_indices)):
for j in range(i+1, len(files_with_indices)):
file_i_idx = files_with_indices[i][1]
file_j_idx = files_with_indices[j][1]
similarity = float(similarity_matrix[file_i_idx][file_j_idx])
group_similarities.append(similarity)
pairwise_similarities.append({
'file1': files_with_indices[i][0],
'file2': files_with_indices[j][0],
'similarity': similarity
})
avg_similarity = sum(group_similarities) / len(group_similarities) if group_similarities else 0
min_similarity = min(group_similarities) if group_similarities else 0
max_similarity = max(group_similarities) if group_similarities else 0
result['similar_file_groups'].append({
'cluster_id': cluster_id,
'files': files_in_cluster,
'avg_similarity': avg_similarity,
'min_similarity': min_similarity,
'max_similarity': max_similarity,
'pairwise_similarities': pairwise_similarities,
'coherence': 'high' if min_similarity > 0.6 else 'medium' if min_similarity > 0.4 else 'low'
})
else:
# This is a singleton cluster - potentially anomalous
file = files_in_cluster[0]
file_idx = files_with_indices[0][1]
# Calculate maximum similarity to any other file
max_similarity = 0
most_similar_file = None
similarities_to_others = []
for other_idx, other_file in enumerate(changed_files):
if other_idx != file_idx:
similarity = float(similarity_matrix[file_idx][other_idx])
similarities_to_others.append({
'file': other_file,
'similarity': similarity
})
if similarity > max_similarity:
max_similarity = similarity
most_similar_file = other_file
result['anomalous_files'].append({
'file': file,
'cluster_id': cluster_id,
'max_similarity_to_others': max_similarity,
'most_similar_file': most_similar_file,
'similarities_to_others': similarities_to_others,
'is_anomaly': max_similarity < 0.5, # Strong anomaly threshold
'anomaly_strength': 'strong' if max_similarity < 0.3 else 'medium' if max_similarity < 0.5 else 'weak',
'reason': 'isolated_cluster'
})
# Additional anomaly detection: files that are far from the group average
if len(changed_files) >= 3:
# Calculate average embedding of all changed files
avg_embedding = torch.mean(file_embeddings, dim=0)
# Find files that are far from the average
for i, file in enumerate(changed_files):
file_embedding = file_embeddings[i]
similarity_to_avg = float(util.pytorch_cos_sim(file_embedding.unsqueeze(0), avg_embedding.unsqueeze(0))[0][0])
# Check if this file is already in anomalous_files
existing_anomaly = next((a for a in result['anomalous_files'] if a['file'] == file), None)
if existing_anomaly:
# Update existing anomaly record
existing_anomaly['similarity_to_group_avg'] = similarity_to_avg
existing_anomaly['is_strong_anomaly'] = (
similarity_to_avg < 0.4 and existing_anomaly['max_similarity_to_others'] < 0.5
)
if existing_anomaly['is_strong_anomaly']:
existing_anomaly['anomaly_strength'] = 'very_strong'
elif similarity_to_avg < 0.4: # Low similarity to group average
# Calculate similarities to all other files
similarities_to_others = []
max_sim = 0
most_sim_file = None
for j, other_file in enumerate(changed_files):
if i != j:
sim = float(similarity_matrix[i][j])
similarities_to_others.append({
'file': other_file,
'similarity': sim
})
if sim > max_sim:
max_sim = sim
most_sim_file = other_file
result['anomalous_files'].append({
'file': file,
'cluster_id': None,
'max_similarity_to_others': max_sim,
'most_similar_file': most_sim_file,
'similarities_to_others': similarities_to_others,
'similarity_to_group_avg': similarity_to_avg,
'is_anomaly': True,
'is_strong_anomaly': max_sim < 0.5,
'anomaly_strength': 'very_strong' if max_sim < 0.3 else 'strong' if max_sim < 0.5 else 'medium',
'reason': 'distant_from_group_average'
})
# Update analysis summary
result['analysis_summary']['num_groups'] = len(result['similar_file_groups'])
result['analysis_summary']['num_anomalies'] = len(result['anomalous_files'])
if result['similar_file_groups']:
total_files_in_groups = sum(len(group['files']) for group in result['similar_file_groups'])
result['analysis_summary']['avg_group_size'] = total_files_in_groups / len(result['similar_file_groups'])
# Log results
logger.info(f"File similarity analysis complete:")
logger.info(f" Total files: {result['analysis_summary']['total_files']}")
logger.info(f" Similar groups: {result['analysis_summary']['num_groups']}")
logger.info(f" Anomalous files: {result['analysis_summary']['num_anomalies']}")
for i, group in enumerate(result['similar_file_groups']):
logger.info(f" Group {i+1} ({group['coherence']} coherence): {group['files']} (avg: {group['avg_similarity']:.3f})")
for anomaly in result['anomalous_files']:
logger.info(f" {anomaly['anomaly_strength'].upper()} ANOMALY: {anomaly['file']} (reason: {anomaly['reason']}, max_sim: {anomaly['max_similarity_to_others']:.3f})")
return result
# TODO: Add local LLM reasoning
# def generate_llm_response(self, prompt: str, max_new_tokens: int = 256) -> str:
# """Generate response using the fine-tuned LLM."""
# if not self.tokenizer or not self.local_llm:
# raise ValueError("LLM not loaded. Call load_llm() first.")
# inputs = self.tokenizer(prompt, return_tensors="pt").to(self.local_llm.device)
# outputs = self.local_llm.generate(
# **inputs,
# max_new_tokens=max_new_tokens,
# pad_token_id=self.tokenizer.eos_token_id
# )
# return self.tokenizer.decode(outputs[0], skip_special_tokens=True)
def search_code_snippets(self, diff_hunks) -> list:
metadata_file = self.ctx.model_dirs["index"] / "metadata.json"
with open(metadata_file, 'r', encoding='utf-8') as f:
metadata = json.load(f)
result = []
# Process each file's diff hunks
for file_path, hunks in diff_hunks.items():
logger.info(f"Searching functions for file: {file_path}")
for hunk in hunks:
line_range = hunk.get('line_range')
if not line_range:
continue
start_line, end_line = line_range
logger.debug(f"Processing hunk at lines {start_line}-{end_line}")
# Find functions that overlap with this line range
overlapping_functions = []
for func_metadata in metadata:
func_file = func_metadata.get('file', '')
func_start = func_metadata.get('start_line')
func_end = func_metadata.get('end_line')
func_name = func_metadata.get('name', 'unknown')
func_description = func_metadata.get('llm_description', '')
# Check if this function is in the same file
if func_file != file_path:
continue
# Check if function line range overlaps with diff hunk line range
if func_start is not None and func_end is not None:
# Check for overlap: function overlaps if it starts before diff ends
# and ends after diff starts
if func_start <= end_line and func_end >= start_line:
overlap_start = max(func_start, start_line)
overlap_end = min(func_end, end_line)
overlapping_functions.append({
'function_name': func_name,
'function_description': func_description,
'function_start_line': func_start,
'function_end_line': func_end,
# 'overlap_start': overlap_start,
# 'overlap_end': overlap_end,
# 'overlap_lines': overlap_end - overlap_start + 1
})
# if len(overlapping_functions) > 0:
hunk_result = {
'file_name': file_path,
'diff_hunk': hunk.get('content', ''),
'overlapping_functions': overlapping_functions
}
result.append(hunk_result)
total_hunks = sum(len(hunks) for hunks in diff_hunks.values())
total_functions = sum(len(entry['overlapping_functions']) for entry in result)
logger.info(f"Processed {total_hunks} diff hunks across {len(diff_hunks)} files, found {total_functions} overlapping functions")
return result
def _select_files_around_changed(self, changed_files: List[str] = None, max_files: int = 500) -> List[str]:
"""Select files to visualize, prioritizing changed files and semantically similar ones."""
logger.info(f"Selecting {max_files} files around {len(changed_files)} changed files...")
# Start with changed files
selected_files = set(changed_files)
# Find files similar to changed files using embeddings
try:
# Encode changed files
changed_embeddings = self.embedder.encode(changed_files, convert_to_tensor=False)
# Calculate target number of similar files to find
target_similar = min(max_files - len(changed_files), 200) # Leave room for random files
# Get a sample of repo files to compare against (for performance)
sample_size = min(2000, len(self.repo_files))
repo_sample = self.repo_files[:sample_size]
# Remove already selected files from sample
repo_sample = [f for f in repo_sample if f not in selected_files]
if len(repo_sample) > 0:
# Encode sample files
sample_embeddings = self.embedder.encode(repo_sample, convert_to_tensor=False, show_progress_bar=False)
# Calculate similarities
similarities = []
for i, repo_file in enumerate(repo_sample):
# Calculate max similarity to any changed file
max_sim = 0
for changed_emb in changed_embeddings:
sim = np.dot(changed_emb, sample_embeddings[i]) / (
np.linalg.norm(changed_emb) * np.linalg.norm(sample_embeddings[i])
)
max_sim = max(max_sim, sim)
# Only add if not already selected (avoid duplicates)
similarities.append((repo_file, max_sim))
# Sort by similarity and take top ones, avoiding duplicates
added = 0
for file_path, sim in sorted(similarities, key=lambda x: x[1], reverse=True):
if file_path not in selected_files:
selected_files.add(file_path)
added += 1
if len(selected_files) >= max_files or added >= target_similar:
break
logger.info(f"Added {len(similarities[:target_similar])} similar files to visualization")
except Exception as e:
logger.warning(f"Could not compute file similarities: {e}")
# Fill remaining slots with random files
remaining_slots = max_files - len(selected_files)
if remaining_slots > 0:
remaining_files = [f for f in self.repo_files if f not in selected_files]
random.shuffle(remaining_files)
for file_path in remaining_files[:remaining_slots]:
selected_files.add(file_path)
result = list(selected_files)
logger.info(f"Selected {len(result)} files total: {len(changed_files)} changed, {len(result) - len(changed_files)} related/random")
return result
def create_repo_visualization(self, changed_files: List[str] = None, max_files: int = 500):
files_to_plot = self._select_files_around_changed(changed_files, max_files * len(changed_files))
logger.info(f"Creating visualization for {len(files_to_plot)} files...")
if len(files_to_plot) < 2:
return self._create_dummy_plot(f"Only {len(files_to_plot)} files available")
embeddings = self.embedder.encode(files_to_plot, convert_to_tensor=False, show_progress_bar=False)
logger.info(f"Embeddings computed successfully: shape {getattr(embeddings, 'shape', None)}")
n = len(files_to_plot)
perplexity = min(30, max(1, n - 1))
tsne = TSNE(n_components=3, perplexity=perplexity, init='random', random_state=42)
reduced = tsne.fit_transform(embeddings)
fig = go.Figure()
colors = []
sizes = []
hover_texts = []
for i, file_path in enumerate(files_to_plot):
if changed_files and file_path in changed_files:
colors.append('red')
else:
# Color by file type
ext = os.path.splitext(file_path)[1].lower()
if ext in ['.py', '.js', '.ts', '.java', '.cpp', '.c', '.cs', '.rb', '.go', '.rs']:
colors.append('blue')
elif ext in ['.md', '.txt', '.rst', '.doc']:
colors.append('green')
elif ext in ['.json', '.yaml', '.yml', '.xml', '.toml', '.ini']:
colors.append('orange')
elif ext in ['.html', '.css', '.scss', '.sass']:
colors.append('purple')
else:
colors.append('gray')
sizes.append(8)
hover_texts.append(f"{os.path.basename(file_path)}")
fig.add_trace(go.Scatter3d(
x=reduced[:, 0].tolist(),
y=reduced[:, 1].tolist(),
z=reduced[:, 2].tolist(),
mode='markers+text',
marker=dict(size=sizes, color=colors),
text=[os.path.basename(f) for f in files_to_plot],
hovertext=hover_texts,
textposition='middle center',
name='Repository Files'
))
title_text = 'Repository File Embeddings (3D t-SNE)'
if changed_files:
title_text += f' - {len(changed_files)} Changed Files Highlighted in Red'
fig.update_layout(
title=title_text,
scene=dict(
xaxis_title='t-SNE 1',
yaxis_title='t-SNE 2',
zaxis_title='t-SNE 3',
camera=dict(eye=dict(x=1.5, y=1.5, z=1.5))
),
width=800,
height=600,
margin=dict(r=20, b=10, l=10, t=60)
)
return fig
def build_structured_prompt(self, data: dict, sim_analysis: dict, code_desc: list) -> str:
# Group clusters
clusters = sim_analysis['similar_file_groups']
anomalies = sim_analysis['anomalous_files']
# Header
prompt = []
prompt.append("You are an expert reviewer. First give group summaries, then detailed line-by-line feedback.")
prompt.append(f"Title: {data['title']}")
prompt.append(f"Description: {data['body']}")
# Clusters
for c in clusters:
prompt.append(f"## Group {c['cluster_id']} ({len(c['files'])} files, avg_sim={c['avg_similarity']:.2f}): {', '.join(c['files'])}")
prompt.append("Files:")
for f in c['files']:
prompt.append(f"- {f}")
prompt.append(f"Summary: Changes in these files share semantic pattern. Focus on shared logic.")
# Anomalies
if anomalies:
prompt.append("## Isolated Files (low similarity with changed files)")
for a in anomalies:
prompt.append(f"- {a['file']} (reason: {a['reason']}, strength: {a.get('anomaly_strength')})")
# Grounding diffs per cluster/files
prompt.append("## Diff Hunks and Context:")
for entry in code_desc:
prompt.append(f"File: {entry['file_name']}\n{entry['diff_hunk']}")
if entry['overlapping_functions']:
prompt.append("Affected functions:")
for f in entry['overlapping_functions']:
prompt.append(f"- {f['function_name']}: {f['function_description']}")
# Request
prompt.append("Provide feedback on groups, then isolated files. After that provide line-by-line feedback in diff format.")
return "\n".join(prompt)
def get_current_logs():
return log_stream.getvalue()
# Pipeline
pipeline = InferencePipeline(InferenceContext("https://github.com/dotnet/xharness"))
def analyze_pr_streaming(pr_url):
log_stream.seek(0)
log_stream.truncate()
response = {}
base_review = ""
final_review = ""
visualization = None
data = pipeline._extract_pr_data(pr_url)
yield base_review, final_review, get_current_logs(), visualization
visualization = pipeline.create_repo_visualization(list(data["changed_files"]), max_files=20)
yield "", "", get_current_logs(), visualization
similarity_analysis = pipeline.analyze_file_similarity(list(data["changed_files"]))
similar_file_groups = similarity_analysis['similar_file_groups']
anomalous_files = similarity_analysis['anomalous_files']
yield "", "", get_current_logs(), visualization
code_description = pipeline.search_code_snippets(data["diff_hunks"])
comprehensive_prompt = pipeline.build_structured_prompt(data, similarity_analysis, code_description)
# Base prompt
base_prompt = f"""You are an expert reviewer. Provide detailed line-by-line feedback.
Title: {data['title']}
Description: {data['body']}
Diff: {data['diff']}
"""
# similar_file_groups_formatted = []
# for i, group in enumerate(similar_file_groups):
# files_str = ", ".join(group['files'])
# similar_file_groups_formatted.append(f"group {i}: {files_str}")
# anomalous_files_formatted = []
# for anomaly in anomalous_files:
# anomalous_files_formatted.append(f"anomaly: {anomaly['file']} (reason: {anomaly['reason']}, strength: {anomaly['anomaly_strength']})")
# grounding_formatted = ""
# for entry in code_description:
# file_name = entry['file_name']
# overlapping_functions = entry['overlapping_functions']
# diff_hunk = entry['diff_hunk']
# if len(overlapping_functions) > 0:
# grounding_formatted += f"In file {file_name}, the following changes were made: {diff_hunk}\n"
# grounding_formatted += f"These changes affected the following functions:\n"
# for func in overlapping_functions:
# grounding_formatted += f"{func['function_name']} - {func['function_description']}\n"
# else:
# grounding_formatted += f"In file {file_name}, the following changes were made: {diff_hunk}\n"
# grounding_formatted += "\n"
# # Create formatted strings for f-string
# similar_groups_text = "\n".join(similar_file_groups_formatted)
# anomalous_files_text = "\n".join(anomalous_files_formatted)
# # TODO: Add local LLM reasoning
# # TODO: Add relevant files from the directory not included
# comprehensive_prompt = f"""{base_prompt}
# FILES THAT ARE SEMANTICALLY CLOSE CHANGED IN THIS PR:
# {similar_groups_text}
# UNEXPECTED CHANGES IN FILES:
# {anomalous_files_text}
# GROUNDING DATA: The following provides specific information about which functions are affected by each diff hunk:
# {grounding_formatted}
# """
base_prompt += f"""
DIFF: {data['diff']}
"""
logger.info(f"Base prompt word count: {len(base_prompt.split())}")
logger.info(f"Base prompt: {base_prompt}")
logger.info(f"Comprehensive prompt word count: {len(comprehensive_prompt.split())}")
logger.info(f"Comprehensive prompt: {comprehensive_prompt}")
logger.info("Calling Azure OpenAI...")
yield "", "", get_current_logs(), visualization
base_review_response = pipeline.enterprise_llm.chat.completions.create(
model=DEPLOYMENT,
messages=[
{"role": "system", "content": "You are an expert code reviewer. Provide thorough, constructive feedback."},
{"role": "user", "content": base_prompt}
],
max_tokens=8192,
temperature=0.3
)
base_review = base_review_response.choices[0].message.content
logger.info("Base review completed")
final_review_response = pipeline.enterprise_llm.chat.completions.create(
model=DEPLOYMENT,
messages=[
{"role": "system", "content": "You are an expert code reviewer. Provide thorough, constructive feedback."},
{"role": "user", "content": comprehensive_prompt}
],
max_tokens=8192,
temperature=0.3
)
final_review = final_review_response.choices[0].message.content
logger.info("Final review completed")
yield base_review, final_review, get_current_logs(), visualization
with gr.Blocks(title="PR Code Review Assistant") as demo:
gr.Markdown("# PR Code Review Assistant")
gr.Markdown("Enter a GitHub PR URL to get comprehensive code review analysis with interactive repository visualization.")
with gr.Row():
pr_url_input = gr.Textbox(
label="GitHub PR URL",
placeholder="https://github.com/owner/repo/pull/123",
value="https://github.com/dotnet/xharness/pull/1416"
)
analyze_btn = gr.Button("Analyze PR", variant="primary")
with gr.Row():
with gr.Column(scale=1):
base_review_output = gr.Textbox(
label="Base Review",
lines=15,
max_lines=30,
interactive=False
)
with gr.Column(scale=1):
final_review_output = gr.Textbox(
label="Comprehensive Review",
lines=15,
max_lines=30,
interactive=False
)
with gr.Row():
with gr.Column(scale=1):
visualization_output = gr.Plot(
label="Repository Files Visualization (3D)",
value=None
)
with gr.Column(scale=1):
logs_output = gr.Textbox(
label="Analysis Logs",
lines=15,
max_lines=25,
interactive=False,
show_copy_button=True
)
analyze_btn.click(
fn=analyze_pr_streaming,
inputs=[pr_url_input],
outputs=[base_review_output, final_review_output, logs_output, visualization_output],
show_progress=True
)
if __name__ == "__main__":
demo.launch()
|