Spaces:
Running
Running
File size: 12,491 Bytes
5f3e9f5 | 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 | """AI API client for processing text input."""
import os
import sys
import threading
import httpx # type: ignore
from openai import OpenAI # type: ignore
# Add config to path
config_path = os.path.join(os.path.dirname(__file__), '..', '..', 'config')
sys.path.insert(0, config_path)
from config import API_URL, MODELS_CONFIG # type: ignore
# Add utils to path
utils_path = os.path.join(os.path.dirname(__file__), '..', 'utils')
sys.path.insert(0, utils_path)
from cache_manager import CacheManager # type: ignore
from retry_handler import retry_with_backoff # type: ignore
# Initialize cache manager
cache = CacheManager()
def load_system_prompt():
"""Load the system prompt from file."""
try:
prompt_path = os.path.join(os.path.dirname(__file__), '..', '..', 'config', 'system_prompt.txt')
with open(prompt_path, 'r', encoding='utf-8') as f:
return f.read()
except FileNotFoundError:
return """You are an expert web developer. Convert the raw text notes into properly formatted HTML content using CSS classes: .exercise-title, .question, .answer, .vocabulary-item, .section-number. Output ONLY the HTML content without DOCTYPE, html, head, or body tags."""
@retry_with_backoff(max_retries=3, base_delay=2, max_delay=30)
def _make_ai_request(client, system_prompt, user_text, model_config, cancel_event=None):
"""Make the actual AI request with proper system/user roles (wrapped with retry logic)."""
# Safely handle extra_body params if they are defined
kwargs = {
"model": model_config['model'],
"messages": [
{"role": "system", "content": system_prompt},
{"role": "user", "content": user_text}
],
"temperature": model_config['temperature'],
"top_p": model_config['top_p'],
"max_tokens": model_config['max_tokens'],
"stream": True,
}
if model_config.get('extra_body'):
kwargs['extra_body'] = model_config['extra_body']
if model_config.get('seed'):
kwargs['seed'] = model_config['seed']
completion = client.chat.completions.create(**kwargs)
# Collect streamed response
full_response = ""
print("π₯ Receiving response:", flush=True)
last_print_len: int = 0
for chunk in completion:
# Check for cancellation
if cancel_event and cancel_event.is_set():
print("\nβ οΈ Request cancelled by user", flush=True)
raise CancelledError("Generation cancelled by user")
if not getattr(chunk, "choices", None):
continue
# Handle reasoning content (thought process) if present
reasoning = getattr(chunk.choices[0].delta, "reasoning_content", None)
if reasoning:
print(reasoning, end="", flush=True)
continue # Don't add reasoning to full_response to avoid corrupting HTML
if chunk.choices[0].delta.content is not None:
content = chunk.choices[0].delta.content
full_response += content
# Print a dot for every 50 characters received
if len(full_response) >= last_print_len + 50: # type: ignore
print(".", end="", flush=True)
last_print_len = len(full_response)
print("\n", flush=True)
return full_response
class CancelledError(Exception):
"""Raised when a generation is cancelled by the user."""
pass
# Store active cancel events keyed by operation_id
_active_operations = {}
_operations_lock = threading.Lock()
def register_operation(operation_id):
"""Register a new operation and return its cancel event."""
event = threading.Event()
with _operations_lock:
_active_operations[operation_id] = event
return event
def cancel_operation(operation_id):
"""Cancel an active operation by setting its cancel event."""
with _operations_lock:
event = _active_operations.get(operation_id)
if event:
event.set()
return True
return False
def unregister_operation(operation_id):
"""Clean up a completed operation."""
with _operations_lock:
_active_operations.pop(operation_id, None)
def verify_html_content(input_text, html_content, cancel_event=None, model_choice='default'):
"""Verify that the generated HTML preserves all content from the input text."""
print("=" * 60, flush=True)
print("π€ Verifying HTML content against original text...", flush=True)
verification_sys_prompt = (
"You are an expert quality assurance reviewer. Your job is to compare the original raw text with the generated HTML output.\n"
"Check line by line to ensure NO content, questions, answers, or vocabulary from the original text has been skipped, summarized, or omitted in the HTML.\n"
"If ALL content is carefully preserved in the HTML, output EXACTLY the word 'PASS' and nothing else.\n"
"If ANY content was removed, summarized, or omitted, output a list of the specific missing content and instructions on what needs to be added back. Do not output 'PASS'."
)
verification_user_prompt = (
f"--- ORIGINAL RAW TEXT ---\n{input_text}\n\n"
f"--- GENERATED HTML ---\n{html_content}\n\n"
"Did the HTML preserve all the content? Output 'PASS' or list the missing content."
)
try:
model_config = MODELS_CONFIG.get(model_choice, MODELS_CONFIG['default'])
client = OpenAI(base_url=API_URL, api_key=model_config['api_key'])
response = _make_ai_request(client, verification_sys_prompt, verification_user_prompt, model_config, cancel_event=cancel_event)
response = response.strip()
print(f"β
Verification result: {response[:100]}...", flush=True)
if response.upper() == "PASS" or response.upper().startswith("PASS"):
return "PASS"
else:
return response
except CancelledError:
print("β οΈ Verification was cancelled")
return None
except Exception as e:
print(f"β Verification failed: {e}")
return "PASS" # Fail open if verification errors
def get_ai_revision(input_text, previous_html, feedback, cancel_event=None, model_choice='default'):
"""Ask the AI to revise the HTML based on verification feedback."""
print("=" * 60, flush=True)
print("π€ Requesting AI revision based on feedback...", flush=True)
base_sys_prompt = load_system_prompt()
revision_sys_prompt = (
f"{base_sys_prompt}\n\n"
"CRITICAL REVISION INSTRUCTIONS:\n"
"You previously generated HTML for this text, but the quality assurance reviewer found that you skipped or summarized some content.\n"
"Here is the exact feedback on what is missing:\n"
"-------------------------------------\n"
f"{feedback}\n"
"-------------------------------------\n"
"Your task:\n"
"1. Rewrite the ENTIRE HTML document from start to finish.\n"
"2. You MUST include ALL content from the original text.\n"
"3. Pay special attention to the feedback above and guarantee that all missing parts are inserted in the correct locations.\n"
"4. This is a strict test. If you skip, omit, or summarize ANY paragraph, question, or option, you will fail.\n"
"DO NOT output anything other than raw HTML. No markdown code blocks, no explanations. Start with <!DOCTYPE html>."
)
return get_ai_response(input_text, use_cache=False, cancel_event=cancel_event, system_prompt=revision_sys_prompt, model_choice=model_choice)
def _cache_key(input_text, model_choice, system_prompt):
"""Cache key varies on (model_choice, system_prompt, input_text) β switching
model or adding a custom system prompt must produce a different key so the
cache doesn't return the response from a previous configuration."""
return f"{model_choice}|{system_prompt or ''}|{input_text}"
def get_ai_response(input_text, use_cache=True, cancel_event=None, system_prompt=None, model_choice='default'):
"""Send text to AI model and get response with proper system/user message roles."""
print("=" * 60, flush=True)
print("π€ Sending request to AI using OpenAI library...", flush=True)
# Check cache first β key includes model + system prompt, not just the input.
resolved_system_prompt = system_prompt if system_prompt is not None else load_system_prompt()
cache_key = _cache_key(input_text, model_choice, resolved_system_prompt)
if use_cache:
cached_response = cache.get(cache_key)
if cached_response:
print("=" * 60, flush=True)
return cached_response
try:
model_config = MODELS_CONFIG.get(model_choice, MODELS_CONFIG['default'])
# Fail fast on an obviously-unconfigured key rather than making the
# UI sit at 0% while the OpenAI client retries a bogus endpoint.
placeholder_keys = {'', 'your-api-key-here', 'REPLACE_ME'}
resolved_key = (model_config.get('api_key') or '').strip()
if resolved_key in placeholder_keys:
raise RuntimeError(
"AI is not configured: API_KEY is missing or a placeholder. "
"Edit backend/config/config.py (or set the API_KEY env var) "
"with a real key and restart the backend."
)
# Initialize OpenAI client with per-phase timeouts. A single
# scalar `timeout=60` was previously used, which treated the
# whole streaming completion as one 60-second budget and meant
# any AI response taking longer than 60s (common for 1000-char
# inputs on slower endpoints) timed out, got retried 3x by the
# backoff decorator, and pushed a normal 3-4 min run past 8 min.
# httpx.Timeout gives us fast-fail on connect + generous read
# time for the streamed body. Override via env vars.
connect_timeout = float(os.environ.get('AI_CONNECT_TIMEOUT', '15'))
read_timeout = float(os.environ.get('AI_READ_TIMEOUT', '600'))
client = OpenAI(
base_url=API_URL,
api_key=resolved_key,
timeout=httpx.Timeout(
connect=connect_timeout,
read=read_timeout,
write=30.0,
pool=30.0,
),
max_retries=0, # _make_ai_request already handles retries
)
system_prompt = resolved_system_prompt
print(f"π Input length: {len(input_text)} characters", flush=True)
print(f"π System prompt length: {len(system_prompt)} characters", flush=True)
print(f"π API URL: {API_URL}", flush=True)
print(f"π Using model config: {model_choice} -> {model_config['model']}", flush=True)
print(f"β³ Sending request with streaming...\n", flush=True)
# Make request with retry logic and proper roles
full_response = _make_ai_request(client, system_prompt, input_text, model_config, cancel_event=cancel_event)
print(f"β
Response received successfully", flush=True)
print(f"π Content length: {len(full_response)} characters", flush=True)
# Clean up response - remove markdown code blocks if present
full_response = full_response.strip()
if full_response.startswith("```html"):
full_response = full_response[7:] # Remove ```html
if full_response.startswith("```"):
full_response = full_response[3:] # Remove ```
if full_response.endswith("```"):
full_response = full_response[:-3] # Remove trailing ```
full_response = full_response.strip()
print(f"π First 100 chars: {full_response[:100]}...", flush=True)
# Cache the response using the composite key so later requests with a
# different model / system_prompt don't silently get this response back.
if use_cache:
cache.set(cache_key, full_response)
return full_response
except CancelledError:
print("β οΈ Generation was cancelled")
return None
except Exception as e:
print(f"β Request failed: {type(e).__name__}: {e}")
import traceback
traceback.print_exc()
return None
finally:
print("=" * 60)
|