Skip to content

A
aisbf
Commit 32380465 authored by Stefy Lanza (nextime / spora )'s avatar Stefy Lanza (nextime / spora )
Browse files
Options

notifyerrors fix

parent b3b44f6f
Hide whitespace changes
Inline Side-by-side
Showing with 28 additions and 53 deletions
aisbf/config.py
View file @ 32380465
...@@ -182,7 +182,22 @@ class Config: ...@@ -182,7 +182,22 @@ class Config:
logger.info(f"Loading rotations from: {rotations_path}") logger.info(f"Loading rotations from: {rotations_path}")
with open(rotations_path) as f: with open(rotations_path) as f:
data = json.load(f) data = json.load(f)
self.rotations = {k: RotationConfig(**v) for k, v in data['rotations'].items()}
# Extract global notifyerrors setting (top-level, outside rotations)
self.global_notifyerrors = data.get('notifyerrors', False)
logger.info(f"Global notifyerrors setting: {self.global_notifyerrors}")
# Load rotations, merging global notifyerrors with rotation-specific settings
self.rotations = {}
for k, v in data['rotations'].items():
# If rotation doesn't have its own notifyerrors, use global setting
if 'notifyerrors' not in v:
v['notifyerrors'] = self.global_notifyerrors
logger.info(f"Rotation '{k}' using global notifyerrors: {self.global_notifyerrors}")
else:
logger.info(f"Rotation '{k}' has own notifyerrors: {v['notifyerrors']}")
self.rotations[k] = RotationConfig(**v)
logger.info(f"Loaded {len(self.rotations)} rotations: {list(self.rotations.keys())}") logger.info(f"Loaded {len(self.rotations)} rotations: {list(self.rotations.keys())}")
# Validate that all providers referenced in rotations exist # Validate that all providers referenced in rotations exist
......
aisbf/handlers.py
View file @ 32380465
...@@ -1428,39 +1428,14 @@ class RotationHandler: ...@@ -1428,39 +1428,14 @@ class RotationHandler:
tool_calls = None tool_calls = None
final_text = accumulated_response_text final_text = accumulated_response_text
logger.debug(f"=== ACCUMULATED RESPONSE TEXT ===")
logger.debug(f"Total length: {len(accumulated_response_text)}")
logger.debug(f"First 500 chars: {accumulated_response_text[:500]}")
logger.debug(f"Last 200 chars: {accumulated_response_text[-200:]}")
# Check for tool call patterns in the accumulated text # Check for tool call patterns in the accumulated text
if accumulated_response_text: if accumulated_response_text:
import re as re_module import re as re_module
# Initialize tool_match to None # Simple approach: just look for "tool: {...}" pattern and extract the JSON
tool_match = None # This avoids complex nested parsing issues
tool_pattern = r'tool:\s*(\{[^{}]*\{[^{}]*\}[^{}]*\}|\{[^{}]+\})'
# First, check if the response is wrapped in "assistant: [{'type': 'text', 'text': '...'}]" tool_match = re_module.search(tool_pattern, accumulated_response_text, re_module.DOTALL)
# This is a common pattern where the model wraps its response instead of returning plain text
assistant_wrapper_pattern = r"^assistant:\s*\[\s*\{\s*'type':\s*'text',\s*'text':\s*['\"](.+?)['\"]\s*\}\s*\]\s*$"
assistant_wrapper_match = re_module.match(assistant_wrapper_pattern, accumulated_response_text.strip(), re_module.DOTALL)
if assistant_wrapper_match:
# Extract the plain text from the wrapper
extracted_text = assistant_wrapper_match.group(1)
# Unescape common escape sequences
extracted_text = extracted_text.replace("\\'", "'").replace('\\"', '"').replace("\\n", "\n")
logger.debug(f"Extracted text from assistant wrapper: {extracted_text[:200]}...")
final_text = extracted_text
tool_calls = None
else:
# Check for tool call pattern
# Simple approach: just look for "tool: {...}" pattern and extract the JSON
# This avoids complex nested parsing issues
tool_pattern = r'tool:\s*(\{[^{}]*\{[^{}]*\}[^{}]*\}|\{[^{}]+\})'
tool_match = re_module.search(tool_pattern, accumulated_response_text, re_module.DOTALL)
logger.debug(f"Tool pattern match: {tool_match is not None}")
if tool_match: if tool_match:
try: try:
...@@ -1481,29 +1456,16 @@ class RotationHandler: ...@@ -1481,29 +1456,16 @@ class RotationHandler:
break break
tool_json_str = accumulated_response_text[json_start:json_end] tool_json_str = accumulated_response_text[json_start:json_end]
logger.debug(f"=== TOOL JSON EXTRACTION ===") logger.debug(f"Extracted tool JSON: {tool_json_str[:200]}...")
logger.debug(f"Extracted tool JSON length: {len(tool_json_str)}")
logger.debug(f"Extracted tool JSON (first 500 chars): {tool_json_str[:500]}")
logger.debug(f"First 20 bytes (repr): {repr(tool_json_str[:20])}")
logger.debug(f"ASCII codes for first 20 chars: {[ord(c) for c in tool_json_str[:20]]}")
try: try:
parsed_tool = json.loads(tool_json_str) parsed_tool = json.loads(tool_json_str)
logger.debug(f"Successfully parsed tool JSON") except json.JSONDecodeError:
except json.JSONDecodeError as e:
logger.debug(f"JSON parse error: {e}")
logger.debug(f"Error at position {e.pos if hasattr(e, 'pos') else 'unknown'}")
# Try fixing common issues: single quotes, trailing commas # Try fixing common issues: single quotes, trailing commas
fixed_json = tool_json_str.replace("'", '"') fixed_json = tool_json_str.replace("'", '"')
fixed_json = re_module.sub(r',\s*}', '}', fixed_json) fixed_json = re_module.sub(r',\s*}', '}', fixed_json)
fixed_json = re_module.sub(r',\s*]', ']', fixed_json) fixed_json = re_module.sub(r',\s*]', ']', fixed_json)
logger.debug(f"Fixed JSON (first 200 chars): {fixed_json[:200]}") parsed_tool = json.loads(fixed_json)
try:
parsed_tool = json.loads(fixed_json)
logger.debug(f"Successfully parsed fixed JSON")
except json.JSONDecodeError as e2:
logger.debug(f"Fixed JSON also failed: {e2}")
raise e # Re-raise original error
# Convert to OpenAI tool_calls format # Convert to OpenAI tool_calls format
tool_calls = [{ tool_calls = [{
...@@ -1586,8 +1548,8 @@ class RotationHandler: ...@@ -1586,8 +1548,8 @@ class RotationHandler:
yield f"data: {json.dumps(text_chunk)}\n\n".encode('utf-8') yield f"data: {json.dumps(text_chunk)}\n\n".encode('utf-8')
else: else:
# No tool calls detected, send text normally # No tool calls detected, send text normally
# Send the final text (which may have been extracted from wrapper) # Send the accumulated text as a single chunk
if final_text: if accumulated_response_text:
text_chunk = { text_chunk = {
"id": response_id, "id": response_id,
"object": "chat.completion.chunk", "object": "chat.completion.chunk",
...@@ -1600,7 +1562,7 @@ class RotationHandler: ...@@ -1600,7 +1562,7 @@ class RotationHandler:
"choices": [{ "choices": [{
"index": 0, "index": 0,
"delta": { "delta": {
"content": final_text, "content": accumulated_response_text,
"refusal": None, "refusal": None,
"role": "assistant", "role": "assistant",
"tool_calls": None "tool_calls": None
...@@ -1613,10 +1575,8 @@ class RotationHandler: ...@@ -1613,10 +1575,8 @@ class RotationHandler:
yield f"data: {json.dumps(text_chunk)}\n\n".encode('utf-8') yield f"data: {json.dumps(text_chunk)}\n\n".encode('utf-8')
# Send final chunk with finish reason and usage statistics # Send final chunk with finish reason and usage statistics
# Use final_text for token counting (which may have been extracted from wrapper) if accumulated_response_text:
text_for_token_count = final_text if final_text else accumulated_response_text completion_tokens = count_messages_tokens([{"role": "assistant", "content": accumulated_response_text}], model_name)
if text_for_token_count:
completion_tokens = count_messages_tokens([{"role": "assistant", "content": text_for_token_count}], model_name)
total_tokens = effective_context + completion_tokens total_tokens = effective_context + completion_tokens
final_chunk = { final_chunk = {
"id": response_id, "id": response_id,
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment