Update src/txagent/txagent.py
Browse files- src/txagent/txagent.py +135 -141
src/txagent/txagent.py
CHANGED
@@ -14,7 +14,6 @@ from .toolrag import ToolRAGModel
|
|
14 |
import torch
|
15 |
import logging
|
16 |
from difflib import SequenceMatcher
|
17 |
-
import asyncio
|
18 |
import threading
|
19 |
|
20 |
logger = logging.getLogger(__name__)
|
@@ -455,6 +454,140 @@ Patient Record Excerpt:
|
|
455 |
logger.debug("Quick summary output: %s", output[:100])
|
456 |
return output
|
457 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
458 |
def run_self_agent(self, message: str, temperature: float, max_new_tokens: int, max_token: int):
|
459 |
logger.debug("Starting self agent")
|
460 |
conversation = self.set_system_prompt([], self.self_prompt)
|
@@ -583,143 +716,4 @@ Summarize the function responses in one sentence with all necessary information.
|
|
583 |
setattr(self, key, value)
|
584 |
updated_attributes[key] = value
|
585 |
logger.debug("Updated parameters: %s", updated_attributes)
|
586 |
-
return updated_attributes
|
587 |
-
|
588 |
-
async def run_background_report(self, message: str, history: list, temperature: float,
|
589 |
-
max_new_tokens: int, max_token: int, call_agent: bool,
|
590 |
-
conversation: gr.State, max_round: int, seed: int,
|
591 |
-
call_agent_level: int, report_path: str):
|
592 |
-
"""Run detailed report generation in the background and save to file"""
|
593 |
-
logger.debug("Starting background report for message: %s", message[:100])
|
594 |
-
combined_response = ""
|
595 |
-
history_copy = history.copy()
|
596 |
-
|
597 |
-
picked_tools_prompt, call_agent_level = self.initialize_tools_prompt(
|
598 |
-
call_agent, call_agent_level, message)
|
599 |
-
conversation = self.initialize_conversation(message, conversation, history_copy)
|
600 |
-
|
601 |
-
next_round = True
|
602 |
-
current_round = 0
|
603 |
-
enable_summary = False
|
604 |
-
last_status = {}
|
605 |
-
token_overflow = False
|
606 |
-
|
607 |
-
if self.enable_checker:
|
608 |
-
checker = ReasoningTraceChecker(message, conversation, init_index=len(conversation))
|
609 |
-
|
610 |
-
try:
|
611 |
-
while next_round and current_round < max_round:
|
612 |
-
current_round += 1
|
613 |
-
last_outputs = []
|
614 |
-
if last_outputs:
|
615 |
-
function_call_messages, picked_tools_prompt, special_tool_call, _ = yield from self.run_function_call_stream(
|
616 |
-
last_outputs, return_message=True, existing_tools_prompt=picked_tools_prompt,
|
617 |
-
message_for_call_agent=message, call_agent=call_agent,
|
618 |
-
call_agent_level=call_agent_level, temperature=temperature,
|
619 |
-
return_gradio_history=False)
|
620 |
-
|
621 |
-
if special_tool_call == 'Finish':
|
622 |
-
next_round = False
|
623 |
-
conversation.extend(function_call_messages)
|
624 |
-
combined_response += function_call_messages[0]['content'] + "\n"
|
625 |
-
break
|
626 |
-
|
627 |
-
if (self.enable_summary or token_overflow) and not call_agent:
|
628 |
-
enable_summary = True
|
629 |
-
last_status = self.function_result_summary(
|
630 |
-
conversation, status=last_status, enable_summary=enable_summary)
|
631 |
-
|
632 |
-
if function_call_messages:
|
633 |
-
conversation.extend(function_call_messages)
|
634 |
-
combined_response += tool_result_format(function_call_messages) + "\n"
|
635 |
-
else:
|
636 |
-
next_round = False
|
637 |
-
combined_response += ''.join(last_outputs).replace("</s>", "") + "\n"
|
638 |
-
break
|
639 |
-
|
640 |
-
if self.enable_checker:
|
641 |
-
good_status, wrong_info = checker.check_conversation()
|
642 |
-
if not good_status:
|
643 |
-
logger.warning("Checker error: %s", wrong_info)
|
644 |
-
break
|
645 |
-
|
646 |
-
tools = picked_tools_prompt
|
647 |
-
last_outputs_str, token_overflow = self.llm_infer(
|
648 |
-
messages=conversation, temperature=temperature, tools=tools,
|
649 |
-
max_new_tokens=max_new_tokens, max_token=max_token, seed=seed, check_token_status=True)
|
650 |
-
|
651 |
-
if last_outputs_str is None:
|
652 |
-
if self.force_finish:
|
653 |
-
last_outputs_str = self.get_answer_based_on_unfinished_reasoning(
|
654 |
-
conversation, temperature, max_new_tokens, max_token)
|
655 |
-
combined_response += last_outputs_str + "\n"
|
656 |
-
break
|
657 |
-
combined_response += "Token limit exceeded.\n"
|
658 |
-
break
|
659 |
-
|
660 |
-
combined_response += last_outputs_str + "\n"
|
661 |
-
last_outputs.append(last_outputs_str)
|
662 |
-
|
663 |
-
if next_round and self.force_finish:
|
664 |
-
last_outputs_str = self.get_answer_based_on_unfinished_reasoning(
|
665 |
-
conversation, temperature, max_new_tokens, max_token)
|
666 |
-
combined_response += last_outputs_str + "\n"
|
667 |
-
|
668 |
-
# Save report
|
669 |
-
try:
|
670 |
-
with open(report_path, "w", encoding="utf-8") as f:
|
671 |
-
f.write(combined_response)
|
672 |
-
logger.info("Detailed report saved to %s", report_path)
|
673 |
-
except Exception as e:
|
674 |
-
logger.error("Failed to save report: %s", e)
|
675 |
-
|
676 |
-
except Exception as e:
|
677 |
-
logger.error("Background report error: %s", e)
|
678 |
-
combined_response += f"Error: {e}\n"
|
679 |
-
with open(report_path, "w", encoding="utf-8") as f:
|
680 |
-
f.write(combined_response)
|
681 |
-
|
682 |
-
finally:
|
683 |
-
torch.cuda.empty_cache()
|
684 |
-
gc.collect()
|
685 |
-
|
686 |
-
def run_gradio_chat(self, message: str, history: list, temperature: float,
|
687 |
-
max_new_tokens: int, max_token: int, call_agent: bool,
|
688 |
-
conversation: gr.State, max_round: int = 3, seed: int = None,
|
689 |
-
call_agent_level: int = 0, sub_agent_task: str = None,
|
690 |
-
uploaded_files: list = None, report_path: str = None):
|
691 |
-
logger.debug("Chat started, message: %s", message[:100])
|
692 |
-
if not message or len(message.strip()) < 5:
|
693 |
-
yield "Please provide a valid message or upload files to analyze."
|
694 |
-
return
|
695 |
-
|
696 |
-
if message.startswith("[\U0001f9f0 Tool_RAG") or message.startswith("⚒️"):
|
697 |
-
return
|
698 |
-
|
699 |
-
clinical_keywords = ['medication', 'symptom', 'evaluation', 'diagnosis']
|
700 |
-
has_clinical_data = any(keyword in message.lower() for keyword in clinical_keywords)
|
701 |
-
call_agent = call_agent and not has_clinical_data
|
702 |
-
|
703 |
-
# Generate quick summary
|
704 |
-
quick_summary = self.run_quick_summary(
|
705 |
-
message, temperature=temperature, max_new_tokens=256, max_token=1024)
|
706 |
-
history.append(ChatMessage(role="assistant", content=f"**Quick Summary:**\n{quick_summary}"))
|
707 |
-
yield history
|
708 |
-
|
709 |
-
# Start background report generation
|
710 |
-
if report_path:
|
711 |
-
loop = asyncio.get_event_loop()
|
712 |
-
threading.Thread(
|
713 |
-
target=lambda: loop.run_until_complete(
|
714 |
-
self.run_background_report(
|
715 |
-
message, history, temperature, max_new_tokens, max_token, call_agent,
|
716 |
-
conversation, max_round, seed, call_agent_level, report_path
|
717 |
-
)
|
718 |
-
),
|
719 |
-
daemon=True
|
720 |
-
).start()
|
721 |
-
history.append(ChatMessage(
|
722 |
-
role="assistant",
|
723 |
-
content="Generating detailed report in the background. Download will be available when ready."
|
724 |
-
))
|
725 |
-
yield history
|
|
|
14 |
import torch
|
15 |
import logging
|
16 |
from difflib import SequenceMatcher
|
|
|
17 |
import threading
|
18 |
|
19 |
logger = logging.getLogger(__name__)
|
|
|
454 |
logger.debug("Quick summary output: %s", output[:100])
|
455 |
return output
|
456 |
|
457 |
+
def run_background_report(self, message: str, history: list, temperature: float,
|
458 |
+
max_new_tokens: int, max_token: int, call_agent: bool,
|
459 |
+
conversation: gr.State, max_round: int, seed: int,
|
460 |
+
call_agent_level: int, report_path: str):
|
461 |
+
"""Run detailed report generation in the background and save to file"""
|
462 |
+
logger.debug("Starting background report for message: %s", message[:100])
|
463 |
+
combined_response = ""
|
464 |
+
history_copy = history.copy()
|
465 |
+
|
466 |
+
picked_tools_prompt, call_agent_level = self.initialize_tools_prompt(
|
467 |
+
call_agent, call_agent_level, message)
|
468 |
+
conversation = self.initialize_conversation(message, conversation, history_copy)
|
469 |
+
|
470 |
+
next_round = True
|
471 |
+
current_round = 0
|
472 |
+
enable_summary = False
|
473 |
+
last_status = {}
|
474 |
+
token_overflow = False
|
475 |
+
|
476 |
+
if self.enable_checker:
|
477 |
+
checker = ReasoningTraceChecker(message, conversation, init_index=len(conversation))
|
478 |
+
|
479 |
+
try:
|
480 |
+
while next_round and current_round < max_round:
|
481 |
+
current_round += 1
|
482 |
+
last_outputs = []
|
483 |
+
if last_outputs:
|
484 |
+
function_call_messages, picked_tools_prompt, special_tool_call = self.run_function_call(
|
485 |
+
last_outputs, return_message=True, existing_tools_prompt=picked_tools_prompt,
|
486 |
+
message_for_call_agent=message, call_agent=call_agent,
|
487 |
+
call_agent_level=call_agent_level, temperature=temperature)
|
488 |
+
|
489 |
+
if special_tool_call == 'Finish':
|
490 |
+
next_round = False
|
491 |
+
conversation.extend(function_call_messages)
|
492 |
+
combined_response += function_call_messages[0]['content'] + "\n"
|
493 |
+
break
|
494 |
+
|
495 |
+
if (self.enable_summary or token_overflow) and not call_agent:
|
496 |
+
enable_summary = True
|
497 |
+
last_status = self.function_result_summary(
|
498 |
+
conversation, status=last_status, enable_summary=enable_summary)
|
499 |
+
|
500 |
+
if function_call_messages:
|
501 |
+
conversation.extend(function_call_messages)
|
502 |
+
combined_response += tool_result_format(function_call_messages) + "\n"
|
503 |
+
else:
|
504 |
+
next_round = False
|
505 |
+
combined_response += ''.join(last_outputs).replace("</s>", "") + "\n"
|
506 |
+
break
|
507 |
+
|
508 |
+
if self.enable_checker:
|
509 |
+
good_status, wrong_info = checker.check_conversation()
|
510 |
+
if not good_status:
|
511 |
+
logger.warning("Checker error: %s", wrong_info)
|
512 |
+
break
|
513 |
+
|
514 |
+
tools = picked_tools_prompt
|
515 |
+
last_outputs_str, token_overflow = self.llm_infer(
|
516 |
+
messages=conversation, temperature=temperature, tools=tools,
|
517 |
+
max_new_tokens=max_new_tokens, max_token=max_token, seed=seed, check_token_status=True)
|
518 |
+
|
519 |
+
if last_outputs_str is None:
|
520 |
+
if self.force_finish:
|
521 |
+
last_outputs_str = self.get_answer_based_on_unfinished_reasoning(
|
522 |
+
conversation, temperature, max_new_tokens, max_token)
|
523 |
+
combined_response += last_outputs_str + "\n"
|
524 |
+
break
|
525 |
+
combined_response += "Token limit exceeded.\n"
|
526 |
+
break
|
527 |
+
|
528 |
+
combined_response += last_outputs_str + "\n"
|
529 |
+
last_outputs.append(last_outputs_str)
|
530 |
+
|
531 |
+
if next_round and self.force_finish:
|
532 |
+
last_outputs_str = self.get_answer_based_on_unfinished_reasoning(
|
533 |
+
conversation, temperature, max_new_tokens, max_token)
|
534 |
+
combined_response += last_outputs_str + "\n"
|
535 |
+
|
536 |
+
# Save report
|
537 |
+
try:
|
538 |
+
with open(report_path, "w", encoding="utf-8") as f:
|
539 |
+
f.write(combined_response)
|
540 |
+
logger.info("Detailed report saved to %s", report_path)
|
541 |
+
except Exception as e:
|
542 |
+
logger.error("Failed to save report: %s", e)
|
543 |
+
|
544 |
+
except Exception as e:
|
545 |
+
logger.error("Background report error: %s", e)
|
546 |
+
combined_response += f"Error: {e}\n"
|
547 |
+
with open(report_path, "w", encoding="utf-8") as f:
|
548 |
+
f.write(combined_response)
|
549 |
+
|
550 |
+
finally:
|
551 |
+
torch.cuda.empty_cache()
|
552 |
+
gc.collect()
|
553 |
+
|
554 |
+
def run_gradio_chat(self, message: str, history: list, temperature: float,
|
555 |
+
max_new_tokens: int, max_token: int, call_agent: bool,
|
556 |
+
conversation: gr.State, max_round: int = 3, seed: int = None,
|
557 |
+
call_agent_level: int = 0, sub_agent_task: str = None,
|
558 |
+
uploaded_files: list = None, report_path: str = None):
|
559 |
+
logger.debug("Chat started, message: %s", message[:100])
|
560 |
+
if not message or len(message.strip()) < 5:
|
561 |
+
yield "Please provide a valid message or upload files to analyze."
|
562 |
+
return
|
563 |
+
|
564 |
+
if message.startswith("[\U0001f9f0 Tool_RAG") or message.startswith("⚒️"):
|
565 |
+
return
|
566 |
+
|
567 |
+
clinical_keywords = ['medication', 'symptom', 'evaluation', 'diagnosis']
|
568 |
+
has_clinical_data = any(keyword in message.lower() for keyword in clinical_keywords)
|
569 |
+
call_agent = call_agent and not has_clinical_data
|
570 |
+
|
571 |
+
# Generate quick summary
|
572 |
+
quick_summary = self.run_quick_summary(
|
573 |
+
message, temperature=temperature, max_new_tokens=256, max_token=1024)
|
574 |
+
history.append(ChatMessage(role="assistant", content=f"**Quick Summary:**\n{quick_summary}"))
|
575 |
+
yield history
|
576 |
+
|
577 |
+
# Start background report generation
|
578 |
+
if report_path:
|
579 |
+
threading.Thread(
|
580 |
+
target=self.run_background_report,
|
581 |
+
args=(message, history, temperature, max_new_tokens, max_token, call_agent,
|
582 |
+
conversation, max_round, seed, call_agent_level, report_path),
|
583 |
+
daemon=True
|
584 |
+
).start()
|
585 |
+
history.append(ChatMessage(
|
586 |
+
role="assistant",
|
587 |
+
content="Generating detailed report in the background. Download will be available when ready."
|
588 |
+
))
|
589 |
+
yield history
|
590 |
+
|
591 |
def run_self_agent(self, message: str, temperature: float, max_new_tokens: int, max_token: int):
|
592 |
logger.debug("Starting self agent")
|
593 |
conversation = self.set_system_prompt([], self.self_prompt)
|
|
|
716 |
setattr(self, key, value)
|
717 |
updated_attributes[key] = value
|
718 |
logger.debug("Updated parameters: %s", updated_attributes)
|
719 |
+
return updated_attributes
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|