Update app.py
Browse files
app.py
CHANGED
|
@@ -431,51 +431,139 @@ def handsome_chat_completions():
|
|
| 431 |
def generate():
|
| 432 |
first_chunk_time = None
|
| 433 |
full_response_content = ""
|
| 434 |
-
|
|
|
|
|
|
|
|
|
|
| 435 |
for chunk in response.iter_content(chunk_size=1024):
|
| 436 |
if chunk:
|
| 437 |
if first_chunk_time is None:
|
| 438 |
first_chunk_time = time.time()
|
|
|
|
| 439 |
full_response_content += chunk.decode("utf-8")
|
| 440 |
|
| 441 |
-
|
| 442 |
-
|
| 443 |
-
|
| 444 |
-
|
| 445 |
-
|
| 446 |
-
|
| 447 |
-
|
| 448 |
-
|
| 449 |
-
|
| 450 |
-
|
| 451 |
-
|
| 452 |
-
|
| 453 |
-
|
| 454 |
-
|
| 455 |
-
|
| 456 |
-
|
| 457 |
-
|
| 458 |
-
|
| 459 |
-
if "
|
| 460 |
-
|
| 461 |
-
|
| 462 |
-
|
| 463 |
-
|
| 464 |
-
|
| 465 |
-
|
| 466 |
-
|
| 467 |
-
|
| 468 |
-
|
| 469 |
-
|
| 470 |
-
|
| 471 |
-
|
| 472 |
-
|
| 473 |
-
|
| 474 |
-
|
| 475 |
-
|
| 476 |
-
|
| 477 |
-
|
| 478 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 479 |
|
| 480 |
end_time = time.time()
|
| 481 |
first_token_time = (
|
|
@@ -484,41 +572,6 @@ def handsome_chat_completions():
|
|
| 484 |
)
|
| 485 |
total_time = end_time - start_time
|
| 486 |
|
| 487 |
-
prompt_tokens = 0
|
| 488 |
-
completion_tokens = 0
|
| 489 |
-
for line in full_response_content.splitlines():
|
| 490 |
-
if line.startswith("data:"):
|
| 491 |
-
line = line[5:].strip()
|
| 492 |
-
if line == "[DONE]":
|
| 493 |
-
continue
|
| 494 |
-
try:
|
| 495 |
-
response_json = json.loads(line)
|
| 496 |
-
|
| 497 |
-
if (
|
| 498 |
-
"usage" in response_json and
|
| 499 |
-
"completion_tokens" in response_json["usage"]
|
| 500 |
-
):
|
| 501 |
-
completion_tokens += response_json[
|
| 502 |
-
"usage"
|
| 503 |
-
]["completion_tokens"]
|
| 504 |
-
|
| 505 |
-
if (
|
| 506 |
-
"usage" in response_json and
|
| 507 |
-
"prompt_tokens" in response_json["usage"]
|
| 508 |
-
):
|
| 509 |
-
prompt_tokens = response_json[
|
| 510 |
-
"usage"
|
| 511 |
-
]["prompt_tokens"]
|
| 512 |
-
|
| 513 |
-
except (
|
| 514 |
-
KeyError,
|
| 515 |
-
ValueError,
|
| 516 |
-
IndexError
|
| 517 |
-
) as e:
|
| 518 |
-
logging.error(
|
| 519 |
-
f"解析流式响应单行 JSON 失败: {e}, "
|
| 520 |
-
f"行内容: {line}"
|
| 521 |
-
)
|
| 522 |
|
| 523 |
user_content = ""
|
| 524 |
messages = data.get("messages", [])
|
|
@@ -542,7 +595,7 @@ def handsome_chat_completions():
|
|
| 542 |
user_content_replaced = user_content.replace(
|
| 543 |
'\n', '\\n'
|
| 544 |
).replace('\r', '\\n')
|
| 545 |
-
|
| 546 |
logging.info(
|
| 547 |
f"使用的key: {api_key}, "
|
| 548 |
f"提示token: {prompt_tokens}, "
|
|
@@ -562,6 +615,7 @@ def handsome_chat_completions():
|
|
| 562 |
content_type=response.headers['Content-Type']
|
| 563 |
)
|
| 564 |
else:
|
|
|
|
| 565 |
response.raise_for_status()
|
| 566 |
end_time = time.time()
|
| 567 |
response_json = response.json()
|
|
@@ -663,6 +717,7 @@ def handsome_chat_completions():
|
|
| 663 |
logging.error(f"请求转发异常: {e}")
|
| 664 |
return jsonify({"error": str(e)}), 500
|
| 665 |
|
|
|
|
| 666 |
if __name__ == '__main__':
|
| 667 |
logging.info(f"环境变量:{os.environ}")
|
| 668 |
|
|
|
|
| 431 |
def generate():
|
| 432 |
first_chunk_time = None
|
| 433 |
full_response_content = ""
|
| 434 |
+
reasoning_content_accumulated = ""
|
| 435 |
+
content_accumulated = ""
|
| 436 |
+
|
| 437 |
+
|
| 438 |
for chunk in response.iter_content(chunk_size=1024):
|
| 439 |
if chunk:
|
| 440 |
if first_chunk_time is None:
|
| 441 |
first_chunk_time = time.time()
|
| 442 |
+
|
| 443 |
full_response_content += chunk.decode("utf-8")
|
| 444 |
|
| 445 |
+
try:
|
| 446 |
+
for line in chunk.decode("utf-8").splitlines():
|
| 447 |
+
if line.startswith("data:"):
|
| 448 |
+
line = line[5:].strip()
|
| 449 |
+
if line == "[DONE]":
|
| 450 |
+
continue
|
| 451 |
+
try:
|
| 452 |
+
response_json = json.loads(line)
|
| 453 |
+
|
| 454 |
+
if (
|
| 455 |
+
"usage" in response_json and
|
| 456 |
+
"completion_tokens" in response_json["usage"]
|
| 457 |
+
):
|
| 458 |
+
completion_tokens = response_json[
|
| 459 |
+
"usage"
|
| 460 |
+
]["completion_tokens"]
|
| 461 |
+
|
| 462 |
+
# Special handling for deepseek-reasoner in streaming mode
|
| 463 |
+
if model_name == "deepseek-reasoner" and "choices" in response_json and len(response_json["choices"]) > 0:
|
| 464 |
+
delta = response_json["choices"][0].get("delta", {})
|
| 465 |
+
|
| 466 |
+
if "reasoning_content" in delta and delta["reasoning_content"]:
|
| 467 |
+
reasoning_content = delta["reasoning_content"]
|
| 468 |
+
|
| 469 |
+
formatted_reasoning_chunk = {
|
| 470 |
+
"id": response_json.get("id", ""),
|
| 471 |
+
"object": "chat.completion.chunk",
|
| 472 |
+
"created": response_json.get("created", int(time.time())),
|
| 473 |
+
"model": model_name,
|
| 474 |
+
"choices": [
|
| 475 |
+
{
|
| 476 |
+
"index": 0,
|
| 477 |
+
"delta": {
|
| 478 |
+
"content": f"```Thinking\n{reasoning_content}\n```",
|
| 479 |
+
},
|
| 480 |
+
"finish_reason": None
|
| 481 |
+
}
|
| 482 |
+
],
|
| 483 |
+
"usage": None,
|
| 484 |
+
}
|
| 485 |
+
yield f"data: {json.dumps(formatted_reasoning_chunk)}\n\n".encode('utf-8')
|
| 486 |
+
if "content" in delta and delta["content"]:
|
| 487 |
+
content = delta["content"]
|
| 488 |
+
formatted_content_chunk = {
|
| 489 |
+
"id": response_json.get("id", ""),
|
| 490 |
+
"object": "chat.completion.chunk",
|
| 491 |
+
"created": response_json.get("created", int(time.time())),
|
| 492 |
+
"model": model_name,
|
| 493 |
+
"choices": [
|
| 494 |
+
{
|
| 495 |
+
"index": 0,
|
| 496 |
+
"delta": {
|
| 497 |
+
"content": content,
|
| 498 |
+
},
|
| 499 |
+
"finish_reason": None
|
| 500 |
+
}
|
| 501 |
+
],
|
| 502 |
+
"usage": None,
|
| 503 |
+
}
|
| 504 |
+
yield f"data: {json.dumps(formatted_content_chunk)}\n\n".encode('utf-8')
|
| 505 |
+
elif "choices" in response_json and len(response_json["choices"]) > 0:
|
| 506 |
+
# Handle other models normally
|
| 507 |
+
delta = response_json["choices"][0].get("delta", {})
|
| 508 |
+
if "content" in delta and delta["content"]:
|
| 509 |
+
formatted_content_chunk = {
|
| 510 |
+
"id": response_json.get("id", ""),
|
| 511 |
+
"object": "chat.completion.chunk",
|
| 512 |
+
"created": response_json.get("created", int(time.time())),
|
| 513 |
+
"model": model_name,
|
| 514 |
+
"choices": [
|
| 515 |
+
{
|
| 516 |
+
"index": 0,
|
| 517 |
+
"delta": {
|
| 518 |
+
"content": delta["content"],
|
| 519 |
+
},
|
| 520 |
+
"finish_reason": None
|
| 521 |
+
}
|
| 522 |
+
],
|
| 523 |
+
"usage": None,
|
| 524 |
+
}
|
| 525 |
+
yield f"data: {json.dumps(formatted_content_chunk)}\n\n".encode('utf-8')
|
| 526 |
+
|
| 527 |
+
if (
|
| 528 |
+
"usage" in response_json and
|
| 529 |
+
"prompt_tokens" in response_json["usage"]
|
| 530 |
+
):
|
| 531 |
+
prompt_tokens = response_json[
|
| 532 |
+
"usage"
|
| 533 |
+
]["prompt_tokens"]
|
| 534 |
+
|
| 535 |
+
except (
|
| 536 |
+
KeyError,
|
| 537 |
+
ValueError,
|
| 538 |
+
IndexError
|
| 539 |
+
) as e:
|
| 540 |
+
logging.error(
|
| 541 |
+
f"解析流式响应单行 JSON 失败: {e}, "
|
| 542 |
+
f"行内容: {line}"
|
| 543 |
+
)
|
| 544 |
+
except Exception as e:
|
| 545 |
+
logging.error(f"处理流式响应失败:{e}")
|
| 546 |
+
|
| 547 |
+
# Send the [DONE] message after all chunks have been processed
|
| 548 |
+
done_chunk = {
|
| 549 |
+
"id": response_json.get("id", ""),
|
| 550 |
+
"object": "chat.completion.chunk",
|
| 551 |
+
"created": response_json.get("created", int(time.time())),
|
| 552 |
+
"model": model_name,
|
| 553 |
+
"choices": [
|
| 554 |
+
{
|
| 555 |
+
"index": 0,
|
| 556 |
+
"delta": {},
|
| 557 |
+
"finish_reason": "stop"
|
| 558 |
+
}
|
| 559 |
+
],
|
| 560 |
+
"usage": {
|
| 561 |
+
"completion_tokens": completion_tokens,
|
| 562 |
+
"prompt_tokens": prompt_tokens,
|
| 563 |
+
"total_tokens": prompt_tokens + completion_tokens
|
| 564 |
+
},
|
| 565 |
+
}
|
| 566 |
+
yield f"data: {json.dumps(done_chunk)}\n\n".encode('utf-8')
|
| 567 |
|
| 568 |
end_time = time.time()
|
| 569 |
first_token_time = (
|
|
|
|
| 572 |
)
|
| 573 |
total_time = end_time - start_time
|
| 574 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 575 |
|
| 576 |
user_content = ""
|
| 577 |
messages = data.get("messages", [])
|
|
|
|
| 595 |
user_content_replaced = user_content.replace(
|
| 596 |
'\n', '\\n'
|
| 597 |
).replace('\r', '\\n')
|
| 598 |
+
|
| 599 |
logging.info(
|
| 600 |
f"使用的key: {api_key}, "
|
| 601 |
f"提示token: {prompt_tokens}, "
|
|
|
|
| 615 |
content_type=response.headers['Content-Type']
|
| 616 |
)
|
| 617 |
else:
|
| 618 |
+
# ... (Non-streaming part remains the same as in the previous response)
|
| 619 |
response.raise_for_status()
|
| 620 |
end_time = time.time()
|
| 621 |
response_json = response.json()
|
|
|
|
| 717 |
logging.error(f"请求转发异常: {e}")
|
| 718 |
return jsonify({"error": str(e)}), 500
|
| 719 |
|
| 720 |
+
|
| 721 |
if __name__ == '__main__':
|
| 722 |
logging.info(f"环境变量:{os.environ}")
|
| 723 |
|