Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -5,11 +5,8 @@ import requests
|
|
5 |
import json
|
6 |
import random
|
7 |
import uuid
|
8 |
-
import concurrent.
|
9 |
-
import base64
|
10 |
-
import io
|
11 |
import threading
|
12 |
-
from PIL import Imagetures
|
13 |
from datetime import datetime, timedelta
|
14 |
from apscheduler.schedulers.background import BackgroundScheduler
|
15 |
from flask import Flask, request, jsonify, Response, stream_with_context
|
@@ -562,10 +559,8 @@ def handsome_chat_completions():
|
|
562 |
|
563 |
request_type = determine_request_type(
|
564 |
model_name,
|
565 |
-
text_models,
|
566 |
-
free_text_models
|
567 |
-
image_models,
|
568 |
-
free_image_models
|
569 |
)
|
570 |
|
571 |
api_key = select_key(request_type, model_name)
|
@@ -580,44 +575,46 @@ def handsome_chat_completions():
|
|
580 |
)
|
581 |
}
|
582 |
), 429
|
583 |
-
|
584 |
headers = {
|
585 |
"Authorization": f"Bearer {api_key}",
|
586 |
"Content-Type": "application/json"
|
587 |
}
|
588 |
-
|
589 |
-
if model_name in image_models
|
590 |
-
# Handle image
|
591 |
-
|
592 |
-
messages = data.get("messages", [])
|
593 |
-
for message in messages:
|
594 |
-
if message["role"] == "user":
|
595 |
-
if isinstance(message["content"], str):
|
596 |
-
user_content += message["content"] + " "
|
597 |
-
elif isinstance(message["content"], list):
|
598 |
-
for item in message["content"]:
|
599 |
-
if (
|
600 |
-
isinstance(item, dict) and
|
601 |
-
item.get("type") == "text"
|
602 |
-
):
|
603 |
-
user_content += (
|
604 |
-
item.get("text", "") +
|
605 |
-
" "
|
606 |
-
)
|
607 |
-
user_content = user_content.strip()
|
608 |
-
|
609 |
siliconflow_data = {
|
610 |
"model": model_name,
|
611 |
-
"prompt":
|
612 |
-
"image_size": "1024x1024",
|
613 |
-
"batch_size": 1,
|
614 |
-
"num_inference_steps": 20,
|
615 |
-
"guidance_scale": 7.5,
|
616 |
-
"negative_prompt":
|
617 |
-
"seed":
|
618 |
"prompt_enhancement": False,
|
619 |
}
|
620 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
621 |
try:
|
622 |
start_time = time.time()
|
623 |
response = requests.post(
|
@@ -626,7 +623,7 @@ def handsome_chat_completions():
|
|
626 |
json=siliconflow_data,
|
627 |
timeout=120
|
628 |
)
|
629 |
-
|
630 |
if response.status_code == 429:
|
631 |
return jsonify(response.json()), 429
|
632 |
|
@@ -637,13 +634,32 @@ def handsome_chat_completions():
|
|
637 |
|
638 |
try:
|
639 |
images = response_json.get("images", [])
|
640 |
-
|
641 |
-
|
642 |
-
|
643 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
644 |
response_data = {
|
645 |
-
|
646 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
647 |
}
|
648 |
|
649 |
except (KeyError, ValueError, IndexError) as e:
|
@@ -652,15 +668,26 @@ def handsome_chat_completions():
|
|
652 |
f"完整内容: {response_json}"
|
653 |
)
|
654 |
response_data = {
|
655 |
-
|
656 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
657 |
}
|
658 |
|
659 |
logging.info(
|
660 |
f"使用的key: {api_key}, "
|
661 |
f"总共用时: {total_time:.4f}秒, "
|
662 |
-
f"使用的模型: {model_name}
|
663 |
-
f"用户的内容: {user_content}"
|
664 |
)
|
665 |
|
666 |
with data_lock:
|
@@ -668,13 +695,11 @@ def handsome_chat_completions():
|
|
668 |
token_counts.append(0) # Image generation doesn't use tokens
|
669 |
|
670 |
return jsonify(response_data)
|
671 |
-
|
672 |
except requests.exceptions.RequestException as e:
|
673 |
logging.error(f"请求转发异常: {e}")
|
674 |
return jsonify({"error": str(e)}), 500
|
675 |
-
|
676 |
else:
|
677 |
-
|
678 |
try:
|
679 |
start_time = time.time()
|
680 |
response = requests.post(
|
@@ -684,7 +709,6 @@ def handsome_chat_completions():
|
|
684 |
stream=data.get("stream", False),
|
685 |
timeout=60
|
686 |
)
|
687 |
-
|
688 |
if response.status_code == 429:
|
689 |
return jsonify(response.json()), 429
|
690 |
|
@@ -836,7 +860,8 @@ def handsome_chat_completions():
|
|
836 |
item.get("type") == "text"
|
837 |
):
|
838 |
user_content += (
|
839 |
-
item.get("text", "") +
|
|
|
840 |
)
|
841 |
|
842 |
user_content = user_content.strip()
|
@@ -878,7 +903,6 @@ def list_models():
|
|
878 |
|
879 |
detailed_models = []
|
880 |
|
881 |
-
# 添加文本模型
|
882 |
for model in text_models:
|
883 |
detailed_models.append({
|
884 |
"id": model,
|
@@ -905,7 +929,6 @@ def list_models():
|
|
905 |
"parent": None
|
906 |
})
|
907 |
|
908 |
-
# 添加 embedding 模型
|
909 |
for model in embedding_models:
|
910 |
detailed_models.append({
|
911 |
"id": model,
|
@@ -931,36 +954,9 @@ def list_models():
|
|
931 |
"root": model,
|
932 |
"parent": None
|
933 |
})
|
934 |
-
|
935 |
-
# 添加图像模型
|
936 |
-
for model in image_models:
|
937 |
-
detailed_models.append({
|
938 |
-
"id": model,
|
939 |
-
"object": "model",
|
940 |
-
"created": 1678888888,
|
941 |
-
"owned_by": "openai",
|
942 |
-
"permission": [
|
943 |
-
{
|
944 |
-
"id": f"modelperm-{uuid.uuid4().hex}",
|
945 |
-
"object": "model_permission",
|
946 |
-
"created": 1678888888,
|
947 |
-
"allow_create_engine": False,
|
948 |
-
"allow_sampling": True,
|
949 |
-
"allow_logprobs": False,
|
950 |
-
"allow_search_indices": False,
|
951 |
-
"allow_view": True,
|
952 |
-
"allow_fine_tuning": False,
|
953 |
-
"organization": "*",
|
954 |
-
"group": None,
|
955 |
-
"is_blocking": False
|
956 |
-
}
|
957 |
-
],
|
958 |
-
"root": model,
|
959 |
-
"parent": None
|
960 |
-
})
|
961 |
|
962 |
-
for model in
|
963 |
-
|
964 |
"id": model,
|
965 |
"object": "model",
|
966 |
"created": 1678888888,
|
@@ -972,7 +968,7 @@ def list_models():
|
|
972 |
"created": 1678888888,
|
973 |
"allow_create_engine": False,
|
974 |
"allow_sampling": True,
|
975 |
-
"allow_logprobs":
|
976 |
"allow_search_indices": False,
|
977 |
"allow_view": True,
|
978 |
"allow_fine_tuning": False,
|
@@ -1154,6 +1150,10 @@ def handsome_embeddings():
|
|
1154 |
except requests.exceptions.RequestException as e:
|
1155 |
return jsonify({"error": str(e)}), 500
|
1156 |
|
|
|
|
|
|
|
|
|
1157 |
@app.route('/handsome/v1/images/generations', methods=['POST'])
|
1158 |
def handsome_images_generations():
|
1159 |
if not check_authorization(request):
|
|
|
5 |
import json
|
6 |
import random
|
7 |
import uuid
|
8 |
+
import concurrent.futures
|
|
|
|
|
9 |
import threading
|
|
|
10 |
from datetime import datetime, timedelta
|
11 |
from apscheduler.schedulers.background import BackgroundScheduler
|
12 |
from flask import Flask, request, jsonify, Response, stream_with_context
|
|
|
559 |
|
560 |
request_type = determine_request_type(
|
561 |
model_name,
|
562 |
+
text_models + image_models,
|
563 |
+
free_text_models + free_image_models
|
|
|
|
|
564 |
)
|
565 |
|
566 |
api_key = select_key(request_type, model_name)
|
|
|
575 |
)
|
576 |
}
|
577 |
), 429
|
578 |
+
|
579 |
headers = {
|
580 |
"Authorization": f"Bearer {api_key}",
|
581 |
"Content-Type": "application/json"
|
582 |
}
|
583 |
+
|
584 |
+
if model_name in image_models:
|
585 |
+
# Handle image generation
|
586 |
+
# Map OpenAI-style parameters to SiliconFlow's parameters
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
587 |
siliconflow_data = {
|
588 |
"model": model_name,
|
589 |
+
"prompt": data.get("messages", [{}])[0].get("content") if isinstance(data.get("messages"), list) else "",
|
590 |
+
"image_size": data.get("size", "1024x1024"),
|
591 |
+
"batch_size": data.get("n", 1),
|
592 |
+
"num_inference_steps": data.get("steps", 20),
|
593 |
+
"guidance_scale": data.get("guidance_scale", 7.5),
|
594 |
+
"negative_prompt": data.get("negative_prompt"),
|
595 |
+
"seed": data.get("seed"),
|
596 |
"prompt_enhancement": False,
|
597 |
}
|
598 |
|
599 |
+
# Parameter validation and adjustments
|
600 |
+
if siliconflow_data["batch_size"] < 1:
|
601 |
+
siliconflow_data["batch_size"] = 1
|
602 |
+
if siliconflow_data["batch_size"] > 4:
|
603 |
+
siliconflow_data["batch_size"] = 4
|
604 |
+
|
605 |
+
if siliconflow_data["num_inference_steps"] < 1:
|
606 |
+
siliconflow_data["num_inference_steps"] = 1
|
607 |
+
if siliconflow_data["num_inference_steps"] > 50:
|
608 |
+
siliconflow_data["num_inference_steps"] = 50
|
609 |
+
|
610 |
+
if siliconflow_data["guidance_scale"] < 0:
|
611 |
+
siliconflow_data["guidance_scale"] = 0
|
612 |
+
if siliconflow_data["guidance_scale"] > 100:
|
613 |
+
siliconflow_data["guidance_scale"] = 100
|
614 |
+
|
615 |
+
if siliconflow_data["image_size"] not in ["1024x1024", "512x1024", "768x512", "768x1024", "1024x576", "576x1024"]:
|
616 |
+
siliconflow_data["image_size"] = "1024x1024"
|
617 |
+
|
618 |
try:
|
619 |
start_time = time.time()
|
620 |
response = requests.post(
|
|
|
623 |
json=siliconflow_data,
|
624 |
timeout=120
|
625 |
)
|
626 |
+
|
627 |
if response.status_code == 429:
|
628 |
return jsonify(response.json()), 429
|
629 |
|
|
|
634 |
|
635 |
try:
|
636 |
images = response_json.get("images", [])
|
637 |
+
|
638 |
+
# Extract the first URL if available
|
639 |
+
image_url = ""
|
640 |
+
if images and isinstance(images[0], dict) and "url" in images[0]:
|
641 |
+
image_url = images[0]["url"]
|
642 |
+
logging.info(f"Extracted image URL: {image_url}")
|
643 |
+
elif images and isinstance(images[0], str):
|
644 |
+
image_url = images[0]
|
645 |
+
logging.info(f"Extracted image URL: {image_url}")
|
646 |
+
|
647 |
+
# Construct the expected JSON output - Mimicking OpenAI
|
648 |
response_data = {
|
649 |
+
"id": f"chatcmpl-{uuid.uuid4()}",
|
650 |
+
"object": "chat.completion",
|
651 |
+
"created": int(time.time()),
|
652 |
+
"model": model_name,
|
653 |
+
"choices": [
|
654 |
+
{
|
655 |
+
"index": 0,
|
656 |
+
"message": {
|
657 |
+
"role": "assistant",
|
658 |
+
"content": image_url if image_url else "Failed to generate image", # Directly return the URL in content
|
659 |
+
},
|
660 |
+
"finish_reason": "stop",
|
661 |
+
}
|
662 |
+
],
|
663 |
}
|
664 |
|
665 |
except (KeyError, ValueError, IndexError) as e:
|
|
|
668 |
f"完整内容: {response_json}"
|
669 |
)
|
670 |
response_data = {
|
671 |
+
"id": f"chatcmpl-{uuid.uuid4()}",
|
672 |
+
"object": "chat.completion",
|
673 |
+
"created": int(time.time()),
|
674 |
+
"model": model_name,
|
675 |
+
"choices": [
|
676 |
+
{
|
677 |
+
"index": 0,
|
678 |
+
"message": {
|
679 |
+
"role": "assistant",
|
680 |
+
"content": "Failed to process image data",
|
681 |
+
},
|
682 |
+
"finish_reason": "stop",
|
683 |
+
}
|
684 |
+
],
|
685 |
}
|
686 |
|
687 |
logging.info(
|
688 |
f"使用的key: {api_key}, "
|
689 |
f"总共用时: {total_time:.4f}秒, "
|
690 |
+
f"使用的模型: {model_name}"
|
|
|
691 |
)
|
692 |
|
693 |
with data_lock:
|
|
|
695 |
token_counts.append(0) # Image generation doesn't use tokens
|
696 |
|
697 |
return jsonify(response_data)
|
|
|
698 |
except requests.exceptions.RequestException as e:
|
699 |
logging.error(f"请求转发异常: {e}")
|
700 |
return jsonify({"error": str(e)}), 500
|
|
|
701 |
else:
|
702 |
+
# Existing text-based model handling logic
|
703 |
try:
|
704 |
start_time = time.time()
|
705 |
response = requests.post(
|
|
|
709 |
stream=data.get("stream", False),
|
710 |
timeout=60
|
711 |
)
|
|
|
712 |
if response.status_code == 429:
|
713 |
return jsonify(response.json()), 429
|
714 |
|
|
|
860 |
item.get("type") == "text"
|
861 |
):
|
862 |
user_content += (
|
863 |
+
item.get("text", "") +
|
864 |
+
" "
|
865 |
)
|
866 |
|
867 |
user_content = user_content.strip()
|
|
|
903 |
|
904 |
detailed_models = []
|
905 |
|
|
|
906 |
for model in text_models:
|
907 |
detailed_models.append({
|
908 |
"id": model,
|
|
|
929 |
"parent": None
|
930 |
})
|
931 |
|
|
|
932 |
for model in embedding_models:
|
933 |
detailed_models.append({
|
934 |
"id": model,
|
|
|
954 |
"root": model,
|
955 |
"parent": None
|
956 |
})
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
957 |
|
958 |
+
for model in image_models:
|
959 |
+
detailed_models.append({
|
960 |
"id": model,
|
961 |
"object": "model",
|
962 |
"created": 1678888888,
|
|
|
968 |
"created": 1678888888,
|
969 |
"allow_create_engine": False,
|
970 |
"allow_sampling": True,
|
971 |
+
"allow_logprobs": True,
|
972 |
"allow_search_indices": False,
|
973 |
"allow_view": True,
|
974 |
"allow_fine_tuning": False,
|
|
|
1150 |
except requests.exceptions.RequestException as e:
|
1151 |
return jsonify({"error": str(e)}), 500
|
1152 |
|
1153 |
+
import base64
|
1154 |
+
import io
|
1155 |
+
from PIL import Image
|
1156 |
+
|
1157 |
@app.route('/handsome/v1/images/generations', methods=['POST'])
|
1158 |
def handsome_images_generations():
|
1159 |
if not check_authorization(request):
|