LiYongHui
yonghui li
commited on
Commit
·
29b8637
1
Parent(s):
69e5f04
Integrates LLM Azure OpenAI (#1318)
Browse files### What problem does this PR solve?
feat: Integrates LLM Azure OpenAI #716
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### Other
It's just the back-end code, the front-end needs to provide the Azure
OpenAI model addition form.
#### Required parameters
- base_url
- api_key
---------
Co-authored-by: yonghui li <[email protected]>
- api/db/init_data.py +82 -0
- api/settings.py +6 -0
- rag/llm/__init__.py +3 -0
- rag/llm/chat_model.py +6 -0
- rag/llm/cv_model.py +20 -0
- rag/llm/embedding_model.py +6 -0
api/db/init_data.py
CHANGED
@@ -165,6 +165,11 @@ factory_infos = [{
|
|
165 |
"logo": "",
|
166 |
"tags": "LLM,TEXT EMBEDDING",
|
167 |
"status": "1",
|
|
|
|
|
|
|
|
|
|
|
168 |
}
|
169 |
# {
|
170 |
# "name": "文心一言",
|
@@ -649,6 +654,83 @@ def init_llm_factory():
|
|
649 |
"max_tokens": 8192,
|
650 |
"model_type": LLMType.EMBEDDING
|
651 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
652 |
]
|
653 |
for info in factory_infos:
|
654 |
try:
|
|
|
165 |
"logo": "",
|
166 |
"tags": "LLM,TEXT EMBEDDING",
|
167 |
"status": "1",
|
168 |
+
},{
|
169 |
+
"name": "Azure-OpenAI",
|
170 |
+
"logo": "",
|
171 |
+
"tags": "LLM,TEXT EMBEDDING,SPEECH2TEXT,MODERATION",
|
172 |
+
"status": "1",
|
173 |
}
|
174 |
# {
|
175 |
# "name": "文心一言",
|
|
|
654 |
"max_tokens": 8192,
|
655 |
"model_type": LLMType.EMBEDDING
|
656 |
},
|
657 |
+
# ------------------------ Azure OpenAI -----------------------
|
658 |
+
# Please ensure the llm_name is the same as the name in Azure
|
659 |
+
# OpenAI deployment name (e.g., azure-gpt-4o). And the llm_name
|
660 |
+
# must different from the OpenAI llm_name
|
661 |
+
#
|
662 |
+
# Each model must be deployed in the Azure OpenAI service, otherwise,
|
663 |
+
# you will receive an error message 'The API deployment for
|
664 |
+
# this resource does not exist'
|
665 |
+
{
|
666 |
+
"fid": factory_infos[15]["name"],
|
667 |
+
"llm_name": "azure-gpt-4o",
|
668 |
+
"tags": "LLM,CHAT,128K",
|
669 |
+
"max_tokens": 128000,
|
670 |
+
"model_type": LLMType.CHAT.value + "," + LLMType.IMAGE2TEXT.value
|
671 |
+
}, {
|
672 |
+
"fid": factory_infos[15]["name"],
|
673 |
+
"llm_name": "azure-gpt-35-turbo",
|
674 |
+
"tags": "LLM,CHAT,4K",
|
675 |
+
"max_tokens": 4096,
|
676 |
+
"model_type": LLMType.CHAT.value
|
677 |
+
}, {
|
678 |
+
"fid": factory_infos[15]["name"],
|
679 |
+
"llm_name": "azure-gpt-35-turbo-16k",
|
680 |
+
"tags": "LLM,CHAT,16k",
|
681 |
+
"max_tokens": 16385,
|
682 |
+
"model_type": LLMType.CHAT.value
|
683 |
+
}, {
|
684 |
+
"fid": factory_infos[15]["name"],
|
685 |
+
"llm_name": "azure-text-embedding-ada-002",
|
686 |
+
"tags": "TEXT EMBEDDING,8K",
|
687 |
+
"max_tokens": 8191,
|
688 |
+
"model_type": LLMType.EMBEDDING.value
|
689 |
+
}, {
|
690 |
+
"fid": factory_infos[15]["name"],
|
691 |
+
"llm_name": "azure-text-embedding-3-small",
|
692 |
+
"tags": "TEXT EMBEDDING,8K",
|
693 |
+
"max_tokens": 8191,
|
694 |
+
"model_type": LLMType.EMBEDDING.value
|
695 |
+
}, {
|
696 |
+
"fid": factory_infos[15]["name"],
|
697 |
+
"llm_name": "azure-text-embedding-3-large",
|
698 |
+
"tags": "TEXT EMBEDDING,8K",
|
699 |
+
"max_tokens": 8191,
|
700 |
+
"model_type": LLMType.EMBEDDING.value
|
701 |
+
},{
|
702 |
+
"fid": factory_infos[15]["name"],
|
703 |
+
"llm_name": "azure-whisper-1",
|
704 |
+
"tags": "SPEECH2TEXT",
|
705 |
+
"max_tokens": 25 * 1024 * 1024,
|
706 |
+
"model_type": LLMType.SPEECH2TEXT.value
|
707 |
+
},
|
708 |
+
{
|
709 |
+
"fid": factory_infos[15]["name"],
|
710 |
+
"llm_name": "azure-gpt-4",
|
711 |
+
"tags": "LLM,CHAT,8K",
|
712 |
+
"max_tokens": 8191,
|
713 |
+
"model_type": LLMType.CHAT.value
|
714 |
+
}, {
|
715 |
+
"fid": factory_infos[15]["name"],
|
716 |
+
"llm_name": "azure-gpt-4-turbo",
|
717 |
+
"tags": "LLM,CHAT,8K",
|
718 |
+
"max_tokens": 8191,
|
719 |
+
"model_type": LLMType.CHAT.value
|
720 |
+
}, {
|
721 |
+
"fid": factory_infos[15]["name"],
|
722 |
+
"llm_name": "azure-gpt-4-32k",
|
723 |
+
"tags": "LLM,CHAT,32K",
|
724 |
+
"max_tokens": 32768,
|
725 |
+
"model_type": LLMType.CHAT.value
|
726 |
+
}, {
|
727 |
+
"fid": factory_infos[15]["name"],
|
728 |
+
"llm_name": "azure-gpt-4-vision-preview",
|
729 |
+
"tags": "LLM,CHAT,IMAGE2TEXT",
|
730 |
+
"max_tokens": 765,
|
731 |
+
"model_type": LLMType.IMAGE2TEXT.value
|
732 |
+
},
|
733 |
+
|
734 |
]
|
735 |
for info in factory_infos:
|
736 |
try:
|
api/settings.py
CHANGED
@@ -69,6 +69,12 @@ default_llm = {
|
|
69 |
"image2text_model": "gpt-4-vision-preview",
|
70 |
"asr_model": "whisper-1",
|
71 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
72 |
"ZHIPU-AI": {
|
73 |
"chat_model": "glm-3-turbo",
|
74 |
"embedding_model": "embedding-2",
|
|
|
69 |
"image2text_model": "gpt-4-vision-preview",
|
70 |
"asr_model": "whisper-1",
|
71 |
},
|
72 |
+
"Azure-OpenAI": {
|
73 |
+
"chat_model": "azure-gpt-35-turbo",
|
74 |
+
"embedding_model": "azure-text-embedding-ada-002",
|
75 |
+
"image2text_model": "azure-gpt-4-vision-preview",
|
76 |
+
"asr_model": "azure-whisper-1",
|
77 |
+
},
|
78 |
"ZHIPU-AI": {
|
79 |
"chat_model": "glm-3-turbo",
|
80 |
"embedding_model": "embedding-2",
|
rag/llm/__init__.py
CHANGED
@@ -22,6 +22,7 @@ from .rerank_model import *
|
|
22 |
EmbeddingModel = {
|
23 |
"Ollama": OllamaEmbed,
|
24 |
"OpenAI": OpenAIEmbed,
|
|
|
25 |
"Xinference": XinferenceEmbed,
|
26 |
"Tongyi-Qianwen": QWenEmbed,
|
27 |
"ZHIPU-AI": ZhipuEmbed,
|
@@ -36,6 +37,7 @@ EmbeddingModel = {
|
|
36 |
|
37 |
CvModel = {
|
38 |
"OpenAI": GptV4,
|
|
|
39 |
"Ollama": OllamaCV,
|
40 |
"Xinference": XinferenceCV,
|
41 |
"Tongyi-Qianwen": QWenCV,
|
@@ -46,6 +48,7 @@ CvModel = {
|
|
46 |
|
47 |
ChatModel = {
|
48 |
"OpenAI": GptTurbo,
|
|
|
49 |
"ZHIPU-AI": ZhipuChat,
|
50 |
"Tongyi-Qianwen": QWenChat,
|
51 |
"Ollama": OllamaChat,
|
|
|
22 |
EmbeddingModel = {
|
23 |
"Ollama": OllamaEmbed,
|
24 |
"OpenAI": OpenAIEmbed,
|
25 |
+
"Azure-OpenAI": AzureEmbed,
|
26 |
"Xinference": XinferenceEmbed,
|
27 |
"Tongyi-Qianwen": QWenEmbed,
|
28 |
"ZHIPU-AI": ZhipuEmbed,
|
|
|
37 |
|
38 |
CvModel = {
|
39 |
"OpenAI": GptV4,
|
40 |
+
"Azure-OpenAI": AzureGptV4,
|
41 |
"Ollama": OllamaCV,
|
42 |
"Xinference": XinferenceCV,
|
43 |
"Tongyi-Qianwen": QWenCV,
|
|
|
48 |
|
49 |
ChatModel = {
|
50 |
"OpenAI": GptTurbo,
|
51 |
+
"Azure-OpenAI": AzureChat,
|
52 |
"ZHIPU-AI": ZhipuChat,
|
53 |
"Tongyi-Qianwen": QWenChat,
|
54 |
"Ollama": OllamaChat,
|
rag/llm/chat_model.py
CHANGED
@@ -13,6 +13,7 @@
|
|
13 |
# See the License for the specific language governing permissions and
|
14 |
# limitations under the License.
|
15 |
#
|
|
|
16 |
from zhipuai import ZhipuAI
|
17 |
from dashscope import Generation
|
18 |
from abc import ABC
|
@@ -94,6 +95,11 @@ class DeepSeekChat(Base):
|
|
94 |
if not base_url: base_url="https://api.deepseek.com/v1"
|
95 |
super().__init__(key, model_name, base_url)
|
96 |
|
|
|
|
|
|
|
|
|
|
|
97 |
|
98 |
class BaiChuanChat(Base):
|
99 |
def __init__(self, key, model_name="Baichuan3-Turbo", base_url="https://api.baichuan-ai.com/v1"):
|
|
|
13 |
# See the License for the specific language governing permissions and
|
14 |
# limitations under the License.
|
15 |
#
|
16 |
+
from openai.lib.azure import AzureOpenAI
|
17 |
from zhipuai import ZhipuAI
|
18 |
from dashscope import Generation
|
19 |
from abc import ABC
|
|
|
95 |
if not base_url: base_url="https://api.deepseek.com/v1"
|
96 |
super().__init__(key, model_name, base_url)
|
97 |
|
98 |
+
class AzureChat(Base):
|
99 |
+
def __init__(self, key, model_name, **kwargs):
|
100 |
+
self.client = AzureOpenAI(api_key=key, azure_endpoint=kwargs["base_url"], api_version="2024-02-01")
|
101 |
+
self.model_name = model_name
|
102 |
+
|
103 |
|
104 |
class BaiChuanChat(Base):
|
105 |
def __init__(self, key, model_name="Baichuan3-Turbo", base_url="https://api.baichuan-ai.com/v1"):
|
rag/llm/cv_model.py
CHANGED
@@ -13,6 +13,7 @@
|
|
13 |
# See the License for the specific language governing permissions and
|
14 |
# limitations under the License.
|
15 |
#
|
|
|
16 |
from zhipuai import ZhipuAI
|
17 |
import io
|
18 |
from abc import ABC
|
@@ -87,6 +88,25 @@ class GptV4(Base):
|
|
87 |
)
|
88 |
return res.choices[0].message.content.strip(), res.usage.total_tokens
|
89 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
90 |
|
91 |
class QWenCV(Base):
|
92 |
def __init__(self, key, model_name="qwen-vl-chat-v1", lang="Chinese", **kwargs):
|
|
|
13 |
# See the License for the specific language governing permissions and
|
14 |
# limitations under the License.
|
15 |
#
|
16 |
+
from openai.lib.azure import AzureOpenAI
|
17 |
from zhipuai import ZhipuAI
|
18 |
import io
|
19 |
from abc import ABC
|
|
|
88 |
)
|
89 |
return res.choices[0].message.content.strip(), res.usage.total_tokens
|
90 |
|
91 |
+
class AzureGptV4(Base):
|
92 |
+
def __init__(self, key, model_name, lang="Chinese", **kwargs):
|
93 |
+
self.client = AzureOpenAI(api_key=key, azure_endpoint=kwargs["base_url"], api_version="2024-02-01")
|
94 |
+
self.model_name = model_name
|
95 |
+
self.lang = lang
|
96 |
+
|
97 |
+
def describe(self, image, max_tokens=300):
|
98 |
+
b64 = self.image2base64(image)
|
99 |
+
prompt = self.prompt(b64)
|
100 |
+
for i in range(len(prompt)):
|
101 |
+
for c in prompt[i]["content"]:
|
102 |
+
if "text" in c: c["type"] = "text"
|
103 |
+
|
104 |
+
res = self.client.chat.completions.create(
|
105 |
+
model=self.model_name,
|
106 |
+
messages=prompt,
|
107 |
+
max_tokens=max_tokens,
|
108 |
+
)
|
109 |
+
return res.choices[0].message.content.strip(), res.usage.total_tokens
|
110 |
|
111 |
class QWenCV(Base):
|
112 |
def __init__(self, key, model_name="qwen-vl-chat-v1", lang="Chinese", **kwargs):
|
rag/llm/embedding_model.py
CHANGED
@@ -18,6 +18,7 @@ from typing import Optional
|
|
18 |
import threading
|
19 |
import requests
|
20 |
from huggingface_hub import snapshot_download
|
|
|
21 |
from zhipuai import ZhipuAI
|
22 |
import os
|
23 |
from abc import ABC
|
@@ -110,6 +111,11 @@ class OpenAIEmbed(Base):
|
|
110 |
return np.array(res.data[0].embedding), res.usage.total_tokens
|
111 |
|
112 |
|
|
|
|
|
|
|
|
|
|
|
113 |
class BaiChuanEmbed(OpenAIEmbed):
|
114 |
def __init__(self, key,
|
115 |
model_name='Baichuan-Text-Embedding',
|
|
|
18 |
import threading
|
19 |
import requests
|
20 |
from huggingface_hub import snapshot_download
|
21 |
+
from openai.lib.azure import AzureOpenAI
|
22 |
from zhipuai import ZhipuAI
|
23 |
import os
|
24 |
from abc import ABC
|
|
|
111 |
return np.array(res.data[0].embedding), res.usage.total_tokens
|
112 |
|
113 |
|
114 |
+
class AzureEmbed(Base):
|
115 |
+
def __init__(self, key, model_name, **kwargs):
|
116 |
+
self.client = AzureOpenAI(api_key=key, azure_endpoint=kwargs["base_url"], api_version="2024-02-01")
|
117 |
+
self.model_name = model_name
|
118 |
+
|
119 |
class BaiChuanEmbed(OpenAIEmbed):
|
120 |
def __init__(self, key,
|
121 |
model_name='Baichuan-Text-Embedding',
|