H
commited on
Commit
路
afe703e
1
Parent(s):
f515d5f
Chat Use CVmodel (#1607)
Browse files### What problem does this PR solve?
#1230
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
api/db/services/dialog_service.py
CHANGED
@@ -13,6 +13,8 @@
|
|
13 |
# See the License for the specific language governing permissions and
|
14 |
# limitations under the License.
|
15 |
#
|
|
|
|
|
16 |
import re
|
17 |
from copy import deepcopy
|
18 |
|
@@ -26,6 +28,7 @@ from rag.app.resume import forbidden_select_fields4resume
|
|
26 |
from rag.nlp import keyword_extraction
|
27 |
from rag.nlp.search import index_name
|
28 |
from rag.utils import rmSpace, num_tokens_from_string, encoder
|
|
|
29 |
|
30 |
|
31 |
class DialogService(CommonService):
|
@@ -73,6 +76,15 @@ def message_fit_in(msg, max_length=4000):
|
|
73 |
return max_length, msg
|
74 |
|
75 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
76 |
def chat(dialog, messages, stream=True, **kwargs):
|
77 |
assert messages[-1]["role"] == "user", "The last content of this conversation is not from user."
|
78 |
llm = LLMService.query(llm_name=dialog.llm_id)
|
@@ -91,7 +103,10 @@ def chat(dialog, messages, stream=True, **kwargs):
|
|
91 |
|
92 |
questions = [m["content"] for m in messages if m["role"] == "user"]
|
93 |
embd_mdl = LLMBundle(dialog.tenant_id, LLMType.EMBEDDING, embd_nms[0])
|
94 |
-
|
|
|
|
|
|
|
95 |
|
96 |
prompt_config = dialog.prompt_config
|
97 |
field_map = KnowledgebaseService.get_field_map(dialog.kb_ids)
|
@@ -328,7 +343,10 @@ def use_sql(question, field_map, tenant_id, chat_mdl, quota=True):
|
|
328 |
|
329 |
|
330 |
def relevant(tenant_id, llm_id, question, contents: list):
|
331 |
-
|
|
|
|
|
|
|
332 |
prompt = """
|
333 |
You are a grader assessing relevance of a retrieved document to a user question.
|
334 |
It does not need to be a stringent test. The goal is to filter out erroneous retrievals.
|
@@ -347,7 +365,10 @@ def relevant(tenant_id, llm_id, question, contents: list):
|
|
347 |
|
348 |
|
349 |
def rewrite(tenant_id, llm_id, question):
|
350 |
-
|
|
|
|
|
|
|
351 |
prompt = """
|
352 |
You are an expert at query expansion to generate a paraphrasing of a question.
|
353 |
I can't retrieval relevant information from the knowledge base by using user's question directly.
|
|
|
13 |
# See the License for the specific language governing permissions and
|
14 |
# limitations under the License.
|
15 |
#
|
16 |
+
import os
|
17 |
+
import json
|
18 |
import re
|
19 |
from copy import deepcopy
|
20 |
|
|
|
28 |
from rag.nlp import keyword_extraction
|
29 |
from rag.nlp.search import index_name
|
30 |
from rag.utils import rmSpace, num_tokens_from_string, encoder
|
31 |
+
from api.utils.file_utils import get_project_base_directory
|
32 |
|
33 |
|
34 |
class DialogService(CommonService):
|
|
|
76 |
return max_length, msg
|
77 |
|
78 |
|
79 |
+
def llm_id2llm_type(llm_id):
|
80 |
+
fnm = os.path.join(get_project_base_directory(), "conf")
|
81 |
+
llm_factories = json.load(open(os.path.join(fnm, "llm_factories.json"), "r"))
|
82 |
+
for llm_factory in llm_factories["factory_llm_infos"]:
|
83 |
+
for llm in llm_factory["llm"]:
|
84 |
+
if llm_id == llm["llm_name"]:
|
85 |
+
return llm["model_type"].strip(",")[-1]
|
86 |
+
|
87 |
+
|
88 |
def chat(dialog, messages, stream=True, **kwargs):
|
89 |
assert messages[-1]["role"] == "user", "The last content of this conversation is not from user."
|
90 |
llm = LLMService.query(llm_name=dialog.llm_id)
|
|
|
103 |
|
104 |
questions = [m["content"] for m in messages if m["role"] == "user"]
|
105 |
embd_mdl = LLMBundle(dialog.tenant_id, LLMType.EMBEDDING, embd_nms[0])
|
106 |
+
if llm_id2llm_type(dialog.llm_id) == "image2text":
|
107 |
+
chat_mdl = LLMBundle(dialog.tenant_id, LLMType.IMAGE2TEXT, dialog.llm_id)
|
108 |
+
else:
|
109 |
+
chat_mdl = LLMBundle(dialog.tenant_id, LLMType.CHAT, dialog.llm_id)
|
110 |
|
111 |
prompt_config = dialog.prompt_config
|
112 |
field_map = KnowledgebaseService.get_field_map(dialog.kb_ids)
|
|
|
343 |
|
344 |
|
345 |
def relevant(tenant_id, llm_id, question, contents: list):
|
346 |
+
if llm_id2llm_type(llm_id) == "image2text":
|
347 |
+
chat_mdl = LLMBundle(tenant_id, LLMType.IMAGE2TEXT, llm_id)
|
348 |
+
else:
|
349 |
+
chat_mdl = LLMBundle(tenant_id, LLMType.CHAT, llm_id)
|
350 |
prompt = """
|
351 |
You are a grader assessing relevance of a retrieved document to a user question.
|
352 |
It does not need to be a stringent test. The goal is to filter out erroneous retrievals.
|
|
|
365 |
|
366 |
|
367 |
def rewrite(tenant_id, llm_id, question):
|
368 |
+
if llm_id2llm_type(llm_id) == "image2text":
|
369 |
+
chat_mdl = LLMBundle(tenant_id, LLMType.IMAGE2TEXT, llm_id)
|
370 |
+
else:
|
371 |
+
chat_mdl = LLMBundle(tenant_id, LLMType.CHAT, llm_id)
|
372 |
prompt = """
|
373 |
You are an expert at query expansion to generate a paraphrasing of a question.
|
374 |
I can't retrieval relevant information from the knowledge base by using user's question directly.
|
api/db/services/llm_service.py
CHANGED
@@ -70,7 +70,7 @@ class TenantLLMService(CommonService):
|
|
70 |
elif llm_type == LLMType.SPEECH2TEXT.value:
|
71 |
mdlnm = tenant.asr_id
|
72 |
elif llm_type == LLMType.IMAGE2TEXT.value:
|
73 |
-
mdlnm = tenant.img2txt_id
|
74 |
elif llm_type == LLMType.CHAT.value:
|
75 |
mdlnm = tenant.llm_id if not llm_name else llm_name
|
76 |
elif llm_type == LLMType.RERANK:
|
|
|
70 |
elif llm_type == LLMType.SPEECH2TEXT.value:
|
71 |
mdlnm = tenant.asr_id
|
72 |
elif llm_type == LLMType.IMAGE2TEXT.value:
|
73 |
+
mdlnm = tenant.img2txt_id if not llm_name else llm_name
|
74 |
elif llm_type == LLMType.CHAT.value:
|
75 |
mdlnm = tenant.llm_id if not llm_name else llm_name
|
76 |
elif llm_type == LLMType.RERANK:
|
rag/llm/cv_model.py
CHANGED
@@ -26,6 +26,7 @@ from io import BytesIO
|
|
26 |
import json
|
27 |
import requests
|
28 |
|
|
|
29 |
from api.utils import get_uuid
|
30 |
from api.utils.file_utils import get_project_base_directory
|
31 |
|
@@ -36,7 +37,60 @@ class Base(ABC):
|
|
36 |
|
37 |
def describe(self, image, max_tokens=300):
|
38 |
raise NotImplementedError("Please implement encode method!")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
def image2base64(self, image):
|
41 |
if isinstance(image, bytes):
|
42 |
return base64.b64encode(image).decode("utf-8")
|
@@ -68,6 +122,21 @@ class Base(ABC):
|
|
68 |
}
|
69 |
]
|
70 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
71 |
|
72 |
class GptV4(Base):
|
73 |
def __init__(self, key, model_name="gpt-4-vision-preview", lang="Chinese", base_url="https://api.openai.com/v1"):
|
@@ -140,6 +209,12 @@ class QWenCV(Base):
|
|
140 |
}
|
141 |
]
|
142 |
|
|
|
|
|
|
|
|
|
|
|
|
|
143 |
def describe(self, image, max_tokens=300):
|
144 |
from http import HTTPStatus
|
145 |
from dashscope import MultiModalConversation
|
@@ -149,6 +224,66 @@ class QWenCV(Base):
|
|
149 |
return response.output.choices[0]['message']['content'][0]["text"], response.usage.output_tokens
|
150 |
return response.message, 0
|
151 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
152 |
|
153 |
class Zhipu4V(Base):
|
154 |
def __init__(self, key, model_name="glm-4v", lang="Chinese", **kwargs):
|
@@ -166,6 +301,59 @@ class Zhipu4V(Base):
|
|
166 |
)
|
167 |
return res.choices[0].message.content.strip(), res.usage.total_tokens
|
168 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
169 |
|
170 |
class OllamaCV(Base):
|
171 |
def __init__(self, key, model_name, lang="Chinese", **kwargs):
|
@@ -188,6 +376,63 @@ class OllamaCV(Base):
|
|
188 |
except Exception as e:
|
189 |
return "**ERROR**: " + str(e), 0
|
190 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
191 |
|
192 |
class LocalAICV(Base):
|
193 |
def __init__(self, key, model_name, base_url, lang="Chinese"):
|
@@ -236,7 +481,7 @@ class XinferenceCV(Base):
|
|
236 |
|
237 |
class GeminiCV(Base):
|
238 |
def __init__(self, key, model_name="gemini-1.0-pro-vision-latest", lang="Chinese", **kwargs):
|
239 |
-
from google.generativeai import client,GenerativeModel
|
240 |
client.configure(api_key=key)
|
241 |
_client = client.get_default_generative_client()
|
242 |
self.model_name = model_name
|
@@ -258,6 +503,59 @@ class GeminiCV(Base):
|
|
258 |
)
|
259 |
return res.text,res.usage_metadata.total_token_count
|
260 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
261 |
|
262 |
class OpenRouterCV(Base):
|
263 |
def __init__(
|
|
|
26 |
import json
|
27 |
import requests
|
28 |
|
29 |
+
from rag.nlp import is_english
|
30 |
from api.utils import get_uuid
|
31 |
from api.utils.file_utils import get_project_base_directory
|
32 |
|
|
|
37 |
|
38 |
def describe(self, image, max_tokens=300):
|
39 |
raise NotImplementedError("Please implement encode method!")
|
40 |
+
|
41 |
+
def chat(self, system, history, gen_conf, image=""):
|
42 |
+
if system:
|
43 |
+
history[-1]["content"] = system + history[-1]["content"] + "user query: " + history[-1]["content"]
|
44 |
+
try:
|
45 |
+
for his in history:
|
46 |
+
if his["role"] == "user":
|
47 |
+
his["content"] = self.chat_prompt(his["content"], image)
|
48 |
+
|
49 |
+
response = self.client.chat.completions.create(
|
50 |
+
model=self.model_name,
|
51 |
+
messages=history,
|
52 |
+
max_tokens=gen_conf.get("max_tokens", 1000),
|
53 |
+
temperature=gen_conf.get("temperature", 0.3),
|
54 |
+
top_p=gen_conf.get("top_p", 0.7)
|
55 |
+
)
|
56 |
+
return response.choices[0].message.content.strip(), response.usage.total_tokens
|
57 |
+
except Exception as e:
|
58 |
+
return "**ERROR**: " + str(e), 0
|
59 |
+
|
60 |
+
def chat_streamly(self, system, history, gen_conf, image=""):
|
61 |
+
if system:
|
62 |
+
history[-1]["content"] = system + history[-1]["content"] + "user query: " + history[-1]["content"]
|
63 |
+
|
64 |
+
ans = ""
|
65 |
+
tk_count = 0
|
66 |
+
try:
|
67 |
+
for his in history:
|
68 |
+
if his["role"] == "user":
|
69 |
+
his["content"] = self.chat_prompt(his["content"], image)
|
70 |
|
71 |
+
response = self.client.chat.completions.create(
|
72 |
+
model=self.model_name,
|
73 |
+
messages=history,
|
74 |
+
max_tokens=gen_conf.get("max_tokens", 1000),
|
75 |
+
temperature=gen_conf.get("temperature", 0.3),
|
76 |
+
top_p=gen_conf.get("top_p", 0.7),
|
77 |
+
stream=True
|
78 |
+
)
|
79 |
+
for resp in response:
|
80 |
+
if not resp.choices[0].delta.content: continue
|
81 |
+
delta = resp.choices[0].delta.content
|
82 |
+
ans += delta
|
83 |
+
if resp.choices[0].finish_reason == "length":
|
84 |
+
ans += "...\nFor the content length reason, it stopped, continue?" if is_english(
|
85 |
+
[ans]) else "路路路路路路\n鐢变簬闀垮害鐨勫師鍥狅紝鍥炵瓟琚埅鏂簡锛岃缁х画鍚楋紵"
|
86 |
+
tk_count = resp.usage.total_tokens
|
87 |
+
if resp.choices[0].finish_reason == "stop": tk_count = resp.usage.total_tokens
|
88 |
+
yield ans
|
89 |
+
except Exception as e:
|
90 |
+
yield ans + "\n**ERROR**: " + str(e)
|
91 |
+
|
92 |
+
yield tk_count
|
93 |
+
|
94 |
def image2base64(self, image):
|
95 |
if isinstance(image, bytes):
|
96 |
return base64.b64encode(image).decode("utf-8")
|
|
|
122 |
}
|
123 |
]
|
124 |
|
125 |
+
def chat_prompt(self, text, b64):
|
126 |
+
return [
|
127 |
+
{
|
128 |
+
"type": "image_url",
|
129 |
+
"image_url": {
|
130 |
+
"url": f"data:image/jpeg;base64,{b64}",
|
131 |
+
},
|
132 |
+
},
|
133 |
+
{
|
134 |
+
"type": "text",
|
135 |
+
"text": text
|
136 |
+
},
|
137 |
+
]
|
138 |
+
|
139 |
+
|
140 |
|
141 |
class GptV4(Base):
|
142 |
def __init__(self, key, model_name="gpt-4-vision-preview", lang="Chinese", base_url="https://api.openai.com/v1"):
|
|
|
209 |
}
|
210 |
]
|
211 |
|
212 |
+
def chat_prompt(self, text, b64):
|
213 |
+
return [
|
214 |
+
{"image": f"{b64}"},
|
215 |
+
{"text": text},
|
216 |
+
]
|
217 |
+
|
218 |
def describe(self, image, max_tokens=300):
|
219 |
from http import HTTPStatus
|
220 |
from dashscope import MultiModalConversation
|
|
|
224 |
return response.output.choices[0]['message']['content'][0]["text"], response.usage.output_tokens
|
225 |
return response.message, 0
|
226 |
|
227 |
+
def chat(self, system, history, gen_conf, image=""):
|
228 |
+
from http import HTTPStatus
|
229 |
+
from dashscope import MultiModalConversation
|
230 |
+
if system:
|
231 |
+
history[-1]["content"] = system + history[-1]["content"] + "user query: " + history[-1]["content"]
|
232 |
+
|
233 |
+
for his in history:
|
234 |
+
if his["role"] == "user":
|
235 |
+
his["content"] = self.chat_prompt(his["content"], image)
|
236 |
+
response = MultiModalConversation.call(model=self.model_name, messages=history,
|
237 |
+
max_tokens=gen_conf.get("max_tokens", 1000),
|
238 |
+
temperature=gen_conf.get("temperature", 0.3),
|
239 |
+
top_p=gen_conf.get("top_p", 0.7))
|
240 |
+
|
241 |
+
ans = ""
|
242 |
+
tk_count = 0
|
243 |
+
if response.status_code == HTTPStatus.OK:
|
244 |
+
ans += response.output.choices[0]['message']['content']
|
245 |
+
tk_count += response.usage.total_tokens
|
246 |
+
if response.output.choices[0].get("finish_reason", "") == "length":
|
247 |
+
ans += "...\nFor the content length reason, it stopped, continue?" if is_english(
|
248 |
+
[ans]) else "路路路路路路\n鐢变簬闀垮害鐨勫師鍥狅紝鍥炵瓟琚埅鏂簡锛岃缁х画鍚楋紵"
|
249 |
+
return ans, tk_count
|
250 |
+
|
251 |
+
return "**ERROR**: " + response.message, tk_count
|
252 |
+
|
253 |
+
def chat_streamly(self, system, history, gen_conf, image=""):
|
254 |
+
from http import HTTPStatus
|
255 |
+
from dashscope import MultiModalConversation
|
256 |
+
if system:
|
257 |
+
history[-1]["content"] = system + history[-1]["content"] + "user query: " + history[-1]["content"]
|
258 |
+
|
259 |
+
for his in history:
|
260 |
+
if his["role"] == "user":
|
261 |
+
his["content"] = self.chat_prompt(his["content"], image)
|
262 |
+
|
263 |
+
ans = ""
|
264 |
+
tk_count = 0
|
265 |
+
try:
|
266 |
+
response = MultiModalConversation.call(model=self.model_name, messages=history,
|
267 |
+
max_tokens=gen_conf.get("max_tokens", 1000),
|
268 |
+
temperature=gen_conf.get("temperature", 0.3),
|
269 |
+
top_p=gen_conf.get("top_p", 0.7),
|
270 |
+
stream=True)
|
271 |
+
for resp in response:
|
272 |
+
if resp.status_code == HTTPStatus.OK:
|
273 |
+
ans = resp.output.choices[0]['message']['content']
|
274 |
+
tk_count = resp.usage.total_tokens
|
275 |
+
if resp.output.choices[0].get("finish_reason", "") == "length":
|
276 |
+
ans += "...\nFor the content length reason, it stopped, continue?" if is_english(
|
277 |
+
[ans]) else "路路路路路路\n鐢变簬闀垮害鐨勫師鍥狅紝鍥炵瓟琚埅鏂簡锛岃缁х画鍚楋紵"
|
278 |
+
yield ans
|
279 |
+
else:
|
280 |
+
yield ans + "\n**ERROR**: " + resp.message if str(resp.message).find(
|
281 |
+
"Access") < 0 else "Out of credit. Please set the API key in **settings > Model providers.**"
|
282 |
+
except Exception as e:
|
283 |
+
yield ans + "\n**ERROR**: " + str(e)
|
284 |
+
|
285 |
+
yield tk_count
|
286 |
+
|
287 |
|
288 |
class Zhipu4V(Base):
|
289 |
def __init__(self, key, model_name="glm-4v", lang="Chinese", **kwargs):
|
|
|
301 |
)
|
302 |
return res.choices[0].message.content.strip(), res.usage.total_tokens
|
303 |
|
304 |
+
def chat(self, system, history, gen_conf, image=""):
|
305 |
+
if system:
|
306 |
+
history[-1]["content"] = system + history[-1]["content"] + "user query: " + history[-1]["content"]
|
307 |
+
try:
|
308 |
+
for his in history:
|
309 |
+
if his["role"] == "user":
|
310 |
+
his["content"] = self.chat_prompt(his["content"], image)
|
311 |
+
|
312 |
+
response = self.client.chat.completions.create(
|
313 |
+
model=self.model_name,
|
314 |
+
messages=history,
|
315 |
+
max_tokens=gen_conf.get("max_tokens", 1000),
|
316 |
+
temperature=gen_conf.get("temperature", 0.3),
|
317 |
+
top_p=gen_conf.get("top_p", 0.7)
|
318 |
+
)
|
319 |
+
return response.choices[0].message.content.strip(), response.usage.total_tokens
|
320 |
+
except Exception as e:
|
321 |
+
return "**ERROR**: " + str(e), 0
|
322 |
+
|
323 |
+
def chat_streamly(self, system, history, gen_conf, image=""):
|
324 |
+
if system:
|
325 |
+
history[-1]["content"] = system + history[-1]["content"] + "user query: " + history[-1]["content"]
|
326 |
+
|
327 |
+
ans = ""
|
328 |
+
tk_count = 0
|
329 |
+
try:
|
330 |
+
for his in history:
|
331 |
+
if his["role"] == "user":
|
332 |
+
his["content"] = self.chat_prompt(his["content"], image)
|
333 |
+
|
334 |
+
response = self.client.chat.completions.create(
|
335 |
+
model=self.model_name,
|
336 |
+
messages=history,
|
337 |
+
max_tokens=gen_conf.get("max_tokens", 1000),
|
338 |
+
temperature=gen_conf.get("temperature", 0.3),
|
339 |
+
top_p=gen_conf.get("top_p", 0.7),
|
340 |
+
stream=True
|
341 |
+
)
|
342 |
+
for resp in response:
|
343 |
+
if not resp.choices[0].delta.content: continue
|
344 |
+
delta = resp.choices[0].delta.content
|
345 |
+
ans += delta
|
346 |
+
if resp.choices[0].finish_reason == "length":
|
347 |
+
ans += "...\nFor the content length reason, it stopped, continue?" if is_english(
|
348 |
+
[ans]) else "路路路路路路\n鐢变簬闀垮害鐨勫師鍥狅紝鍥炵瓟琚埅鏂簡锛岃缁х画鍚楋紵"
|
349 |
+
tk_count = resp.usage.total_tokens
|
350 |
+
if resp.choices[0].finish_reason == "stop": tk_count = resp.usage.total_tokens
|
351 |
+
yield ans
|
352 |
+
except Exception as e:
|
353 |
+
yield ans + "\n**ERROR**: " + str(e)
|
354 |
+
|
355 |
+
yield tk_count
|
356 |
+
|
357 |
|
358 |
class OllamaCV(Base):
|
359 |
def __init__(self, key, model_name, lang="Chinese", **kwargs):
|
|
|
376 |
except Exception as e:
|
377 |
return "**ERROR**: " + str(e), 0
|
378 |
|
379 |
+
def chat(self, system, history, gen_conf, image=""):
|
380 |
+
if system:
|
381 |
+
history[-1]["content"] = system + history[-1]["content"] + "user query: " + history[-1]["content"]
|
382 |
+
|
383 |
+
try:
|
384 |
+
for his in history:
|
385 |
+
if his["role"] == "user":
|
386 |
+
his["images"] = [image]
|
387 |
+
options = {}
|
388 |
+
if "temperature" in gen_conf: options["temperature"] = gen_conf["temperature"]
|
389 |
+
if "max_tokens" in gen_conf: options["num_predict"] = gen_conf["max_tokens"]
|
390 |
+
if "top_p" in gen_conf: options["top_k"] = gen_conf["top_p"]
|
391 |
+
if "presence_penalty" in gen_conf: options["presence_penalty"] = gen_conf["presence_penalty"]
|
392 |
+
if "frequency_penalty" in gen_conf: options["frequency_penalty"] = gen_conf["frequency_penalty"]
|
393 |
+
response = self.client.chat(
|
394 |
+
model=self.model_name,
|
395 |
+
messages=history,
|
396 |
+
options=options,
|
397 |
+
keep_alive=-1
|
398 |
+
)
|
399 |
+
|
400 |
+
ans = response["message"]["content"].strip()
|
401 |
+
return ans, response["eval_count"] + response.get("prompt_eval_count", 0)
|
402 |
+
except Exception as e:
|
403 |
+
return "**ERROR**: " + str(e), 0
|
404 |
+
|
405 |
+
def chat_streamly(self, system, history, gen_conf, image=""):
|
406 |
+
if system:
|
407 |
+
history[-1]["content"] = system + history[-1]["content"] + "user query: " + history[-1]["content"]
|
408 |
+
|
409 |
+
for his in history:
|
410 |
+
if his["role"] == "user":
|
411 |
+
his["images"] = [image]
|
412 |
+
options = {}
|
413 |
+
if "temperature" in gen_conf: options["temperature"] = gen_conf["temperature"]
|
414 |
+
if "max_tokens" in gen_conf: options["num_predict"] = gen_conf["max_tokens"]
|
415 |
+
if "top_p" in gen_conf: options["top_k"] = gen_conf["top_p"]
|
416 |
+
if "presence_penalty" in gen_conf: options["presence_penalty"] = gen_conf["presence_penalty"]
|
417 |
+
if "frequency_penalty" in gen_conf: options["frequency_penalty"] = gen_conf["frequency_penalty"]
|
418 |
+
ans = ""
|
419 |
+
try:
|
420 |
+
response = self.client.chat(
|
421 |
+
model=self.model_name,
|
422 |
+
messages=history,
|
423 |
+
stream=True,
|
424 |
+
options=options,
|
425 |
+
keep_alive=-1
|
426 |
+
)
|
427 |
+
for resp in response:
|
428 |
+
if resp["done"]:
|
429 |
+
yield resp.get("prompt_eval_count", 0) + resp.get("eval_count", 0)
|
430 |
+
ans += resp["message"]["content"]
|
431 |
+
yield ans
|
432 |
+
except Exception as e:
|
433 |
+
yield ans + "\n**ERROR**: " + str(e)
|
434 |
+
yield 0
|
435 |
+
|
436 |
|
437 |
class LocalAICV(Base):
|
438 |
def __init__(self, key, model_name, base_url, lang="Chinese"):
|
|
|
481 |
|
482 |
class GeminiCV(Base):
|
483 |
def __init__(self, key, model_name="gemini-1.0-pro-vision-latest", lang="Chinese", **kwargs):
|
484 |
+
from google.generativeai import client, GenerativeModel, GenerationConfig
|
485 |
client.configure(api_key=key)
|
486 |
_client = client.get_default_generative_client()
|
487 |
self.model_name = model_name
|
|
|
503 |
)
|
504 |
return res.text,res.usage_metadata.total_token_count
|
505 |
|
506 |
+
def chat(self, system, history, gen_conf, image=""):
|
507 |
+
if system:
|
508 |
+
history[-1]["content"] = system + history[-1]["content"] + "user query: " + history[-1]["content"]
|
509 |
+
try:
|
510 |
+
for his in history:
|
511 |
+
if his["role"] == "assistant":
|
512 |
+
his["role"] = "model"
|
513 |
+
his["parts"] = [his["content"]]
|
514 |
+
his.pop("content")
|
515 |
+
if his["role"] == "user":
|
516 |
+
his["parts"] = [his["content"]]
|
517 |
+
his.pop("content")
|
518 |
+
history[-1]["parts"].append(f"data:image/jpeg;base64," + image)
|
519 |
+
|
520 |
+
response = self.model.generate_content(history, generation_config=GenerationConfig(
|
521 |
+
max_output_tokens=gen_conf.get("max_tokens", 1000), temperature=gen_conf.get("temperature", 0.3),
|
522 |
+
top_p=gen_conf.get("top_p", 0.7)))
|
523 |
+
|
524 |
+
ans = response.text
|
525 |
+
return ans, response.usage_metadata.total_token_count
|
526 |
+
except Exception as e:
|
527 |
+
return "**ERROR**: " + str(e), 0
|
528 |
+
|
529 |
+
def chat_streamly(self, system, history, gen_conf, image=""):
|
530 |
+
if system:
|
531 |
+
history[-1]["content"] = system + history[-1]["content"] + "user query: " + history[-1]["content"]
|
532 |
+
|
533 |
+
ans = ""
|
534 |
+
tk_count = 0
|
535 |
+
try:
|
536 |
+
for his in history:
|
537 |
+
if his["role"] == "assistant":
|
538 |
+
his["role"] = "model"
|
539 |
+
his["parts"] = [his["content"]]
|
540 |
+
his.pop("content")
|
541 |
+
if his["role"] == "user":
|
542 |
+
his["parts"] = [his["content"]]
|
543 |
+
his.pop("content")
|
544 |
+
history[-1]["parts"].append(f"data:image/jpeg;base64," + image)
|
545 |
+
|
546 |
+
response = self.model.generate_content(history, generation_config=GenerationConfig(
|
547 |
+
max_output_tokens=gen_conf.get("max_tokens", 1000), temperature=gen_conf.get("temperature", 0.3),
|
548 |
+
top_p=gen_conf.get("top_p", 0.7)), stream=True)
|
549 |
+
|
550 |
+
for resp in response:
|
551 |
+
if not resp.text: continue
|
552 |
+
ans += resp.text
|
553 |
+
yield ans
|
554 |
+
except Exception as e:
|
555 |
+
yield ans + "\n**ERROR**: " + str(e)
|
556 |
+
|
557 |
+
yield response._chunks[-1].usage_metadata.total_token_count
|
558 |
+
|
559 |
|
560 |
class OpenRouterCV(Base):
|
561 |
def __init__(
|
web/src/components/llm-setting-items/index.tsx
CHANGED
@@ -46,7 +46,7 @@ const LlmSettingItems = ({ prefix, formItemLayout = {} }: IProps) => {
|
|
46 |
{...formItemLayout}
|
47 |
rules={[{ required: true, message: t('modelMessage') }]}
|
48 |
>
|
49 |
-
<Select options={modelOptions[LlmModelType.Chat]} showSearch
|
50 |
</Form.Item>
|
51 |
<Divider></Divider>
|
52 |
<Form.Item
|
|
|
46 |
{...formItemLayout}
|
47 |
rules={[{ required: true, message: t('modelMessage') }]}
|
48 |
>
|
49 |
+
<Select options={[...modelOptions[LlmModelType.Chat], ...modelOptions[LlmModelType.Image2text],]} showSearch/>
|
50 |
</Form.Item>
|
51 |
<Divider></Divider>
|
52 |
<Form.Item
|