pengjunjie/ragflow
1
0
Fork 1
mirror of https://github.com/infiniflow/ragflow.git synced 2026-01-30 23:26:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add bce-embedding and fastembed (#383)

Browse Source 
### What problem does this PR solve?


Issue link:#326

### Type of change

- [x] New Feature (non-breaking change which adds functionality)
This commit is contained in:
KevinHuSh
2024-04-16 16:42:19 +08:00
committed by GitHub
parent a7be5d4e8b
commit 890561703b
15 changed files with 99 additions and 37 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
api/apps/chunk_app.py
View File

@ -252,7 +252,7 @@ def retrieval_test():
return get_data_error_result(retmsg="Knowledgebase not found!")
embd_mdl = TenantLLMService.model_instance(
kb.tenant_id, LLMType.EMBEDDING.value)
kb.tenant_id, LLMType.EMBEDDING.value, llm_name=kb.embd_id)
ranks = retrievaler.retrieval(question, embd_mdl, kb.tenant_id, [kb_id], page, size, similarity_threshold,
vector_similarity_weight, top, doc_ids)
for c in ranks["chunks"]:

1
api/apps/document_app.py
View File

@ -15,6 +15,7 @@
#
import base64
import os
import pathlib
import re

4
api/apps/llm_app.py
View File

@ -28,7 +28,7 @@ from rag.llm import EmbeddingModel, ChatModel
def factories():
try:
fac = LLMFactoriesService.get_all()
return get_json_result(data=[f.to_dict() for f in fac])
return get_json_result(data=[f.to_dict() for f in fac if f.name not in ["QAnything", "FastEmbed"]])
except Exception as e:
return server_error_response(e)
@ -174,7 +174,7 @@ def list():
llms = [m.to_dict()
for m in llms if m.status == StatusEnum.VALID.value]
for m in llms:
m["available"] = m["fid"] in facts or m["llm_name"].lower() == "flag-embedding"
m["available"] = m["fid"] in facts or m["llm_name"].lower() == "flag-embedding" or m["fid"] in ["QAnything","FastEmbed"]
llm_set = set([m["llm_name"] for m in llms])
for o in objs:

30
api/db/init_data.py
View File

@ -18,7 +18,7 @@ import time
import uuid
from api.db import LLMType, UserTenantRole
from api.db.db_models import init_database_tables as init_web_db, LLMFactories, LLM
from api.db.db_models import init_database_tables as init_web_db, LLMFactories, LLM, TenantLLM
from api.db.services import UserService
from api.db.services.llm_service import LLMFactoriesService, LLMService, TenantLLMService, LLMBundle
from api.db.services.user_service import TenantService, UserTenantService
@ -114,12 +114,16 @@ factory_infos = [{
"logo": "",
"tags": "TEXT EMBEDDING",
"status": "1",
},
{
}, {
"name": "Xinference",
"logo": "",
"tags": "LLM,TEXT EMBEDDING,SPEECH2TEXT,MODERATION",
"status": "1",
},{
"name": "QAnything",
"logo": "",
"tags": "LLM,TEXT EMBEDDING,SPEECH2TEXT,MODERATION",
"status": "1",
},
# {
# "name": "文心一言",
@ -254,12 +258,6 @@ def init_llm_factory():
"tags": "LLM,CHAT,",
"max_tokens": 7900,
"model_type": LLMType.CHAT.value
}, {
"fid": factory_infos[4]["name"],
"llm_name": "flag-embedding",
"tags": "TEXT EMBEDDING,",
"max_tokens": 128 * 1000,
"model_type": LLMType.EMBEDDING.value
}, {
"fid": factory_infos[4]["name"],
"llm_name": "moonshot-v1-32k",
@ -325,6 +323,14 @@ def init_llm_factory():
"max_tokens": 2147483648,
"model_type": LLMType.EMBEDDING.value
},
# ------------------------ QAnything -----------------------
{
"fid": factory_infos[7]["name"],
"llm_name": "maidalun1020/bce-embedding-base_v1",
"tags": "TEXT EMBEDDING,",
"max_tokens": 512,
"model_type": LLMType.EMBEDDING.value
},
]
for info in factory_infos:
try:
@ -337,8 +343,10 @@ def init_llm_factory():
except Exception as e:
pass
LLMFactoriesService.filter_delete([LLMFactories.name=="Local"])
LLMService.filter_delete([LLM.fid=="Local"])
LLMFactoriesService.filter_delete([LLMFactories.name == "Local"])
LLMService.filter_delete([LLM.fid == "Local"])
LLMService.filter_delete([LLM.fid == "Moonshot", LLM.llm_name == "flag-embedding"])
TenantLLMService.filter_delete([TenantLLM.llm_factory == "Moonshot", TenantLLM.llm_name == "flag-embedding"])
"""
drop table llm;

6
api/db/services/dialog_service.py
View File

@ -80,8 +80,12 @@ def chat(dialog, messages, **kwargs):
raise LookupError("LLM(%s) not found" % dialog.llm_id)
max_tokens = 1024
else: max_tokens = llm[0].max_tokens
kbs = KnowledgebaseService.get_by_ids(dialog.kb_ids)
embd_nms = list(set([kb.embd_id for kb in kbs]))
assert len(embd_nms) == 1, "Knowledge bases use different embedding models."
questions = [m["content"] for m in messages if m["role"] == "user"]
embd_mdl = LLMBundle(dialog.tenant_id, LLMType.EMBEDDING)
embd_mdl = LLMBundle(dialog.tenant_id, LLMType.EMBEDDING, embd_nms[0])
chat_mdl = LLMBundle(dialog.tenant_id, LLMType.CHAT, dialog.llm_id)
prompt_config = dialog.prompt_config

11
api/db/services/llm_service.py
View File

@ -66,7 +66,7 @@ class TenantLLMService(CommonService):
raise LookupError("Tenant not found")
if llm_type == LLMType.EMBEDDING.value:
mdlnm = tenant.embd_id
mdlnm = tenant.embd_id if not llm_name else llm_name
elif llm_type == LLMType.SPEECH2TEXT.value:
mdlnm = tenant.asr_id
elif llm_type == LLMType.IMAGE2TEXT.value:
@ -77,9 +77,14 @@ class TenantLLMService(CommonService):
assert False, "LLM type error"
model_config = cls.get_api_key(tenant_id, mdlnm)
if model_config: model_config = model_config.to_dict()
if not model_config:
raise LookupError("Model({}) not authorized".format(mdlnm))
model_config = model_config.to_dict()
if llm_type == LLMType.EMBEDDING.value:
llm = LLMService.query(llm_name=llm_name)
if llm and llm[0].fid in ["QAnything", "FastEmbed"]:
model_config = {"llm_factory": llm[0].fid, "api_key":"", "llm_name": llm_name, "api_base": ""}
if not model_config: raise LookupError("Model({}) not authorized".format(mdlnm))
if llm_type == LLMType.EMBEDDING.value:
if model_config["llm_factory"] not in EmbeddingModel:
return

2
api/db/services/task_service.py
View File

@ -41,7 +41,7 @@ class TaskService(CommonService):
Document.size,
Knowledgebase.tenant_id,
Knowledgebase.language,
Tenant.embd_id,
Knowledgebase.embd_id,
Tenant.img2txt_id,
Tenant.asr_id,
cls.model.update_time]