KevinHuSh
commited on
Commit
·
c372afe
1
Parent(s):
6b8fc2c
change licence (#28)
Browse files* add front end code
* change licence
- rag/llm/__init__.py +1 -1
- rag/llm/chat_model.py +1 -1
- rag/llm/cv_model.py +1 -1
- rag/llm/embedding_model.py +1 -1
- rag/nlp/search.py +24 -30
- rag/nlp/term_weight.py +1 -1
- rag/settings.py +1 -1
- rag/svr/parse_user_docs.py +20 -30
- rag/utils/es_conn.py +20 -0
rag/llm/__init__.py
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
#
|
2 |
-
# Copyright 2019 The
|
3 |
#
|
4 |
# Licensed under the Apache License, Version 2.0 (the "License");
|
5 |
# you may not use this file except in compliance with the License.
|
|
|
1 |
#
|
2 |
+
# Copyright 2019 The RAG Flow Authors. All Rights Reserved.
|
3 |
#
|
4 |
# Licensed under the Apache License, Version 2.0 (the "License");
|
5 |
# you may not use this file except in compliance with the License.
|
rag/llm/chat_model.py
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
#
|
2 |
-
# Copyright 2019 The
|
3 |
#
|
4 |
# Licensed under the Apache License, Version 2.0 (the "License");
|
5 |
# you may not use this file except in compliance with the License.
|
|
|
1 |
#
|
2 |
+
# Copyright 2019 The RAG Flow Authors. All Rights Reserved.
|
3 |
#
|
4 |
# Licensed under the Apache License, Version 2.0 (the "License");
|
5 |
# you may not use this file except in compliance with the License.
|
rag/llm/cv_model.py
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
#
|
2 |
-
# Copyright 2019 The
|
3 |
#
|
4 |
# Licensed under the Apache License, Version 2.0 (the "License");
|
5 |
# you may not use this file except in compliance with the License.
|
|
|
1 |
#
|
2 |
+
# Copyright 2019 The RAG Flow Authors. All Rights Reserved.
|
3 |
#
|
4 |
# Licensed under the Apache License, Version 2.0 (the "License");
|
5 |
# you may not use this file except in compliance with the License.
|
rag/llm/embedding_model.py
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
#
|
2 |
-
# Copyright 2019 The
|
3 |
#
|
4 |
# Licensed under the Apache License, Version 2.0 (the "License");
|
5 |
# you may not use this file except in compliance with the License.
|
|
|
1 |
#
|
2 |
+
# Copyright 2019 The RAG Flow Authors. All Rights Reserved.
|
3 |
#
|
4 |
# Licensed under the Apache License, Version 2.0 (the "License");
|
5 |
# you may not use this file except in compliance with the License.
|
rag/nlp/search.py
CHANGED
@@ -1,8 +1,11 @@
|
|
1 |
# -*- coding: utf-8 -*-
|
|
|
2 |
import re
|
3 |
from elasticsearch_dsl import Q, Search, A
|
4 |
from typing import List, Optional, Tuple, Dict, Union
|
5 |
from dataclasses import dataclass
|
|
|
|
|
6 |
from rag.utils import rmSpace
|
7 |
from rag.nlp import huqie, query
|
8 |
import numpy as np
|
@@ -34,30 +37,30 @@ class Dealer:
|
|
34 |
group_docs: List[List] = None
|
35 |
|
36 |
def _vector(self, txt, sim=0.8, topk=10):
|
|
|
37 |
return {
|
38 |
-
"field": "
|
39 |
"k": topk,
|
40 |
"similarity": sim,
|
41 |
"num_candidates": 1000,
|
42 |
-
"query_vector":
|
43 |
}
|
44 |
|
45 |
def search(self, req, idxnm, tks_num=3):
|
46 |
-
keywords = []
|
47 |
qst = req.get("question", "")
|
48 |
-
|
49 |
bqry, keywords = self.qryr.question(qst)
|
50 |
if req.get("kb_ids"):
|
51 |
bqry.filter.append(Q("terms", kb_id=req["kb_ids"]))
|
52 |
-
|
|
|
53 |
bqry.boost = 0.05
|
54 |
-
print(bqry)
|
55 |
|
56 |
s = Search()
|
57 |
pg = int(req.get("page", 1)) - 1
|
58 |
ps = int(req.get("size", 1000))
|
59 |
-
src = req.get("
|
60 |
-
"image_id", "doc_id", "
|
|
|
61 |
|
62 |
s = s.query(bqry)[pg * ps:(pg + 1) * ps]
|
63 |
s = s.highlight("content_ltks")
|
@@ -66,22 +69,24 @@ class Dealer:
|
|
66 |
s = s.sort(
|
67 |
{"create_time": {"order": "desc", "unmapped_type": "date"}})
|
68 |
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
|
|
76 |
s = s.to_dict()
|
77 |
q_vec = []
|
78 |
if req.get("vector"):
|
79 |
s["knn"] = self._vector(qst, req.get("similarity", 0.4), ps)
|
80 |
s["knn"]["filter"] = bqry.to_dict()
|
81 |
-
del s["highlight"]
|
82 |
q_vec = s["knn"]["query_vector"]
|
|
|
83 |
res = self.es.search(s, idxnm=idxnm, timeout="600s", src=src)
|
84 |
-
|
85 |
if self.es.getTotal(res) == 0 and "knn" in s:
|
86 |
bqry, _ = self.qryr.question(qst, min_match="10%")
|
87 |
if req.get("kb_ids"):
|
@@ -109,8 +114,7 @@ class Dealer:
|
|
109 |
query_vector=q_vec,
|
110 |
aggregation=aggs,
|
111 |
highlight=self.getHighlight(res),
|
112 |
-
field=self.getFields(res,
|
113 |
-
"kb_id", "image_id", "doc_id", "q_vec"]),
|
114 |
keywords=list(kwds)
|
115 |
)
|
116 |
|
@@ -237,14 +241,4 @@ class Dealer:
|
|
237 |
return sim
|
238 |
|
239 |
|
240 |
-
|
241 |
-
from util import es_conn
|
242 |
-
SE = Dealer(es_conn.HuEs("infiniflow"))
|
243 |
-
qs = [
|
244 |
-
"胡凯",
|
245 |
-
""
|
246 |
-
]
|
247 |
-
for q in qs:
|
248 |
-
print(">>>>>>>>>>>>>>>>>>>>", q)
|
249 |
-
print(SE.search(
|
250 |
-
{"question": q, "kb_ids": "64f072a75f3b97c865718c4a"}, "infiniflow_*"))
|
|
|
1 |
# -*- coding: utf-8 -*-
|
2 |
+
import json
|
3 |
import re
|
4 |
from elasticsearch_dsl import Q, Search, A
|
5 |
from typing import List, Optional, Tuple, Dict, Union
|
6 |
from dataclasses import dataclass
|
7 |
+
|
8 |
+
from rag.settings import es_logger
|
9 |
from rag.utils import rmSpace
|
10 |
from rag.nlp import huqie, query
|
11 |
import numpy as np
|
|
|
37 |
group_docs: List[List] = None
|
38 |
|
39 |
def _vector(self, txt, sim=0.8, topk=10):
|
40 |
+
qv = self.emb_mdl.encode_queries(txt)
|
41 |
return {
|
42 |
+
"field": "q_%d_vec"%len(qv),
|
43 |
"k": topk,
|
44 |
"similarity": sim,
|
45 |
"num_candidates": 1000,
|
46 |
+
"query_vector": qv
|
47 |
}
|
48 |
|
49 |
def search(self, req, idxnm, tks_num=3):
|
|
|
50 |
qst = req.get("question", "")
|
|
|
51 |
bqry, keywords = self.qryr.question(qst)
|
52 |
if req.get("kb_ids"):
|
53 |
bqry.filter.append(Q("terms", kb_id=req["kb_ids"]))
|
54 |
+
if req.get("doc_ids"):
|
55 |
+
bqry.filter.append(Q("terms", doc_id=req["doc_ids"]))
|
56 |
bqry.boost = 0.05
|
|
|
57 |
|
58 |
s = Search()
|
59 |
pg = int(req.get("page", 1)) - 1
|
60 |
ps = int(req.get("size", 1000))
|
61 |
+
src = req.get("fields", ["docnm_kwd", "content_ltks", "kb_id","img_id",
|
62 |
+
"image_id", "doc_id", "q_512_vec", "q_768_vec",
|
63 |
+
"q_1024_vec", "q_1536_vec"])
|
64 |
|
65 |
s = s.query(bqry)[pg * ps:(pg + 1) * ps]
|
66 |
s = s.highlight("content_ltks")
|
|
|
69 |
s = s.sort(
|
70 |
{"create_time": {"order": "desc", "unmapped_type": "date"}})
|
71 |
|
72 |
+
if qst:
|
73 |
+
s = s.highlight_options(
|
74 |
+
fragment_size=120,
|
75 |
+
number_of_fragments=5,
|
76 |
+
boundary_scanner_locale="zh-CN",
|
77 |
+
boundary_scanner="SENTENCE",
|
78 |
+
boundary_chars=",./;:\\!(),。?:!……()——、"
|
79 |
+
)
|
80 |
s = s.to_dict()
|
81 |
q_vec = []
|
82 |
if req.get("vector"):
|
83 |
s["knn"] = self._vector(qst, req.get("similarity", 0.4), ps)
|
84 |
s["knn"]["filter"] = bqry.to_dict()
|
85 |
+
if "highlight" in s: del s["highlight"]
|
86 |
q_vec = s["knn"]["query_vector"]
|
87 |
+
es_logger.info("【Q】: {}".format(json.dumps(s)))
|
88 |
res = self.es.search(s, idxnm=idxnm, timeout="600s", src=src)
|
89 |
+
es_logger.info("TOTAL: {}".format(self.es.getTotal(res)))
|
90 |
if self.es.getTotal(res) == 0 and "knn" in s:
|
91 |
bqry, _ = self.qryr.question(qst, min_match="10%")
|
92 |
if req.get("kb_ids"):
|
|
|
114 |
query_vector=q_vec,
|
115 |
aggregation=aggs,
|
116 |
highlight=self.getHighlight(res),
|
117 |
+
field=self.getFields(res, src),
|
|
|
118 |
keywords=list(kwds)
|
119 |
)
|
120 |
|
|
|
241 |
return sim
|
242 |
|
243 |
|
244 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
rag/nlp/term_weight.py
CHANGED
@@ -62,7 +62,7 @@ class Dealer:
|
|
62 |
return set(res.keys())
|
63 |
return res
|
64 |
|
65 |
-
fnm = os.path.join(get_project_base_directory(), "res")
|
66 |
self.ne, self.df = {}, {}
|
67 |
try:
|
68 |
self.ne = json.load(open(os.path.join(fnm, "ner.json"), "r"))
|
|
|
62 |
return set(res.keys())
|
63 |
return res
|
64 |
|
65 |
+
fnm = os.path.join(get_project_base_directory(), "rag/res")
|
66 |
self.ne, self.df = {}, {}
|
67 |
try:
|
68 |
self.ne = json.load(open(os.path.join(fnm, "ner.json"), "r"))
|
rag/settings.py
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
#
|
2 |
-
# Copyright 2019 The
|
3 |
#
|
4 |
# Licensed under the Apache License, Version 2.0 (the "License");
|
5 |
# you may not use this file except in compliance with the License.
|
|
|
1 |
#
|
2 |
+
# Copyright 2019 The RAG Flow Authors. All Rights Reserved.
|
3 |
#
|
4 |
# Licensed under the Apache License, Version 2.0 (the "License");
|
5 |
# you may not use this file except in compliance with the License.
|
rag/svr/parse_user_docs.py
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
#
|
2 |
-
# Copyright 2019 The
|
3 |
#
|
4 |
# Licensed under the Apache License, Version 2.0 (the "License");
|
5 |
# you may not use this file except in compliance with the License.
|
@@ -13,6 +13,7 @@
|
|
13 |
# See the License for the specific language governing permissions and
|
14 |
# limitations under the License.
|
15 |
#
|
|
|
16 |
import json
|
17 |
import logging
|
18 |
import os
|
@@ -108,17 +109,17 @@ def build(row, cvmdl):
|
|
108 |
(int(DOC_MAXIMUM_SIZE / 1024 / 1024)))
|
109 |
return []
|
110 |
|
111 |
-
res = ELASTICSEARCH.search(Q("term", doc_id=row["id"]))
|
112 |
-
if ELASTICSEARCH.getTotal(res) > 0:
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
|
123 |
random.seed(time.time())
|
124 |
set_progress(row["id"], random.randint(0, 20) /
|
@@ -155,8 +156,7 @@ def build(row, cvmdl):
|
|
155 |
"doc_id": row["id"],
|
156 |
"kb_id": [str(row["kb_id"])],
|
157 |
"docnm_kwd": os.path.split(row["location"])[-1],
|
158 |
-
"title_tks": huqie.qie(row["name"])
|
159 |
-
"updated_at": str(row["update_time"]).replace("T", " ")[:19]
|
160 |
}
|
161 |
doc["title_sm_tks"] = huqie.qieqie(doc["title_tks"])
|
162 |
output_buffer = BytesIO()
|
@@ -179,6 +179,7 @@ def build(row, cvmdl):
|
|
179 |
|
180 |
MINIO.put(row["kb_id"], d["_id"], output_buffer.getvalue())
|
181 |
d["img_id"] = "{}-{}".format(row["kb_id"], d["_id"])
|
|
|
182 |
docs.append(d)
|
183 |
|
184 |
for arr, img in obj.table_chunks:
|
@@ -193,6 +194,7 @@ def build(row, cvmdl):
|
|
193 |
img.save(output_buffer, format='JPEG')
|
194 |
MINIO.put(row["kb_id"], d["_id"], output_buffer.getvalue())
|
195 |
d["img_id"] = "{}-{}".format(row["kb_id"], d["_id"])
|
|
|
196 |
docs.append(d)
|
197 |
set_progress(row["id"], random.randint(60, 70) /
|
198 |
100., "Continue embedding the content.")
|
@@ -218,23 +220,11 @@ def embedding(docs, mdl):
|
|
218 |
vects = 0.1 * tts + 0.9 * cnts
|
219 |
assert len(vects) == len(docs)
|
220 |
for i, d in enumerate(docs):
|
221 |
-
|
|
|
222 |
return tk_count
|
223 |
|
224 |
|
225 |
-
def model_instance(tenant_id, llm_type):
|
226 |
-
model_config = TenantLLMService.get_api_key(tenant_id, model_type=LLMType.EMBEDDING)
|
227 |
-
if not model_config:
|
228 |
-
model_config = {"llm_factory": "local", "api_key": "", "llm_name": ""}
|
229 |
-
else: model_config = model_config[0].to_dict()
|
230 |
-
if llm_type == LLMType.EMBEDDING:
|
231 |
-
if model_config["llm_factory"] not in EmbeddingModel: return
|
232 |
-
return EmbeddingModel[model_config["llm_factory"]](model_config["api_key"], model_config["llm_name"])
|
233 |
-
if llm_type == LLMType.IMAGE2TEXT:
|
234 |
-
if model_config["llm_factory"] not in CvModel: return
|
235 |
-
return CvModel[model_config.llm_factory](model_config["api_key"], model_config["llm_name"])
|
236 |
-
|
237 |
-
|
238 |
def main(comm, mod):
|
239 |
global model
|
240 |
from rag.llm import HuEmbedding
|
@@ -247,12 +237,12 @@ def main(comm, mod):
|
|
247 |
|
248 |
tmf = open(tm_fnm, "a+")
|
249 |
for _, r in rows.iterrows():
|
250 |
-
embd_mdl = model_instance(r["tenant_id"], LLMType.EMBEDDING)
|
251 |
if not embd_mdl:
|
252 |
set_progress(r["id"], -1, "Can't find embedding model!")
|
253 |
cron_logger.error("Tenant({}) can't find embedding model!".format(r["tenant_id"]))
|
254 |
continue
|
255 |
-
cv_mdl = model_instance(r["tenant_id"], LLMType.IMAGE2TEXT)
|
256 |
st_tm = timer()
|
257 |
cks = build(r, cv_mdl)
|
258 |
if not cks:
|
|
|
1 |
#
|
2 |
+
# Copyright 2019 The RAG Flow Authors. All Rights Reserved.
|
3 |
#
|
4 |
# Licensed under the Apache License, Version 2.0 (the "License");
|
5 |
# you may not use this file except in compliance with the License.
|
|
|
13 |
# See the License for the specific language governing permissions and
|
14 |
# limitations under the License.
|
15 |
#
|
16 |
+
import datetime
|
17 |
import json
|
18 |
import logging
|
19 |
import os
|
|
|
109 |
(int(DOC_MAXIMUM_SIZE / 1024 / 1024)))
|
110 |
return []
|
111 |
|
112 |
+
# res = ELASTICSEARCH.search(Q("term", doc_id=row["id"]))
|
113 |
+
# if ELASTICSEARCH.getTotal(res) > 0:
|
114 |
+
# ELASTICSEARCH.updateScriptByQuery(Q("term", doc_id=row["id"]),
|
115 |
+
# scripts="""
|
116 |
+
# if(!ctx._source.kb_id.contains('%s'))
|
117 |
+
# ctx._source.kb_id.add('%s');
|
118 |
+
# """ % (str(row["kb_id"]), str(row["kb_id"])),
|
119 |
+
# idxnm=search.index_name(row["tenant_id"])
|
120 |
+
# )
|
121 |
+
# set_progress(row["id"], 1, "Done")
|
122 |
+
# return []
|
123 |
|
124 |
random.seed(time.time())
|
125 |
set_progress(row["id"], random.randint(0, 20) /
|
|
|
156 |
"doc_id": row["id"],
|
157 |
"kb_id": [str(row["kb_id"])],
|
158 |
"docnm_kwd": os.path.split(row["location"])[-1],
|
159 |
+
"title_tks": huqie.qie(row["name"])
|
|
|
160 |
}
|
161 |
doc["title_sm_tks"] = huqie.qieqie(doc["title_tks"])
|
162 |
output_buffer = BytesIO()
|
|
|
179 |
|
180 |
MINIO.put(row["kb_id"], d["_id"], output_buffer.getvalue())
|
181 |
d["img_id"] = "{}-{}".format(row["kb_id"], d["_id"])
|
182 |
+
d["create_time"] = str(datetime.datetime.now()).replace("T", " ")[:19]
|
183 |
docs.append(d)
|
184 |
|
185 |
for arr, img in obj.table_chunks:
|
|
|
194 |
img.save(output_buffer, format='JPEG')
|
195 |
MINIO.put(row["kb_id"], d["_id"], output_buffer.getvalue())
|
196 |
d["img_id"] = "{}-{}".format(row["kb_id"], d["_id"])
|
197 |
+
d["create_time"] = str(datetime.datetime.now()).replace("T", " ")[:19]
|
198 |
docs.append(d)
|
199 |
set_progress(row["id"], random.randint(60, 70) /
|
200 |
100., "Continue embedding the content.")
|
|
|
220 |
vects = 0.1 * tts + 0.9 * cnts
|
221 |
assert len(vects) == len(docs)
|
222 |
for i, d in enumerate(docs):
|
223 |
+
v = vects[i].tolist()
|
224 |
+
d["q_%d_vec"%len(v)] = v
|
225 |
return tk_count
|
226 |
|
227 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
228 |
def main(comm, mod):
|
229 |
global model
|
230 |
from rag.llm import HuEmbedding
|
|
|
237 |
|
238 |
tmf = open(tm_fnm, "a+")
|
239 |
for _, r in rows.iterrows():
|
240 |
+
embd_mdl = TenantLLMService.model_instance(r["tenant_id"], LLMType.EMBEDDING)
|
241 |
if not embd_mdl:
|
242 |
set_progress(r["id"], -1, "Can't find embedding model!")
|
243 |
cron_logger.error("Tenant({}) can't find embedding model!".format(r["tenant_id"]))
|
244 |
continue
|
245 |
+
cv_mdl = TenantLLMService.model_instance(r["tenant_id"], LLMType.IMAGE2TEXT)
|
246 |
st_tm = timer()
|
247 |
cks = build(r, cv_mdl)
|
248 |
if not cks:
|
rag/utils/es_conn.py
CHANGED
@@ -241,6 +241,26 @@ class HuEs:
|
|
241 |
es_logger.error("ES search timeout for 3 times!")
|
242 |
raise Exception("ES search timeout.")
|
243 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
244 |
def updateByQuery(self, q, d):
|
245 |
ubq = UpdateByQuery(index=self.idxnm).using(self.es).query(q)
|
246 |
scripts = ""
|
|
|
241 |
es_logger.error("ES search timeout for 3 times!")
|
242 |
raise Exception("ES search timeout.")
|
243 |
|
244 |
+
def get(self, doc_id, idxnm=None):
|
245 |
+
for i in range(3):
|
246 |
+
try:
|
247 |
+
res = self.es.get(index=(self.idxnm if not idxnm else idxnm),
|
248 |
+
id=doc_id)
|
249 |
+
if str(res.get("timed_out", "")).lower() == "true":
|
250 |
+
raise Exception("Es Timeout.")
|
251 |
+
return res
|
252 |
+
except Exception as e:
|
253 |
+
es_logger.error(
|
254 |
+
"ES get exception: " +
|
255 |
+
str(e) +
|
256 |
+
"【Q】:" +
|
257 |
+
doc_id)
|
258 |
+
if str(e).find("Timeout") > 0:
|
259 |
+
continue
|
260 |
+
raise e
|
261 |
+
es_logger.error("ES search timeout for 3 times!")
|
262 |
+
raise Exception("ES search timeout.")
|
263 |
+
|
264 |
def updateByQuery(self, q, d):
|
265 |
ubq = UpdateByQuery(index=self.idxnm).using(self.es).query(q)
|
266 |
scripts = ""
|