Spaces:
Sleeping
Sleeping
File size: 47,308 Bytes
5ec3312 3bc1015 5ec3312 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 |
# important classes
import json
import hnswlib
import numpy as np
from openai import OpenAI
from sentence_transformers import SentenceTransformer, CrossEncoder
# It formats the output by printing it in color.
# Defined arguments
API_KEY = st.write(st.secrets["API_KEY"])# api_key
model_name_tr = "gpt-3.5-turbo" # model name
model_name_ta = "gpt-4-1106-preview"
tool_list_path = './tools.json' # list of tools path[ ]
example_path = './examples.json' # list of examples path
zero_shot = 0
no_of_examples = 2
# ---------------------------------------------------some constants ----------------------------------------------------------------------
EF = 100 # EF
K = 3 # top k number
COSINE_THRESHOLD = 0.3 # cosine threshold
biencoder = SentenceTransformer("BAAI/bge-large-en-v1.5", device="cpu")
cross_encoder = CrossEncoder(
"cross-encoder/ms-marco-MiniLM-L-12-v2", max_length=512, device="cpu"
)
client = OpenAI(api_key=API_KEY, timeout=60, max_retries=2)
class color:
PURPLE = "\033[1;35;48m"
CYAN = "\033[1;36;48m"
BOLD = "\033[1;37;48m"
BLUE = "\033[1;34;48m"
GREEN = "\033[1;32;48m"
YELLOW = "\033[1;33;48m"
RED = "\033[1;31;48m"
BLACK = "\033[1;30;48m"
UNDERLINE = "\033[4;37;48m"
END = "\033[1;37;0m"
class Tools:
'''
A class to represent tools.
...
Attributes
----------
- tools: list
a list of tools (json format as described above)
- examples: list
a list of examples (json format as described above)
Methods
-------
- check_json()
checks whether the tool added is in the defined JSON schema
- build_index()
assigns index values to the examples and their respective tool calls to create an index list.
- add_tool()
adds new tool to the index list.
- add_example()
adds the new example in the example pool and modifies the index list.
- modify_example()
modifies an existing example present in the example list.
- update_tool()
updates an exisiting tool in the tool list.
- replenish_examples()
on deletion of tool replenish the examples for the tools where the num_examples < threshold
- delete_tool()
deletes the tool from the index list and example list.
- similarity_retriever()
creates the hsnw index for the example queries
'''
def __init__(self, tools, examples):
# assuming the self.tools to be a dict of tools with tool name as the key and tool info as the value.
self.tools = tools
self.examples = examples
self.index = {}
self.build_index()
self.th = 2
self.query_embeddings = []
self.queries = []
self.search_index = None
self.similarity_retriever()
def check_json(self,tool_json):
keys = ['argument_list', 'title', 'tool_description', 'tool_name']
argument_keys = ['argument_description', 'argument_name', 'argument_type', 'example']
if type(tool_json) != type({}):
raise Exception('Given tool json is not a dictionary')
if keys != sorted(list(tool_json.keys())):
raise Exception(
"""Keys don't match
Expected Keys: {}
Given Keys: {}
""".format(keys, sorted(list(tool_json.keys())))
)
if type(tool_json['argument_list']) != type([]):
raise Exception('Given argument list is not a list')
for idx, arg in enumerate(tool_json['argument_list']):
if type(arg) != type({}):
raise Exception(f'Argument at the index: {idx} is not a dictionary')
if argument_keys != sorted(list(arg.keys())):
raise Exception(
"""Keys don't match at {}
Expected Keys: {}
Given Keys: {}
""".format(idx, argument_keys, sorted(list(arg.keys())))
)
def build_index(self):
'''
Indexes all the examples with their respective tool calls.
For example: {'<tool_name>': {'num_examples': <no. of examples containing tool_name>, 'indices': <list of example indices containing respective tool_name>}}
Parameters
----------
None
Returns
----------
None
Modifies
----------
self.index
'''
for tool in self.tools:
self.index[tool] = {'num_examples': 0, 'indices': []}
for i, example in enumerate(self.examples):
for tool in example['answer']:
self.index[tool['tool_name']]['num_examples'] += 1
self.index[tool['tool_name']]['indices'].append(i)
def add_tool(self, tool):
'''
On addition of any new tool, this function gets called to add the new tool in self.index.
Parameters
----------
- tool : json (format described above)
Returns
----------
None
Modifies
----------
self.index
'''
# self.check_json(tool)
tool_name = tool['tool_name']
if tool_name in self.tools.keys():
if tool==self.tools[tool_name]:
print(color.YELLOW+'Already Exists!'+color.END)
else:
self.update_tool(tool)
print(color.YELLOW+f"[WARNING] You tried adding a tool that already exists, so updating the tool '{tool_name}'"+color.END)
else:
self.tools[tool_name] = tool
self.index[tool_name] = {'num_examples': 0, 'indices': []}
self.add_example(tool, self.th)
print(color.GREEN+f"[SUCCESS] Added the tool '{tool_name}'"+color.END)
def add_example(self, tool, no_of_eg):
'''
On addition of new tool or replenishing examples on deletion, this function gets called to add the new example in self.examples and modifies index list accordingly.
Parameters
----------
- tool : json (format described above)
- no_of_eg : int
Returns
----------
None
Modifies
----------
self.examples
self.index
'''
tool_name = tool['tool_name']
try:
ex = self.examples[:3]
except:
ex = self.examples
message = create_prompt_for_new_example(list(self.tools.values()), tool, ex, n=no_of_eg)
res = client_conn(message, model_name_tr)
res = res.choices[0].message.content
try:
new_examples = json.loads(res)
except:
new_examples = get_parsed_json(res)
for example in new_examples:
tool_calls = example['answer']
for tool in tool_calls:
new_tool_name = tool['tool_name']
self.index[new_tool_name]['num_examples'] += 1
self.index[new_tool_name]['indices'].append(len(self.examples))
self.examples.append(example)
self.similarity_retriever()
def modify_example(self, tool):
'''
On modification of a tool, this function gets called to modify the examples where the tool is used.
Parameters
----------
None
Returns
----------
None
Modifies
----------
self.examples
'''
tool_name = tool["tool_name"]
indices = self.index[tool_name]["indices"]
relevant_examples = []
for ind in indices:
relevant_examples.append(self.examples[ind])
message = create_prompt_for_modified_example(
list(self.tools.values()), tool, relevant_examples
)
res = client_conn(message, model_name_tr)
res = res.choices[0].message.content
try:
examples = json.loads(res)
except:
examples = get_parsed_json(res)
for i in range(len(examples)):
self.examples[indices[i]] = examples[i]
print(color.GREEN + f"[SUCCESS] Modified Examples!" + color.END)
self.similarity_retriever()
def update_tool(self, tool):
'''
On modification of a tool, this function gets called update the tool.
Parameters
----------
tool : json (format described above)
Returns
-------
None
Modifies
--------
self.tools
'''
# self.check_json(tool)
tool_name = tool['tool_name']
if tool_name in self.tools.keys():
self.modify_example(tool)
self.tools[tool_name] = tool
print(color.GREEN+f"[SUCCESS] Updated the tool '{tool_name}'"+color.END)
else:
print(color.YELLOW+f"[WARNING] You tried updating a tool that does not exist, so added the tool '{tool_name}'"+color.END)
self.add_tool(tool)
def replenish_examples(self):
'''
On deletion of any new tool if number of examples for other tools goes below threshold, this function gets called to create examples.
Parameters
----------
None
Returns
-------
None
Modifies
--------
None
'''
for i,tool in enumerate(self.index):
if ((self.index[tool]['num_examples']<self.th) and (len(self.tools)>4)):
self.add_example(self.tools[tool], self.th-self.index[tool]['num_examples'])
def delete_tool(self, tool_name):
'''
On deletion of any new tool, this function gets called to delete the new tool in self.index and remove the respective examples.
Parameters
----------
- tool_name : str
Returns
-------
None
Modifies
--------
self.index
self.examples
'''
if tool_name not in self.tools.keys():
print(color.RED+"[ERROR] You tried deleting a tool that does not exist."+color.END)
else:
indices = self.index[tool_name]['indices']
del self.index[tool_name]
for idx, tool in enumerate(list(self.tools.values())):
if tool['tool_name'] == tool_name:
del self.tools[tool_name]
break
if len(indices) != 0:
for idx, index in enumerate(indices):
del self.examples[index-idx]
self.index = {}
self.build_index()
print(color.GREEN+f"[SUCCESS] Deleted the tool '{tool_name}'"+color.END)
self.replenish_examples()
self.similarity_retriever()
def similarity_retriever(self):
'''
On addition of any new example, this function gets called to create the new hnsw index for queries.
Parameters
----------
None
Returns
-------
None
Modifies
--------
self.query_embeddings
self.queries
self.search_index
'''
self.query_embeddings, self.queries = create_example_query_embeddings(self.examples)
self.search_index = create_hnsw_index(np.array(self.query_embeddings))
# ------------------------------------------important functions------------------------------------------------------------
def read_file(path):
"""
parameters
----------
- path: str
path of the file to read in json.
returns
---------
- file: object
the json file object.
"""
with open(path, "r") as f:
file = json.load(f)
return file
def create_query_embedding(query):
"""
Encodes the query to get its embedding.
parameters
---------
- query: str
returns
---------
- embedding: numpy array
embedding of the query.
"""
embedding = biencoder.encode([query], normalize_embeddings=True)[0]
return embedding
def create_example_query_embeddings(examples):
"""
creates query embeddings and saves it.
parameters
----------
- examples: list
List of examples in json format.
returns
---------
- query_embeddings: list
a list of query embeddings.
- answers: list
a list of answers corresponding to each query.
- queries: list
a list of queries.
"""
query_embeddings = []
answers = []
queries = []
for example in examples:
query = example["query"]
answer = example["answer"]
queries.append(query)
query_embedding = create_query_embedding(query)
query_embeddings.append(query_embedding)
answers.append(answer)
np.save("query_embeddings.npy", query_embeddings)
return query_embeddings, queries
def create_hnsw_index(embedding, M=16, efC=100):
"""
creates the HNSW index.
parameters
----------
- embedding: list
query embedding
- M: int
default = 16
- efc: int
default = 100
returns
----------
- index: object
"""
embeddings = embedding
num_dim = embeddings.shape[1]
ids = np.arange(embeddings.shape[0])
index = hnswlib.Index(space="ip", dim=num_dim)
index.init_index(max_elements=embeddings.shape[0], ef_construction=efC, M=M)
index.add_items(embeddings, ids)
return index
def find_nearest_neighbors(query_embedding, queries, search_index):
"""
Finds the k nearest neighbors using cosine similarity.
parameters
----------
- query_embedding: list
the query embedding whose similarity check has to be made.
- queries: list
a list of queries in the examples.
- search_index: object
returns
----------
- query_list: list
a list of similar queries.
"""
search_index.set_ef(EF)
labels, distances = search_index.knn_query(
query_embedding, k=K
) # Find the k-nearest neighbors for the query embedding
labels = [
label
for label, distance in zip(labels[0], distances[0])
if (1 - distance) >= COSINE_THRESHOLD
]
query_list = [queries[i] for i in labels]
return query_list
def rerank_queries_with_cross_encoder(query, chunks):
"""
Sorts the chunks based on their scores in descending order.
parameters
---------
- query: str
- chunks: list
the list of similar chunks
returns
---------
- sorted_chunks: list
the response after reranking.
"""
pairs = [(query, chunk) for chunk in chunks]
scores = cross_encoder.predict(pairs)
sorted_chunks = [chunk for _, chunk in sorted(zip(scores, chunks), reverse=True)]
return sorted_chunks
def rerank_queries_with_cross_encoder(query, chunks):
"""
Sorts the chunks based on their scores in descending order.
parameters
---------
- query: str
- chunks: list
the list of similar chunks
returns
---------
- sorted_chunks: list
the response after reranking.
"""
pairs = [(query, chunk) for chunk in chunks]
scores = cross_encoder.predict(pairs)
sorted_chunks = [chunk for _, chunk in sorted(zip(scores, chunks), reverse=True)]
return sorted_chunks
# -------------------------------------------fetch topk examples--------------------------------------------------------------
def get_index(queries, topk_queries):
"""
Fetches the topk_indices given topk queries.
parameters:
- queries: list
A list of all queries.
- topk_queries: list
A list of topk queries.
returns:
- index: list of int
"""
index = []
for i in topk_queries:
index.append(queries.index(i))
return index
def get_topk_examples(topk_index, examples):
"""
Fetches topk examples from the examples given indices of topk examples.
parameters
----------
- topk_index: int
list of indexes of topk tools.
- examples: list
list of all examples.
returns
----------
- res: list
list of topk examples.
"""
res = []
for index in topk_index:
res.append(examples[index])
return res
def get_topk_given_query(query, queries, search_index, examples):
'''
fetches topk examples given the queries.
parameters
----------
- query: str
user query
- queries: list
a list of all queries.
- search_index:
index object
- examples: list
a list of all examples.
returns
-----------
- topk_examples: list
a list of topk examples.
'''
query_embedding = create_query_embedding(query)
topk_queries = find_nearest_neighbors(query_embedding, queries, search_index)
ranked_topk_queries = rerank_queries_with_cross_encoder(query, topk_queries)
topk_indices = get_index(queries,ranked_topk_queries)
topk_examples = get_topk_examples(topk_indices, examples)
return topk_examples
# ---------------------------------------------------prompt generation functions-------------------------------------------------
def client_conn(message, model_name):
'''
Generates an API call instant.
parameters
----------
- message: list
A list of conversation between ChatGPT and user.
returns
----------
- completion: object
An API instant.
'''
# please don't change the timeout as example generation is time consuming
client = OpenAI(api_key=API_KEY, timeout=120, max_retries=2)
completion = client.chat.completions.create(
model=model_name,
messages=message,
temperature=0.2
)
return completion
def create_prompt_for_query(user_query, examples, tools):
message = [
{
"role": "system",
"content": "The following is a friendly conversation between a human and an AI. The AI is professional and parses user input to several tasks. If the AI does not know the answer to a question, it truthfully says it does not know. The AI will be provided with a set of tools their descriptions and the argument in them. Here is the list of tools:"+ json.dumps(tools) + " \n Provide the answer in the exact format as given in the following examples. \nExamples"
}
]
for example in examples:
query = example['query']
answer = example['answer']
user_prompt = "Query: "+ str(query)
assistant_prompt = str(answer)
message.append(
{
"role" : "user",
"content": user_prompt
})
message.append(
{
"role" : "assistant",
"content": assistant_prompt
})
message.append(
{
"role":"user",
"content":"Use the above tools to learn how to use the tool on any query. Analyse how to parse the query and extract the correct information and place in the argument name and value. Use all the required tools and arguments in correct order of its calling based on the query and your learning from all the examples. Do not assume any value, you can take the value from query or the previous called tool as shown in the examples. Also focus on the allowed values argument present in tool definition."
})
message.append({
"role" : "user",
"content" : "Now its your task to respond to the user queries in the same format as that in the above examples which is json."
})
message.append({
"role" : "user",
"content" : "Query: "+ user_query
})
message.append(
{
"role":"system",
"content": "Generate the answer in a json format only. Enclose the strings in double quotes"
})
return message
def create_prompt_zero_shot(user_query, tools):
'''
Creates the prompt given user query, single examples and multi examples.
parameters
----------
- user_query: str
the user query.
- single_example: list
the list of single tool examples.
- multi_example: list
the list of multi tiik examples.
returns
----------
- message: list
the list of user prompts.
'''
message = [
{
"role": "system",
"content": "You are a intelligent AI agent specialized in giving the tool responses given a dictionary of tools. Here is the dictionary of tools: "+ json.dumps(tools)
}
]
message.append({
"role" : "user",
"content" : '''
Now its your task to respond to the user queries in the format given below
FORMAT:[{"tool_name": "...", "arguments": [{"argument_name": "...", "argument_value": ... (depending on the argument_type)}, ...]}, ...]
To reference the value of the ith tool in the chain, use $$PREV[i] as argument value. i = 0, 1, .. j-1; j = current tool’s index in the array If the query could not be answered with the given set of tools, output an empty list instead.
Output in the JSON format
'''
})
message.append({
"role" : "user",
"content" : "Query: "+ user_query
})
return message
def create_prompt_for_modified_example(old_tools, modified_tool, relevant_examples):
message = [
{
"role":"system",
"content":"You are an intelligent AI Agent specialized in modifying the old data and generating the new relevant data."
}
]
message.append({
"role":"user",
"content":"Given a list of old tools : " + json.dumps(old_tools) + "Let us say that I modified the tool" + "'" + modified_tool['tool_name'] + "'" + "to be" + json.dumps(modified_tool)+"""
Now your task is to modify the following examples where this tool was used according to its new definition keeping in mind the new schema of json mentioned above.
""" + json.dumps(relevant_examples)
})
message.append({
"role":"system",
"content":"Your response should be in json format only with the strings enclosed in double quotes ready to go in json.loads"
})
return message
# ---------------------------------------------------------------------------------------------------------------------------
def create_prompt_for_new_example(old_tools, new_tool, examples, n=no_of_examples):
message = [
{
"role":"system",
"content":"You are an intelligent AI Agent specialized in generating the new relevant data."
}
]
message.append({
"role":"user",
"content":"Given a list of old tools : " + json.dumps(old_tools) + "and a new tool" + "'" + new_tool['tool_name'] + "'" + " to be " + json.dumps(new_tool)+"""
Now your task is to create """+str(n)+""" examples of the usage of the new tool along with any of the tools from the old tool list, similar to the following example:.
""" + json.dumps(examples)
})
message.append({
"role":"system",
"content":"Your response should be in json format with the strings enclosed in double quotes. Note that To reference the value of the ith tool in the chain, use $$PREV[i] as argument value. i = 0, 1, .. j-1; j = current tool’s index in the array If the query could not be answered with the given set of tools, output an empty list instead."
})
return message
# -------------------------------------------------retriever---------------------------------------------------------------------
def create_tool_str(tools):
tool_str = ''
for tool in tools:
tool_str += f"Tool: {tool['tool_name']}, Desc: {tool['tool_description']}\n"
return tool_str
def tool_retriever_prompt(tool_str,user_query):
message=[
{"role": "system", "content": "You are an intelligent assistant. Please help the user below."},
{"role": "user", "content": f'''You are given the following set of tools:\n {tool_str} \n Can you please figure out which tools the query "{user_query}" will require to solve, out of these tools? Please return only the tool names inside []. If it does not need any tool, return an empty list'''}]
return message
def tool_retriever(tools, user_query):
tool_str = create_tool_str(tools)
message = tool_retriever_prompt(tool_str, user_query)
output = client_conn(message, model_name_tr)
output = output.choices[0].message.content
list_delim=output[output.find('[')+1:output.find(']')]
tools_retrieved = set()
for i in list_delim.split(','):
if (i.strip()!=''):
tools_retrieved.add(i.strip().replace('\'','').replace('\"',''))
return tools_retrieved
def final_tools(tools, tool_names):
tool = []
for tool_name in tool_names:
tool.append(tools[tool_name])
for tool_check in list(tools.values()):
if not tool_check["argument_list"]:
tool.append(tool_check)
return tool
# -------------------------------------------------------bonus section prompts---------------------------------------------------------------
def create_prompt_for_query_bonus(user_query, examples, tools):
message = [
{
"role": "system",
"content": f"""The following is a friendly conversation between a human and an AI.
The AI is professional and parses user input to several tasks. If the AI does not
know the answer to a question, it truthfully says it does not know. The AI will be
provided with a set of tools their descriptions and the argument in them. Here is
the list of tools:"+ {json.dumps(tools)} + "\n Provide the answer in the
exact format as given in the following examples. \nExamples """
}
]
for example in examples:
query = example['query']
answer = example['answer']
user_prompt = "Query: "+ str(query)
assistant_prompt = str(answer)
message.append(
{
"role" : "user",
"content": user_prompt
})
message.append(
{
"role" : "assistant",
"content": assistant_prompt
})
message.append(
{
"role":"user",
"content":"Use the above tools to learn how to use the tool on any query. Analyse how to parse the query and extract the correct information and place in the argument name and value. Use all the required tools and arguments in correct order of its calling based on the query and your learning from all the examples. Do not assume any value, you can take the value from query or the previous called tool as shown in the examples. Also focus on the allowed values argument present in tool definition."
})
message.append(
{
"role":"user",
"content":f"After producing the list of tools, analyze the query and figure out whether it requires the combination of tool outputs via mathematical operations, iterations, conditional logic etc. or not. In case it does, use the lambda function to produce the required results. Examples of such queries are given below: \n "
})
message.append(
{
"role":"user",
"content":f"Find all tasks created by user 'USER-321' and check if there are more than 10 such tasks"
}
)
message.append(
{
"role":"assistant",
"content":"""
"answer": [
{
"tool_name": "search_object_by_name",
"arguments": [
{
"argument_name": "query",
"argument_value": "USER-321"
}
]
},
{
"tool_name": "works_list",
"arguments": [
{
"argument_name": "created_by",
"argument_value": [
"$$PREV[0]"
]
},
{
"argument_name": "type",
"argument_value": [
"task"
]
}
]},
{
"tool_name": "lambda",
"arguments": [
{
"argument_name": "expression",
"argument_value": "lambda $$PREV[1]: True if len($$PREV[1]) > 10 else False"
}
]
}
]
"""
}
)
message.append({
"role" : "user",
"content" : "Now its your task to respond to the user queries in the same format as that in the above examples which is json. Use the lambda function only when necessary."
})
message.append({
"role" : "user",
"content" : "Query: "+ user_query
})
message.append(
{
"role":"system",
"content": "Generate the answer in a json format only. Enclose the strings in double quotes"
})
return message
# -------------------------------------------------miscellaneous functions------------------------------------------------------------
def create_tool_dict(tools):
tool_dict = {}
for i in tools:
tool_dict[i["tool_name"]] = i
return tool_dict
def get_parsed_json(text_parsed):
ans_list = []
json_str = text_parsed.split('```json')
for i in json_str:
if "```" in i:
json_data = i.split("```")[0].strip()
if json_data: # Check if the JSON data is not empty
json_obj = json.loads(json_data)
if type(json_obj)==list:
ans_list.extend(json_obj)
else:
ans_list.append(json_obj)
return ans_list
# -------------------------------------------------postprocessing---------------------------------------------------------------------
def get_json(pred):
'''
parameters
-----------------
pred: str
Answer predicted by the LLM as a string
returns
-----------------
json_pred: list
List of dictionaries that represents the input string as a json
'''
try:
# Tries to find ```json ``` type json format
return json.loads(pred[pred.find('```json'):-1*("".join(reversed(pred)).find('```')+1)])
except:
# Tries to find first instance of '[' from the left and first instance of ']' from the right, and converts all in between ito a json.
try:
return json.loads(pred[pred.find('['):-1*("".join(reversed(pred)).find(']')+1)] + ']')
except:
# Tries to fix keys/ strings being wrapped in single quotes, then tries to decode as above
pred= pred.replace('\'','\"')
try:
return json.loads(pred[pred.find('['):-1*("".join(reversed(pred)).find(']')+1)] + ']')
# Tries to check for instances of boolean values true and false, that might have been misspelt as True and False
except:
pred = pred.replace("True", "true").replace("False", "false")
return json.loads(pred[pred.find('['):-1*("".join(reversed(pred)).find(']')+1)] + ']')
def dict_unwrap(json_pred):
# Unwraps a dictionary if GPT-4 outputs one instead of a list
if type(json_pred)==dict:
for key in json_pred.keys():
if type(json_pred[key])==list:
print(json_pred[key])
return json_pred[key]
return []
return json_pred
def list_in_str_handler(json_pred):
'''
parameters
-----------------
json_pred: list
List of dictionaries that represents the tool call sequence
-----------------
json_pred: list
Tool call sequence with string arguments like '[arg_val]' turned into 'arg_val'
'''
# Iterate over tool call sequence and get argument values
for i, tool in enumerate(json_pred):
for j, arg in enumerate(tool["arguments"]):
arg_val= arg["argument_value"]
# If value is a string and it starts with [ and ends with ], remove them
if (type(arg_val)==str):
if arg_val.startswith('[') and arg_val.endswith(']'):
arg["argument_value"]=arg_val[1:-1]
# If value is a list, then iterate over it and perform similar operations as above
elif (type(arg_val)==list):
for num_item,arg_val_item in enumerate(arg_val):
if (type(arg_val_item)==str):
if arg_val_item.startswith('[') and arg_val_item.endswith(']'):
arg["argument_value"][num_item]=arg_val_item[1:-1]
return json_pred
def func_name_handler(json_pred,tools,no_arg_tool_list):
'''
parameters
-----------------
json_pred: list
List of dictionaries that represents the tool call sequence
tools: dict
Dict representing tools
no_arg_tool_list: list
List of tool names that do not have arguments
returns
-----------------
json_pred: list
Tool call sequence with arguments with $${function_name} errors removed
'''
# Iteration variable
i=0
# WHILE loop is required, len(range()) does not work as the loop condition is kept static while items are inserted into the loop.
while(i<len(json_pred)):
tool=json_pred[i]
for j, arg in enumerate(tool["arguments"]):
if arg["argument_name"] in tools[tool["tool_name"]]["args"] :
# Check argument value
temp = arg["argument_value"]
if type(temp)==str:
# If argument value starts with $$ but is not $$PREV[i]
if (temp.startswith('$$')) and not temp.startswith('$$PREV['):
# Remove the $$
temp_lowercase_call=temp.lower()[2:]
# Iterate over tools with no arguments to figure the appropriate tool to call
for no_arg_tool in no_arg_tool_list:
if temp_lowercase_call.startswith(no_arg_tool):
# Create tool call
tool_ins = {}
tool_ins['arguments']=[]
tool_ins['tool_name']=no_arg_tool
# Insert tool into the list
json_pred.insert(i,tool_ins)
i+=1
# Iterate over the tools, starting from the tool under consideration
for i_n, tool_n in enumerate(json_pred[i:]):
for j_n, arg_n in enumerate(tool_n["arguments"]):
# Check the argument value
prevset = arg_n["argument_value"]
# If the argument value is
if (type(prevset)==str):
# If the argument is of $$PREV[i] type, and it referenced the returned value of the tool that
# came at the same position as, or after the inserted tool, increment i for it,
if prevset.startswith("$$PREV["):
try:
n=int(prevset[7:-1])
except:
indexing_pos=prevset.find('][')
try:
n=int(prevset[7:indexing_pos])
except:
pass
if n>=i-1:
arg_n["argument_value"]=f"$$PREV[{n}]"
# If the argument value is a list, iterate over the values and perform the same process as above.
elif type(prevset)==list:
for list_num,prev_val in enumerate(prevset):
if prev_val.startswith("$$PREV["):
n=0
try:
n=int(prevset[7:-1])
except:
indexing_pos=prevset.find('][')
try:
n=int(prevset[7:indexing_pos])
except:
pass
if n>=i-1:
arg_n["argument_value"][list_num]=f"$$PREV[{n}]"
arg["argument_value"]=f"$$PREV[{i-1}]"
# otherwise the argument value is a list, iterate over this list, and perform the same operations as above.
elif (type(temp)==list):
for num_arg,temp_el in enumerate(temp):
if type(temp_el)==str:
if (temp_el.startswith('$$')) and not temp_el.startswith('$$PREV'):
temp_lowercase_call=temp_el.lower()[2:]
for no_arg_tool in no_arg_tool_list:
if temp_lowercase_call.startswith(no_arg_tool):
tool_ins = {}
tool_ins['arguments']=[]
tool_ins['tool_name']=no_arg_tool
json_pred.insert(i,tool_ins)
i+=1
for i_n, tool_n in enumerate(json_pred[i:]):
for j_n, arg_n in enumerate(tool_n["arguments"]):
prevset = arg_n["argument_value"]
if (type(prevset) not in [list,bool,float]):
if prevset.startswith("$$PREV[") and int(prevset[7:-1])>=i-1:
n=int(prevset[7:-1])+1
arg_n["argument_value"]=f"$$PREV[{n}]"
elif type(prevset) not in [bool,float]:
for list_num,prev_val in enumerate(prevset):
try:
if prev_val.startswith("$$PREV[") and int(prev_val[7:-1])>=i-1:
n=int(prevset[7:-1])+1
arg_n["argument_value"][list_num]=f"$$PREV[{n}]"
except:
pass
arg["argument_value"][num_arg]=f"$$PREV[{i-1}]"
i+=1
return json_pred
def unknown_tool_remover(json_pred,tools):
for i,tool_call in enumerate(json_pred):
if tool_call["tool_name"] not in tools.keys():
return True
def type_handler(json_pred,tools,array_check,string_check,num_check,bool_check,string_to_boolean):
'''
parameters
-----------------
json_pred: list
List of dictionaries that represents the tool call sequence
tools: dict
Dict representing tools
array_check: list
List of keywords to check for array return types
string_check: list
List of keywords to check for string return types
num_check: list
List of keywords to check for numeral return types
bool_check: list
List of keywords to check for boolean return types
string_to_boolean: dict
Dictionary mapping common representations of True/False values to booleans
returns
-----------------
json_pred: list
Tool call sequence with tool inputs respecting tool input type requirements
'''
for i, tool in enumerate(json_pred):
for j, arg in enumerate(tool["arguments"]):
if arg["argument_name"] in tools[tool["tool_name"]]["args"] :
arg_type = tools[tool["tool_name"]]["args"][arg["argument_name"]]["argument_type"]
temp = json_pred[i]["arguments"][j]["argument_value"]
# Split the argument type by spaces for easier checks later
typcheck = set(arg_type.lower().split(' '))
# To check if arg_type is supposed to be a list, but it is not
if typcheck.intersection(array_check) and type(temp)!=list:
# If argument is not $$PREV type and is a string, convert to array of strings
if not temp.startswith("$$") and typcheck.intersection(string_check) :
temp = [str(temp)]
# If argument type has integer, convert to list of integers
elif typcheck.intersection(num_check):
if type(temp) in [int,float]:
temp = [temp]
else:
try:
temp = [int(temp)]
except:
pass
# If argument type has boolean, convert to list of booleans
elif typcheck.intersection(bool_check):
try:
if type(temp)==str:
temp = [string_to_boolean.get(temp,temp)]
else:
temp=[bool(temp)]
except:
pass
# If argument type is string, convert to string
elif typcheck.intersection(string_check) and type(temp) != str:
# If the argument type is currently a list, convert to string. Only the first argument is going to be considered.
if (type(temp)==list and not temp):
try:
if (not temp[0].startswith("$$")):
temp = str(temp[0])
except:
pass
if (type(temp) in [int,float]):
temp=str(temp)
# If argument type is boolean, convert to boolean
elif typcheck.intersection(bool_check) and type(temp)!= bool :
# If the argument type is currently a list, convert to boolean. Only the first argument is going to be considered.
if (type(temp)==list and temp):
if (type(temp[0])==str):
if not temp[0].startswith("$$"):
try:
temp = string_to_boolean.get(temp[0], temp)
except:
pass
elif (type(temp[0])==bool):
temp=temp[0]
elif (type(temp[0]) in [int,float]):
temp=bool(temp[0])
# If the argument type is currently a string, convert to boolean
elif type(temp)==str:
temp=string_to_boolean.get(temp,temp)
json_pred[i]["arguments"][j]["argument_value"] = temp
else:
print("arg name not match for",arg)
return json_pred
def prev_ret_type_handler(json_pred,tools,array_check,string_check,num_check,bool_check):
'''
parameters
-----------------
json_pred: list
List of dictionaries that represents the tool call sequence
tools: dict
Dict representing tools
array_check: list
List of keywords to check for array return types
string_check: list
List of keywords to check for string return types
num_check: list
List of keywords to check for numeral return types
bool_check: list
List of keywords to check for boolean return types
returns
-----------------
json_pred: list
Tool call sequence with $$PREV[i] type arguments respecting tool input type requirements
'''
# Iterate over list of dictionaries
for i, tool in enumerate(json_pred):
for j, arg in enumerate(tool["arguments"]):
if arg["argument_name"] in tools[tool["tool_name"]]["args"] :
# Fetch argument type, and the current argument value
arg_type = set(tools[tool["tool_name"]]["args"][arg["argument_name"]]["argument_type"].split(' '))
temp = json_pred[i]["arguments"][j]["argument_value"]
# If current argument is a string
if type(temp)== str:
# If it is a $$PREV type argument
if temp.startswith('$$PREV'):
# Identify the return type of the tool call referenced by $$PREV[i],
refer=temp[7:-1]
ref_type= set(tools[json_pred[int(refer)]["tool_name"]]["return_type"].split(' '))
# If the referenced tool does not return an array, but the argument expects one
if not ref_type.intersection(array_check) and arg_type.intersection(array_check):
json_pred[i]["arguments"][j]["argument_value"]=[temp]
# Otherwise, if the argument value is a list, iterate over it
elif (type(temp)==list):
for temp_el in temp:
# Perform similar actions as above
if type(temp_el)==str:
if temp_el.startswith('$$PREV'):
refer=temp_el.lower()[7:-1]
ref_type= set(tools[json_pred[int(refer)]["tool_name"]]["return_type"].split(' '))
if (ref_type.intersection(array_check) and arg_type.intersection(array_check)) or (not ref_type.intersection(array_check) and not arg_type.intersection(array_check)):
json_pred[i]["arguments"][j]["argument_value"]=temp_el
return json_pred
def postprocess(json_pred, tool_data):
'''
parameters
-----------------
json_pred: str
Answer predicted by the LLM as a string
tool_data: dict
dictionary representing the given tools
returns
-----------------
json_pred: list
List of dictionaries that represents the final answer
'''
# turn json in string format into list of dicts
# Get dictionary of tools to simplify work
json_pred = dict_unwrap(json_pred)
tools = {}
for i,tool in enumerate(tool_data):
tools[tool["tool_name"]] = tool
tools[tool["tool_name"]]["args"] = {}
tools[tool["tool_name"]]["return_type"] = tool["return_type"]
for arg in tool["argument_list"]:
tools[tool["tool_name"]]["args"][arg["argument_name"]] = arg
for tool in json_pred:
if not tool.get('arguments',None):
tool["arguments"] = []
# Lists holding keywords to search for in the argument types/ return types
array_check = ["array","list","arrays","lists"]
string_check=["string","str","strings"]
num_check=["integer","int32","number","float","double","float32"]
bool_check=["bool","boolean","true","false"]
# Dictionary mapping common strings to boolean values
string_to_boolean = {"True": True, "False": False, "1": True, "0": False, "yes": True, "no": False, "true" : True, "false": False, 'True':True, 'False':False}
# Get tools for which no argument is required, to fix $${function_name} errors encountered
no_arg_tool_list = []
for i,tool in enumerate(tool_data):
if not tool["argument_list"]:
no_arg_tool_list.append(tool["tool_name"])
try:
if unknown_tool_remover(json_pred, tools):
return []
except:
pass
try:
json_pred = list_in_str_handler(json_pred)
except:
pass
try:
# Fix type errors
json_pred = type_handler(json_pred,tools,array_check,string_check,num_check,bool_check,string_to_boolean)
except:
pass
try:
# Fix $${function_name} errors
json_pred = func_name_handler(json_pred,tools,no_arg_tool_list)
except:
pass
try:
# Fix return types for $$PREV[i] type arguments
json_pred = prev_ret_type_handler(json_pred,tools,array_check,string_check,num_check,bool_check)
except:
pass
return json_pred
tools = read_file(tool_list_path)
examples = read_file(example_path)
tool_dict = create_tool_dict(tools["tools"])
tool_obj = Tools(tool_dict, examples)
def main(query):
if len(tool_obj.examples)<5:
message = create_prompt_zero_shot(query,tool_obj.tools)
res = client_conn(message, model_name_tr)
answer = []
try:
answer = json.loads(res.choices[0].message.content)
answer = postprocess(answer, list(tool_obj.tools.values())) # postprocessing
except:
pass
return answer
topk_examples = get_topk_given_query(query, tool_obj.queries, tool_obj.search_index, tool_obj.examples)
reduced_tools = tool_retriever(list(tool_obj.tools.values()), query)
if not reduced_tools:
return []
tool_list = final_tools(tool_obj.tools, list(tool_obj.tools.keys()))
if 'lambda' in reduced_tools:
message = create_prompt_for_query_bonus(query, topk_examples, tool_list)
else:
message = create_prompt_for_query(query, topk_examples, tool_list)
res = client_conn(message, model_name_tr)
print("GPT4 Response: ",res.choices[0].message.content)
answer = []
try:
answer = json.loads(res.choices[0].message.content)
answer = postprocess(answer, list(tool_obj.tools.values())) # postprocessing
except:
pass
return answer |