-
Notifications
You must be signed in to change notification settings - Fork 0
/
coral.py
259 lines (194 loc) · 8.78 KB
/
coral.py
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
import logging, os
import cohere
import tomli
from dotenv import load_dotenv
from langchain.chat_models import ChatCohere
from langchain.document_loaders import ArxivLoader
from langchain.llms import Cohere
from langchain.output_parsers import PydanticOutputParser
from langchain.prompts import PromptTemplate
from langchain.retrievers import CohereRagRetriever
from langchain.schema.document import Document
from pydantic import BaseModel, Field, field_validator
from tenacity import retry, stop_after_attempt, wait_random_exponential
class Tweet(BaseModel):
"""
Pydantic Model to generate an structured Tweet with Validation
"""
text: str = Field(..., description="Tweet text")
@field_validator('text')
def validate_text(cls, v: str) -> str:
if "https://" not in v and "http://" not in v:
logging.error("Tweet does not include a link to the paper!")
raise ValueError("Tweet must include a link to the paper!")
return v
class Email(BaseModel):
"""
Pydantic Model to generate an structured Email
"""
subject: str = Field(..., description="Email subject")
body: str = Field(..., description="Email body")
class CohereEngine:
def __init__(self) -> None:
logging.basicConfig(level=logging.INFO,
format="%(asctime)s [%(levelname)s] %(message)s")
self.vars = self.__load_environment_vars()
self.cohere = self.__cohere_client(self.vars["COHERE_API_KEY"])
self.templates = self.__load_prompt_templates()
logging.info("Initialized CohereEngine")
@retry(wait=wait_random_exponential(min=1, max=5), stop=stop_after_attempt(3))
def query_article(self, article: str, query: str):
"""
Query Article.
Parameters:
- article (str): Article to query
- query (str): Query to search for
Returns:
- str: Relevant passages from the article
"""
logging.info("query_llm (started)")
rag = CohereRagRetriever(llm=ChatCohere())
docs = rag.get_relevant_documents(query,
source_documents=[Document(page_content=article)])
#ranked_docs = self.cohere.rerank(query=query, documents=docs, top_n=4, model="rerank-english-v2.0")
logging.info("query_llm (OK)")
return docs
@retry(wait=wait_random_exponential(min=1, max=5), stop=stop_after_attempt(3))
def generate_tweet(self, summary: str, link: str) -> Tweet:
"""
Generate an structured Tweet object about a research paper.
Under the hood it uses Cohere's LLM, a custom Pydantic Tweet Model, and Langchain Expression Language with Templates.
Parameters:
- summary (str): Summary of the research paper
- link (str): Link to the research paper
Returns:
- Tweet: Tweet object
"""
logging.info(f"generate_tweet ({link}) (started)")
model = Cohere(model='command', temperature=0.3, max_tokens=250)
prompt = PromptTemplate.from_template(self.templates['tweet']['prompt'])
parser = PydanticOutputParser(pydantic_object=Tweet)
tweet_chain = prompt | model | parser
tweet = tweet_chain.invoke({"summary": summary, "link": link})
logging.info("generate_tweet (OK)")
return tweet
@retry(wait=wait_random_exponential(min=1, max=5), stop=stop_after_attempt(3))
def generate_email(self, sender: str, institution: str, receivers: list, title: str, topic: str) -> Email:
"""
Generate an structured Email object to the authors of a research paper.
Under the hood it uses Cohere's LLM, a custom Pydantic Email Model, and Langchain Expression Language with Templates.
Parameters:
- sender (str): Name of the sender
- institution (str): Institution of the sender
- receivers (list): Names of the receivers
- title (str): Title of the research paper
- topic (str): Topic of the research paper
"""
logging.info("generate_email (started)")
model = Cohere(model='command', temperature=0.1, max_tokens=500)
prompt = PromptTemplate.from_template(self.templates['email']['prompt'])
parser = PydanticOutputParser(pydantic_object=Email)
email_chain = prompt | model | parser
email = email_chain.invoke({"sender": sender,
"institution": institution,
"receivers": receivers,
"title": title,
"topic": topic})
logging.info("generate_email (OK)")
return email
@retry(wait=wait_random_exponential(min=1, max=5), stop=stop_after_attempt(3))
def enrich_abstract(self, text: str) -> str:
"""
Identifies technical Named Entities, and enrich them with Wikipedia Links.
Parameters:
- text (str): Text to be enriched
Returns:
- str: Text enriched with Wikipedia links
"""
logging.info("enrich_abstract (started)")
model = Cohere(model='command', temperature=0.3, max_tokens=4096, truncate=None)
prompt = PromptTemplate.from_template(self.templates['abstract']['prompt'])
abstract_chain = prompt | model
abstract = abstract_chain.invoke({"text": text})
logging.info("enrich_abstract (OK)")
return abstract
@retry(wait=wait_random_exponential(min=1, max=5), stop=stop_after_attempt(3))
def extract_keywords(self, text: str) -> str:
"""
Extract keywords from a research paper. For each keyword, it provides a brief explanation of its significance in the context of this research.
Parameters:
- text (str): Text to extract keywords from
Returns:
- str: Keywords extracted from the text
"""
logging.info("extract_keywords (started)")
model = Cohere(model='command', temperature=0.1, max_tokens=4096)
prompt = PromptTemplate.from_template(self.templates['keywords']['prompt'])
keywords_chain = prompt | model
keywords = keywords_chain.invoke({"text": text})
logging.info("extract_keywords (OK)")
return keywords
@retry(wait=wait_random_exponential(min=1, max=5), stop=stop_after_attempt(3))
def summarize(self, text: str) -> str:
logging.info("summarize (started)")
response = self.cohere.summarize(
text = text,
length='auto',
format='bullets',
model='command',
additional_command='',
temperature=0.8,
)
logging.info("summarize (OK)")
return response.summary
@retry(wait=wait_random_exponential(min=1, max=5), stop=stop_after_attempt(3))
def embed(self, texts: dict) -> dict:
return self.cohere.embed(
model='embed-english-v3.0',
texts=texts,
input_type='search_document',
).embeddings
@retry(wait=wait_random_exponential(min=1, max=5), stop=stop_after_attempt(3))
def load_arxiv_paper(self, paper_id: str) -> (dict, str):
logging.info("load_arxiv_paper (started)")
docs = ArxivLoader(query=paper_id, load_max_docs=2, load_all_available_meta=True).load()
metadata = docs[0].metadata
content = docs[0].page_content
logging.info("load_arxiv_paper (OK)")
return metadata, content
def __load_environment_vars(self):
"""
Load environment variables from .env file
"""
logging.info("load_environment_vars (started)")
load_dotenv()
required_vars = ["COHERE_API_KEY"]
env_vars = {var: os.getenv(var) for var in required_vars}
for var, value in env_vars.items():
if not value:
raise EnvironmentError(f"{var} environment variable not set.")
logging.info("load_environment_vars (OK)")
return env_vars
def __load_prompt_templates(self):
"""
Load prompt templates from prompts/athena.toml
"""
logging.info("load_prompt_templates (started)")
try:
with open("prompts/athena.toml", "rb") as f:
prompts = tomli.load(f)
except FileNotFoundError as e:
logging.error(e)
raise OSError("Prompt templates file not found.")
logging.info("load_prompt_templates (OK)")
return prompts
@retry(wait=wait_random_exponential(min=1, max=5), stop=stop_after_attempt(5))
def __cohere_client(self, cohere_api_key):
"""
Initialize Cohere client
Parameters:
- cohere_api_key (str): Cohere API key
Returns:
- cohere.Client: Cohere client
"""
return cohere.Client(cohere_api_key)