ExternalVendorCode/lollms
1
0
Fork 0
mirror of https://github.com/ParisNeo/lollms.git synced 2024-12-26 15:51:09 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

added skills library

Browse Source
This commit is contained in:
Saifeddine ALOUI 2024-03-16 00:31:21 +01:00
parent 1d16bc48c1
commit 4a70abd9b3
2 changed files with 22 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
lollms/app.py
View File

@ -529,6 +529,7 @@ class LollmsApplication(LoLLMsCom):
internet_search_infos = []
documentation = ""
knowledge = ""
knowledge_infos = {"titles":[],"contents":[]}
# boosting information
@ -633,15 +634,17 @@ class LollmsApplication(LoLLMsCom):
self.personality.step_start("Building skills library")
if discussion is None:
discussion = self.recover_discussion(client_id)
query = self.personality.fast_gen(f"!@>discussion:\n{discussion[-2048:]}\n!@>system: Read the discussion and craft a short skills database search query suited to recover needed information to reply to last {self.config.user_name} message.\nDo not answer the prompt. Do not add explanations.\n!@>search query: ", max_generation_size=256, show_progress=True, callback=self.personality.sink)
query = self.personality.fast_gen(f"!@>system: Read the discussion and reformulate {self.config.user_name}'s request.\nDo not answer the request.\nDo not add explanations.\n!@>discussion:\n{discussion[-2048:]}\n!@>search query: ", max_generation_size=256, show_progress=True, callback=self.personality.sink)
# skills = self.skills_library.query_entry(query)
skills, sorted_similarities, document_ids = self.skills_library.query_vector_db(query, top_k=3, max_dist=1000)#query_entry_fts(query)
if self.config.debug:
ASCIIColors.info(f"Query : {query}")
skill_titles, skills = self.skills_library.query_vector_db(query, top_k=3, max_dist=1000)#query_entry_fts(query)
knowledge_infos={"titles":skill_titles,"contents":skills}
if len(skills)>0:
if knowledge=="":
knowledge=f"!@>knowledge:\n!@>instructions: Use the knowledge to answer {self.config.user_name}'s message. If you don't have enough information or you don't know how to answer, just say you do not know.\n"
for i,(category, title, content) in enumerate(skills):
knowledge += f"!@>knowledge {i}:\n!@>category:\n{category}\n!@>title:\n{title}\ncontent:\n{content}"
knowledge=f"!@>knowledge:\n"
for i,(title, content) in enumerate(zip(skill_titles,skills)):
knowledge += f"!@>knowledge {i}:\n!@>title:\n{title}\ncontent:\n{content}"
self.personality.step_end("Building skills library")
except Exception as ex:
ASCIIColors.error(ex)
@ -803,6 +806,7 @@ class LollmsApplication(LoLLMsCom):
"internet_search_results":internet_search_results,
"documentation":documentation,
"knowledge":knowledge,
"knowledge_infos":knowledge_infos,
"user_description":user_description,
"discussion_messages":discussion_messages,
"positive_boost":positive_boost,

24
lollms/databases/skills_database.py
View File

@ -1,11 +1,11 @@
import sqlite3
from safe_store.text_vectorizer import TextVectorizer, VectorizationMethod, VisualizationMethod
import numpy as np
class SkillsLibrary:
def __init__(self, db_path):
self.db_path =db_path
self._initialize_db()
self.vectorizer = TextVectorizer(VectorizationMethod.TFIDF_VECTORIZER)
def _initialize_db(self):
@ -120,33 +120,33 @@ class SkillsLibrary:
conn.close()
return res
def query_vector_db(self, query, top_k=3, max_dist=1000):
def query_vector_db(self, query_, top_k=3, max_dist=1000):
vectorizer = TextVectorizer(VectorizationMethod.TFIDF_VECTORIZER)
conn = sqlite3.connect(self.db_path)
cursor = conn.cursor()
# Use direct string concatenation for the MATCH expression.
# Ensure text is safely escaped to avoid SQL injection.
query = "SELECT title FROM skills_library"
query = "SELECT id, title FROM skills_library"
cursor.execute(query)
res = cursor.fetchall()
cursor.close()
conn.close()
for entry in res:
self.vectorizer.add_document(entry[0])
self.vectorizer.index()
vectorizer.add_document(entry[0],entry[1])
vectorizer.index()
skill_titles, sorted_similarities, document_ids = self.vectorizer.recover_text(query, top_k)
skill_titles, sorted_similarities, document_ids = vectorizer.recover_text(query_, top_k)
skills = []
for skill, sim in zip(skill_titles, sorted_similarities):
if sim>max_dist:
for skill_title, sim, id in zip(skill_titles, sorted_similarities, document_ids):
if np.linalg.norm(sim[1])<max_dist:
conn = sqlite3.connect(self.db_path)
cursor = conn.cursor()
# Use direct string concatenation for the MATCH expression.
# Ensure text is safely escaped to avoid SQL injection.
query = "SELECT content FROM skills_library WHERE title LIKE ?"
res = cursor.execute(query, (skill,))
skills.append(res[0])
cursor.execute(query)
query = "SELECT content FROM skills_library WHERE id = ?"
cursor.execute(query, (id,))
res = cursor.fetchall()
skills.append(res[0])
cursor.close()
conn.close()
return skill_titles, skills