app
Browse files- .gitignore +6 -0
- Dockerfile +13 -0
- app.py +277 -0
- requirements.txt +15 -0
.gitignore
ADDED
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
data
|
2 |
+
.sesskey
|
3 |
+
__pycache__
|
4 |
+
*.pyc
|
5 |
+
.env
|
6 |
+
log_data
|
Dockerfile
ADDED
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
FROM python:3.10
|
2 |
+
|
3 |
+
WORKDIR /code
|
4 |
+
|
5 |
+
COPY --link --chown=1000 . .
|
6 |
+
|
7 |
+
RUN mkdir -p /tmp/cache/
|
8 |
+
RUN chmod a+rwx -R /tmp/cache/
|
9 |
+
ENV HF_HUB_CACHE=HF_HOME
|
10 |
+
RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt
|
11 |
+
|
12 |
+
ENV PYTHONUNBUFFERED=1 PORT=7860
|
13 |
+
CMD ["python", "app.py"]
|
app.py
ADDED
@@ -0,0 +1,277 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
import os
|
3 |
+
from datetime import datetime
|
4 |
+
|
5 |
+
import dotenv
|
6 |
+
import lancedb
|
7 |
+
from datasets import load_dataset
|
8 |
+
from fasthtml.common import * # noqa
|
9 |
+
from huggingface_hub import login, whoami
|
10 |
+
from rerankers import Reranker
|
11 |
+
|
12 |
+
dotenv.load_dotenv()
|
13 |
+
login(token=os.environ.get("HF_TOKEN"))
|
14 |
+
|
15 |
+
hf_user = whoami(os.environ.get("HF_TOKEN"))["name"]
|
16 |
+
HF_REPO_ID_TXT = f"{hf_user}/zotero-answer-ai-texts"
|
17 |
+
|
18 |
+
abstract_ds = load_dataset(HF_REPO_ID_TXT, "abstracts")["train"]
|
19 |
+
article_ds = load_dataset(HF_REPO_ID_TXT, "articles")["train"]
|
20 |
+
|
21 |
+
ranker = Reranker("answerdotai/answerai-colbert-small-v1", model_type="colbert")
|
22 |
+
|
23 |
+
|
24 |
+
uri = "data/zotero-fts"
|
25 |
+
db = lancedb.connect(uri)
|
26 |
+
|
27 |
+
id2abstract = {example["arxiv_id"]: example["abstract"] for example in abstract_ds}
|
28 |
+
id2content = {example["arxiv_id"]: example["contents"] for example in article_ds}
|
29 |
+
id2title = {example["arxiv_id"]: example["title"] for example in article_ds}
|
30 |
+
|
31 |
+
arxiv_ids = set(list(id2abstract.keys()))
|
32 |
+
|
33 |
+
data = []
|
34 |
+
for arxiv_id in arxiv_ids:
|
35 |
+
abstract = id2abstract[arxiv_id]
|
36 |
+
title = id2title[arxiv_id]
|
37 |
+
full_text = title
|
38 |
+
|
39 |
+
for item in id2content[arxiv_id]:
|
40 |
+
full_text += f"{item['title']}\n\n{item['content']}"
|
41 |
+
|
42 |
+
data.append(
|
43 |
+
{
|
44 |
+
"arxiv_id": arxiv_id,
|
45 |
+
"title": title,
|
46 |
+
"abstract": abstract,
|
47 |
+
"full_text": full_text,
|
48 |
+
}
|
49 |
+
)
|
50 |
+
|
51 |
+
|
52 |
+
table = db.create_table("articles", data=data, mode="overwrite")
|
53 |
+
|
54 |
+
table.create_fts_index("full_text", replace=True)
|
55 |
+
|
56 |
+
|
57 |
+
# format results ----
|
58 |
+
def _format_results(results):
|
59 |
+
ret = []
|
60 |
+
|
61 |
+
for result in results:
|
62 |
+
arx_id = result["arxiv_id"]
|
63 |
+
title = result["title"]
|
64 |
+
abstract = result["abstract"]
|
65 |
+
|
66 |
+
if "Abstract\n\n" in abstract:
|
67 |
+
abstract = abstract.split("Abstract\n\n")[-1]
|
68 |
+
|
69 |
+
this_ex = {
|
70 |
+
"title": title,
|
71 |
+
"url": f"https://arxiv.org/abs/{arx_id}",
|
72 |
+
"abstract": abstract,
|
73 |
+
}
|
74 |
+
|
75 |
+
ret.append(this_ex)
|
76 |
+
|
77 |
+
return ret
|
78 |
+
|
79 |
+
|
80 |
+
def retrieve_and_rerank(query, k=5):
|
81 |
+
# retrieve ---
|
82 |
+
n_fetch = 25
|
83 |
+
|
84 |
+
retrieved = (
|
85 |
+
table.search(query, vector_column_name="", query_type="fts")
|
86 |
+
.limit(n_fetch)
|
87 |
+
.select(["arxiv_id", "title", "abstract"])
|
88 |
+
.to_list()
|
89 |
+
)
|
90 |
+
|
91 |
+
# re-rank
|
92 |
+
docs = [f"{item['title']} {item['abstract']}" for item in retrieved]
|
93 |
+
results = ranker.rank(query=query, docs=docs)
|
94 |
+
|
95 |
+
ranked_doc_ids = []
|
96 |
+
for result in results[:k]:
|
97 |
+
ranked_doc_ids.append(result.doc_id)
|
98 |
+
|
99 |
+
final_results = [retrieved[idx] for idx in ranked_doc_ids]
|
100 |
+
final_results = _format_results(final_results)
|
101 |
+
return final_results
|
102 |
+
|
103 |
+
|
104 |
+
###########################################################################
|
105 |
+
# FastHTML app -----
|
106 |
+
###########################################################################
|
107 |
+
|
108 |
+
style = Style("""
|
109 |
+
:root {
|
110 |
+
color-scheme: dark;
|
111 |
+
}
|
112 |
+
body {
|
113 |
+
max-width: 1200px;
|
114 |
+
margin: 0 auto;
|
115 |
+
padding: 20px;
|
116 |
+
line-height: 1.6;
|
117 |
+
}
|
118 |
+
#query {
|
119 |
+
width: 100%;
|
120 |
+
margin-bottom: 1rem;
|
121 |
+
}
|
122 |
+
#search-form button {
|
123 |
+
width: 100%;
|
124 |
+
}
|
125 |
+
#search-results, #log-entries {
|
126 |
+
margin-top: 2rem;
|
127 |
+
}
|
128 |
+
.log-entry {
|
129 |
+
border: 1px solid #ccc;
|
130 |
+
padding: 10px;
|
131 |
+
margin-bottom: 10px;
|
132 |
+
}
|
133 |
+
.log-entry pre {
|
134 |
+
white-space: pre-wrap;
|
135 |
+
word-wrap: break-word;
|
136 |
+
}
|
137 |
+
.htmx-indicator {
|
138 |
+
display: none;
|
139 |
+
}
|
140 |
+
.htmx-request .htmx-indicator {
|
141 |
+
display: inline-block;
|
142 |
+
}
|
143 |
+
.spinner {
|
144 |
+
display: inline-block;
|
145 |
+
width: 2.5em;
|
146 |
+
height: 2.5em;
|
147 |
+
border: 0.3em solid rgba(255,255,255,.3);
|
148 |
+
border-radius: 50%;
|
149 |
+
border-top-color: #fff;
|
150 |
+
animation: spin 1s ease-in-out infinite;
|
151 |
+
margin-left: 10px;
|
152 |
+
vertical-align: middle;
|
153 |
+
}
|
154 |
+
@keyframes spin {
|
155 |
+
to { transform: rotate(360deg); }
|
156 |
+
}
|
157 |
+
.searching-text {
|
158 |
+
font-size: 1.2em;
|
159 |
+
font-weight: bold;
|
160 |
+
color: #fff;
|
161 |
+
margin-right: 10px;
|
162 |
+
vertical-align: middle;
|
163 |
+
}
|
164 |
+
""")
|
165 |
+
|
166 |
+
# get the fast app and route
|
167 |
+
app, rt = fast_app(hdrs=(style,))
|
168 |
+
|
169 |
+
# Initialize a database to store search logs --
|
170 |
+
db = database("log_data/search_logs.db")
|
171 |
+
search_logs = db.t.search_logs
|
172 |
+
|
173 |
+
if search_logs not in db.t:
|
174 |
+
search_logs.create(
|
175 |
+
id=int,
|
176 |
+
timestamp=str,
|
177 |
+
query=str,
|
178 |
+
results=str,
|
179 |
+
pk="id",
|
180 |
+
)
|
181 |
+
|
182 |
+
SearchLog = search_logs.dataclass()
|
183 |
+
|
184 |
+
|
185 |
+
def insert_log_entry(log_entry):
|
186 |
+
"Insert a log entry into the database"
|
187 |
+
return search_logs.insert(
|
188 |
+
SearchLog(
|
189 |
+
timestamp=log_entry["timestamp"].isoformat(),
|
190 |
+
query=log_entry["query"],
|
191 |
+
results=json.dumps(log_entry["results"]),
|
192 |
+
)
|
193 |
+
)
|
194 |
+
|
195 |
+
|
196 |
+
@rt("/")
|
197 |
+
async def get():
|
198 |
+
query_form = Form(
|
199 |
+
Textarea(id="query", name="query", placeholder="Enter your query..."),
|
200 |
+
Button("Submit", type="submit"),
|
201 |
+
Div(
|
202 |
+
Span("Searching...", cls="searching-text htmx-indicator"),
|
203 |
+
Span(cls="spinner htmx-indicator"),
|
204 |
+
cls="indicator-container",
|
205 |
+
),
|
206 |
+
id="search-form",
|
207 |
+
hx_post="/search",
|
208 |
+
hx_target="#search-results",
|
209 |
+
hx_indicator=".indicator-container",
|
210 |
+
)
|
211 |
+
|
212 |
+
results_div = Div(Div(id="search-results", cls="results-container"))
|
213 |
+
|
214 |
+
view_logs_link = A("View Logs", href="/logs", cls="view-logs-link")
|
215 |
+
|
216 |
+
return Titled(
|
217 |
+
"Zotero Search", Div(query_form, results_div, view_logs_link, cls="container")
|
218 |
+
)
|
219 |
+
|
220 |
+
|
221 |
+
def SearchResult(result):
|
222 |
+
"Custom component for displaying a search result"
|
223 |
+
return Card(
|
224 |
+
H4(A(result["title"], href=result["url"], target="_blank")),
|
225 |
+
P(result["abstract"]),
|
226 |
+
footer=A("Read more →", href=result["url"], target="_blank"),
|
227 |
+
)
|
228 |
+
|
229 |
+
|
230 |
+
def log_query_and_results(query, results):
|
231 |
+
log_entry = {
|
232 |
+
"timestamp": datetime.now(),
|
233 |
+
"query": query,
|
234 |
+
"results": [{"title": r["title"], "url": r["url"]} for r in results],
|
235 |
+
}
|
236 |
+
insert_log_entry(log_entry)
|
237 |
+
|
238 |
+
|
239 |
+
@rt("/search")
|
240 |
+
async def post(query: str):
|
241 |
+
results = retrieve_and_rerank(query)
|
242 |
+
log_query_and_results(query, results)
|
243 |
+
|
244 |
+
return Div(*[SearchResult(r) for r in results], id="search-results")
|
245 |
+
|
246 |
+
|
247 |
+
def LogEntry(entry):
|
248 |
+
return Div(
|
249 |
+
H4(f"Query: {entry.query}"),
|
250 |
+
P(f"Timestamp: {entry.timestamp}"),
|
251 |
+
H5("Results:"),
|
252 |
+
Pre(entry.results),
|
253 |
+
cls="log-entry",
|
254 |
+
)
|
255 |
+
|
256 |
+
|
257 |
+
@rt("/logs")
|
258 |
+
async def get():
|
259 |
+
logs = search_logs(order_by="-id", limit=50) # Get the latest 50 logs
|
260 |
+
log_entries = [LogEntry(log) for log in logs]
|
261 |
+
return Titled(
|
262 |
+
"Logs",
|
263 |
+
Div(
|
264 |
+
H2("Recent Search Logs"),
|
265 |
+
Div(*log_entries, id="log-entries"),
|
266 |
+
A("Back to Search", href="/", cls="back-link"),
|
267 |
+
cls="container",
|
268 |
+
),
|
269 |
+
)
|
270 |
+
|
271 |
+
|
272 |
+
if __name__ == "__main__":
|
273 |
+
import uvicorn
|
274 |
+
|
275 |
+
uvicorn.run(app, host="0.0.0.0", port=int(os.environ.get("PORT", 7860)))
|
276 |
+
|
277 |
+
# run_uv()
|
requirements.txt
ADDED
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
fasthtml-hf>=0.1.1
|
2 |
+
python-fasthtml>=0.5.2
|
3 |
+
huggingface-hub>=0.20.0
|
4 |
+
uvicorn>=0.29
|
5 |
+
requests
|
6 |
+
srsly
|
7 |
+
python-dotenv
|
8 |
+
retry
|
9 |
+
pandas
|
10 |
+
datasets
|
11 |
+
tqdm
|
12 |
+
tantivy==0.22.0
|
13 |
+
lancedb
|
14 |
+
rerankers
|
15 |
+
transformers
|