Spaces:
Running
on
T4
Running
on
T4
some refactor
Browse files- .vscode/launch.json +14 -0
- app.py +21 -63
- src/templates.py +3 -11
- src/viewer_api.py +51 -0
.vscode/launch.json
ADDED
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"version": "0.2.0",
|
3 |
+
"configurations": [
|
4 |
+
{
|
5 |
+
"name": "Python Debugger: Current File",
|
6 |
+
"type": "debugpy",
|
7 |
+
"request": "launch",
|
8 |
+
"program": "${file}",
|
9 |
+
"console": "integratedTerminal",
|
10 |
+
"purpose": ["debug-test"],
|
11 |
+
"justMyCode": false
|
12 |
+
}
|
13 |
+
]
|
14 |
+
}
|
app.py
CHANGED
@@ -5,9 +5,7 @@ import logging
|
|
5 |
import os
|
6 |
|
7 |
import datamapplot
|
8 |
-
import duckdb
|
9 |
import numpy as np
|
10 |
-
import requests
|
11 |
|
12 |
from dotenv import load_dotenv
|
13 |
from gradio_huggingfacehub_search import HuggingfaceHubSearch
|
@@ -28,26 +26,28 @@ from transformers import (
|
|
28 |
|
29 |
from src.hub import create_space_with_content
|
30 |
from src.templates import REPRESENTATION_PROMPT, SPACE_REPO_CARD_CONTENT
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
31 |
|
|
|
32 |
load_dotenv()
|
33 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
34 |
assert HF_TOKEN is not None, "You need to set HF_TOKEN in your environment variables"
|
35 |
-
|
36 |
-
|
37 |
-
EXPORTS_REPOSITORY = os.getenv("EXPORTS_REPOSITORY")
|
38 |
-
assert (
|
39 |
-
EXPORTS_REPOSITORY is not None
|
40 |
-
), "You need to set EXPORTS_REPOSITORY in your environment variables"
|
41 |
-
|
42 |
MAX_ROWS = int(os.getenv("MAX_ROWS", "8_000"))
|
43 |
CHUNK_SIZE = int(os.getenv("CHUNK_SIZE", "2_000"))
|
44 |
-
DATASET_VIEWE_API_URL = "https://datasets-server.huggingface.co/"
|
45 |
DATASETS_TOPICS_ORGANIZATION = os.getenv(
|
46 |
"DATASETS_TOPICS_ORGANIZATION", "datasets-topics"
|
47 |
)
|
48 |
USE_ARROW_STYLE = int(os.getenv("USE_ARROW_STYLE", "0"))
|
49 |
-
USE_CUML = int(os.getenv("USE_CUML", "
|
|
|
50 |
|
|
|
51 |
if USE_CUML:
|
52 |
from cuml.manifold import UMAP
|
53 |
from cuml.cluster import HDBSCAN
|
@@ -55,14 +55,12 @@ else:
|
|
55 |
from umap import UMAP
|
56 |
from hdbscan import HDBSCAN
|
57 |
|
58 |
-
USE_LLM_TEXT_GENERATION = int(os.getenv("USE_LLM_TEXT_GENERATION", "1"))
|
59 |
|
60 |
logging.basicConfig(
|
61 |
level=logging.INFO, format="%(asctime)s - %(name)s - %(levelname)s - %(message)s"
|
62 |
)
|
63 |
|
64 |
api = HfApi(token=HF_TOKEN)
|
65 |
-
session = requests.Session()
|
66 |
sentence_model = SentenceTransformer("all-MiniLM-L6-v2")
|
67 |
|
68 |
# Representation model
|
@@ -98,41 +96,6 @@ else:
|
|
98 |
vectorizer_model = CountVectorizer(stop_words="english")
|
99 |
|
100 |
|
101 |
-
def get_split_rows(dataset, config, split):
|
102 |
-
config_size = session.get(
|
103 |
-
f"{DATASET_VIEWE_API_URL}/size?dataset={dataset}&config={config}",
|
104 |
-
timeout=20,
|
105 |
-
).json()
|
106 |
-
if "error" in config_size:
|
107 |
-
raise Exception(f"Error fetching config size: {config_size['error']}")
|
108 |
-
split_size = next(
|
109 |
-
(s for s in config_size["size"]["splits"] if s["split"] == split),
|
110 |
-
None,
|
111 |
-
)
|
112 |
-
if split_size is None:
|
113 |
-
raise Exception(f"Error fetching split {split} in config {config}")
|
114 |
-
return split_size["num_rows"]
|
115 |
-
|
116 |
-
|
117 |
-
def get_parquet_urls(dataset, config, split):
|
118 |
-
parquet_files = session.get(
|
119 |
-
f"{DATASET_VIEWE_API_URL}/parquet?dataset={dataset}&config={config}&split={split}",
|
120 |
-
timeout=20,
|
121 |
-
).json()
|
122 |
-
if "error" in parquet_files:
|
123 |
-
raise Exception(f"Error fetching parquet files: {parquet_files['error']}")
|
124 |
-
parquet_urls = [file["url"] for file in parquet_files["parquet_files"]]
|
125 |
-
logging.debug(f"Parquet files: {parquet_urls}")
|
126 |
-
return ",".join(f"'{url}'" for url in parquet_urls)
|
127 |
-
|
128 |
-
|
129 |
-
def get_docs_from_parquet(parquet_urls, column, offset, limit):
|
130 |
-
SQL_QUERY = f"SELECT {column} FROM read_parquet([{parquet_urls}]) LIMIT {limit} OFFSET {offset};"
|
131 |
-
df = duckdb.sql(SQL_QUERY).to_df()
|
132 |
-
return df[column].tolist()
|
133 |
-
|
134 |
-
|
135 |
-
# @spaces.GPU
|
136 |
def calculate_embeddings(docs):
|
137 |
return sentence_model.encode(docs, show_progress_bar=True, batch_size=32)
|
138 |
|
@@ -143,7 +106,6 @@ def calculate_n_neighbors_and_components(n_rows):
|
|
143 |
return n_neighbors, n_components
|
144 |
|
145 |
|
146 |
-
# @spaces.GPU
|
147 |
def fit_model(docs, embeddings, n_neighbors, n_components):
|
148 |
umap_model = UMAP(
|
149 |
n_neighbors=n_neighbors,
|
@@ -254,18 +216,16 @@ def generate_topics(dataset, config, split, column, nested_column, plot_type):
|
|
254 |
reduced_embeddings_array = np.vstack(reduced_embeddings_list)
|
255 |
|
256 |
topics_info = base_model.get_topic_info()
|
257 |
-
all_topics
|
258 |
-
all_topics = np.array(all_topics)
|
259 |
-
|
260 |
sub_title = (
|
261 |
f"Data map for the entire dataset ({limit} rows) using the column '{column}'"
|
262 |
if full_processing
|
263 |
else f"Data map for a sample of the dataset (first {limit} rows) using the column '{column}'"
|
264 |
)
|
265 |
-
|
266 |
topic_plot = (
|
267 |
base_model.visualize_document_datamap(
|
268 |
docs=all_docs,
|
|
|
269 |
reduced_embeddings=reduced_embeddings_array,
|
270 |
title=dataset,
|
271 |
sub_title=sub_title,
|
@@ -291,7 +251,6 @@ def generate_topics(dataset, config, split, column, nested_column, plot_type):
|
|
291 |
title=dataset,
|
292 |
)
|
293 |
)
|
294 |
-
|
295 |
rows_processed += len(docs)
|
296 |
progress = min(rows_processed / limit, 1.0)
|
297 |
logging.info(f"Progress: {progress} % - {rows_processed} of {limit}")
|
@@ -320,10 +279,10 @@ def generate_topics(dataset, config, split, column, nested_column, plot_type):
|
|
320 |
else:
|
321 |
topic_plot.write_image(plot_png)
|
322 |
|
323 |
-
all_topics
|
324 |
topic_info = base_model.get_topic_info()
|
325 |
|
326 |
-
topic_names = {row["Topic"]: row["Name"] for
|
327 |
topic_names_array = np.array(
|
328 |
[
|
329 |
topic_names.get(topic, "No Topic").split("_")[1].strip("-")
|
@@ -461,21 +420,20 @@ with gr.Blocks() as demo:
|
|
461 |
text_column_dropdown: gr.Dropdown(label="Text column name"),
|
462 |
nested_text_column_dropdown: gr.Dropdown(visible=False),
|
463 |
}
|
464 |
-
|
465 |
-
|
466 |
-
|
467 |
-
if "error" in info_resp:
|
468 |
return {
|
469 |
subset_dropdown: gr.Dropdown(visible=False),
|
470 |
split_dropdown: gr.Dropdown(visible=False),
|
471 |
text_column_dropdown: gr.Dropdown(label="Text column name"),
|
472 |
nested_text_column_dropdown: gr.Dropdown(visible=False),
|
473 |
}
|
474 |
-
subsets: list[str] = list(info_resp
|
475 |
subset = default_subset if default_subset in subsets else subsets[0]
|
476 |
-
splits: list[str] = list(info_resp[
|
477 |
split = default_split if default_split in splits else splits[0]
|
478 |
-
features = info_resp[
|
479 |
|
480 |
def _is_string_feature(feature):
|
481 |
return isinstance(feature, dict) and feature.get("dtype") == "string"
|
|
|
5 |
import os
|
6 |
|
7 |
import datamapplot
|
|
|
8 |
import numpy as np
|
|
|
9 |
|
10 |
from dotenv import load_dotenv
|
11 |
from gradio_huggingfacehub_search import HuggingfaceHubSearch
|
|
|
26 |
|
27 |
from src.hub import create_space_with_content
|
28 |
from src.templates import REPRESENTATION_PROMPT, SPACE_REPO_CARD_CONTENT
|
29 |
+
from src.viewer_api import (
|
30 |
+
get_split_rows,
|
31 |
+
get_parquet_urls,
|
32 |
+
get_docs_from_parquet,
|
33 |
+
get_info,
|
34 |
+
)
|
35 |
+
|
36 |
|
37 |
+
# Load environment variables
|
38 |
load_dotenv()
|
39 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
40 |
assert HF_TOKEN is not None, "You need to set HF_TOKEN in your environment variables"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
MAX_ROWS = int(os.getenv("MAX_ROWS", "8_000"))
|
42 |
CHUNK_SIZE = int(os.getenv("CHUNK_SIZE", "2_000"))
|
|
|
43 |
DATASETS_TOPICS_ORGANIZATION = os.getenv(
|
44 |
"DATASETS_TOPICS_ORGANIZATION", "datasets-topics"
|
45 |
)
|
46 |
USE_ARROW_STYLE = int(os.getenv("USE_ARROW_STYLE", "0"))
|
47 |
+
USE_CUML = int(os.getenv("USE_CUML", "1"))
|
48 |
+
USE_LLM_TEXT_GENERATION = int(os.getenv("USE_LLM_TEXT_GENERATION", "1"))
|
49 |
|
50 |
+
# Use cuml lib only if configured
|
51 |
if USE_CUML:
|
52 |
from cuml.manifold import UMAP
|
53 |
from cuml.cluster import HDBSCAN
|
|
|
55 |
from umap import UMAP
|
56 |
from hdbscan import HDBSCAN
|
57 |
|
|
|
58 |
|
59 |
logging.basicConfig(
|
60 |
level=logging.INFO, format="%(asctime)s - %(name)s - %(levelname)s - %(message)s"
|
61 |
)
|
62 |
|
63 |
api = HfApi(token=HF_TOKEN)
|
|
|
64 |
sentence_model = SentenceTransformer("all-MiniLM-L6-v2")
|
65 |
|
66 |
# Representation model
|
|
|
96 |
vectorizer_model = CountVectorizer(stop_words="english")
|
97 |
|
98 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
99 |
def calculate_embeddings(docs):
|
100 |
return sentence_model.encode(docs, show_progress_bar=True, batch_size=32)
|
101 |
|
|
|
106 |
return n_neighbors, n_components
|
107 |
|
108 |
|
|
|
109 |
def fit_model(docs, embeddings, n_neighbors, n_components):
|
110 |
umap_model = UMAP(
|
111 |
n_neighbors=n_neighbors,
|
|
|
216 |
reduced_embeddings_array = np.vstack(reduced_embeddings_list)
|
217 |
|
218 |
topics_info = base_model.get_topic_info()
|
219 |
+
all_topics = base_model.topics_
|
|
|
|
|
220 |
sub_title = (
|
221 |
f"Data map for the entire dataset ({limit} rows) using the column '{column}'"
|
222 |
if full_processing
|
223 |
else f"Data map for a sample of the dataset (first {limit} rows) using the column '{column}'"
|
224 |
)
|
|
|
225 |
topic_plot = (
|
226 |
base_model.visualize_document_datamap(
|
227 |
docs=all_docs,
|
228 |
+
topics=all_topics,
|
229 |
reduced_embeddings=reduced_embeddings_array,
|
230 |
title=dataset,
|
231 |
sub_title=sub_title,
|
|
|
251 |
title=dataset,
|
252 |
)
|
253 |
)
|
|
|
254 |
rows_processed += len(docs)
|
255 |
progress = min(rows_processed / limit, 1.0)
|
256 |
logging.info(f"Progress: {progress} % - {rows_processed} of {limit}")
|
|
|
279 |
else:
|
280 |
topic_plot.write_image(plot_png)
|
281 |
|
282 |
+
all_topics = base_model.topics_
|
283 |
topic_info = base_model.get_topic_info()
|
284 |
|
285 |
+
topic_names = {row["Topic"]: row["Name"] for _, row in topic_info.iterrows()}
|
286 |
topic_names_array = np.array(
|
287 |
[
|
288 |
topic_names.get(topic, "No Topic").split("_")[1].strip("-")
|
|
|
420 |
text_column_dropdown: gr.Dropdown(label="Text column name"),
|
421 |
nested_text_column_dropdown: gr.Dropdown(visible=False),
|
422 |
}
|
423 |
+
try:
|
424 |
+
info_resp = get_info(dataset)
|
425 |
+
except Exception:
|
|
|
426 |
return {
|
427 |
subset_dropdown: gr.Dropdown(visible=False),
|
428 |
split_dropdown: gr.Dropdown(visible=False),
|
429 |
text_column_dropdown: gr.Dropdown(label="Text column name"),
|
430 |
nested_text_column_dropdown: gr.Dropdown(visible=False),
|
431 |
}
|
432 |
+
subsets: list[str] = list(info_resp)
|
433 |
subset = default_subset if default_subset in subsets else subsets[0]
|
434 |
+
splits: list[str] = list(info_resp[subset]["splits"])
|
435 |
split = default_split if default_split in splits else splits[0]
|
436 |
+
features = info_resp[subset]["features"]
|
437 |
|
438 |
def _is_string_feature(feature):
|
439 |
return isinstance(feature, dict) and feature.get("dtype") == "string"
|
src/templates.py
CHANGED
@@ -5,12 +5,7 @@ You are a helpful, respectful and honest assistant for labeling topics.
|
|
5 |
"""
|
6 |
|
7 |
EXAMPLE_PROMPT = """
|
8 |
-
I have a topic that
|
9 |
-
- Traditional diets in most cultures were primarily plant-based with a little meat on top, but with the rise of industrial style meat production and factory farming, meat has become a staple food.
|
10 |
-
- Meat, but especially beef, is the word food in terms of emissions.
|
11 |
-
- Eating meat doesn't make you a bad person, not eating meat doesn't make you a good one.
|
12 |
-
|
13 |
-
The topic is described by the following keywords: 'meat, beef, eat, eating, emissions, steak, food, health, processed, chicken'.
|
14 |
|
15 |
Based on the information about the topic above, please create a short label of this topic. Make sure you to only return the label and nothing more.
|
16 |
|
@@ -19,10 +14,7 @@ Based on the information about the topic above, please create a short label of t
|
|
19 |
|
20 |
MAIN_PROMPT = """
|
21 |
[INST]
|
22 |
-
I have a topic that
|
23 |
-
[DOCUMENTS]
|
24 |
-
|
25 |
-
The topic is described by the following keywords: '[KEYWORDS]'.
|
26 |
|
27 |
Based on the information about the topic above, please create a short label of this topic. Make sure you to only return the label and nothing more.
|
28 |
[/INST]
|
@@ -32,7 +24,7 @@ REPRESENTATION_PROMPT = SYSTEM_PROMPT + EXAMPLE_PROMPT + MAIN_PROMPT
|
|
32 |
|
33 |
SPACE_REPO_CARD_CONTENT = """
|
34 |
---
|
35 |
-
title: {dataset_id}
|
36 |
sdk: static
|
37 |
pinned: false
|
38 |
datasets:
|
|
|
5 |
"""
|
6 |
|
7 |
EXAMPLE_PROMPT = """
|
8 |
+
I have a topic that is described by the following keywords: 'meat, beef, eat, eating, emissions, steak, food, health, processed, chicken'.
|
|
|
|
|
|
|
|
|
|
|
9 |
|
10 |
Based on the information about the topic above, please create a short label of this topic. Make sure you to only return the label and nothing more.
|
11 |
|
|
|
14 |
|
15 |
MAIN_PROMPT = """
|
16 |
[INST]
|
17 |
+
I have a topic that is described by the following keywords: '[KEYWORDS]'.
|
|
|
|
|
|
|
18 |
|
19 |
Based on the information about the topic above, please create a short label of this topic. Make sure you to only return the label and nothing more.
|
20 |
[/INST]
|
|
|
24 |
|
25 |
SPACE_REPO_CARD_CONTENT = """
|
26 |
---
|
27 |
+
title: {dataset_id}
|
28 |
sdk: static
|
29 |
pinned: false
|
30 |
datasets:
|
src/viewer_api.py
ADDED
@@ -0,0 +1,51 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import requests
|
2 |
+
import duckdb
|
3 |
+
|
4 |
+
DATASET_VIEWER_API_URL = "https://datasets-server.huggingface.co/"
|
5 |
+
session = requests.Session()
|
6 |
+
|
7 |
+
|
8 |
+
def fetch_json(url, params=None, timeout=20):
|
9 |
+
response = session.get(url, params=params, timeout=timeout)
|
10 |
+
response.raise_for_status()
|
11 |
+
data = response.json()
|
12 |
+
if "error" in data:
|
13 |
+
raise Exception(f"Error fetching data: {data['error']}")
|
14 |
+
return data
|
15 |
+
|
16 |
+
|
17 |
+
def get_split_rows(dataset, config, split):
|
18 |
+
url = f"{DATASET_VIEWER_API_URL}/size"
|
19 |
+
params = {"dataset": dataset, "config": config}
|
20 |
+
config_size = fetch_json(url, params)
|
21 |
+
|
22 |
+
split_size = next(
|
23 |
+
(s for s in config_size["size"]["splits"] if s["split"] == split), None
|
24 |
+
)
|
25 |
+
if split_size is None:
|
26 |
+
raise Exception(f"Error fetching split {split} in config {config}")
|
27 |
+
|
28 |
+
return split_size["num_rows"]
|
29 |
+
|
30 |
+
|
31 |
+
def get_parquet_urls(dataset, config, split):
|
32 |
+
url = f"{DATASET_VIEWER_API_URL}/parquet"
|
33 |
+
params = {"dataset": dataset, "config": config, "split": split}
|
34 |
+
parquet_files = fetch_json(url, params)
|
35 |
+
|
36 |
+
parquet_urls = [file["url"] for file in parquet_files["parquet_files"]]
|
37 |
+
return ",".join(f"'{url}'" for url in parquet_urls)
|
38 |
+
|
39 |
+
|
40 |
+
def get_docs_from_parquet(parquet_urls, column, offset, limit):
|
41 |
+
sql_query = f"SELECT {column} FROM read_parquet([{parquet_urls}]) LIMIT {limit} OFFSET {offset};"
|
42 |
+
df = duckdb.sql(sql_query).to_df()
|
43 |
+
return df[column].tolist()
|
44 |
+
|
45 |
+
|
46 |
+
def get_info(dataset):
|
47 |
+
url = f"{DATASET_VIEWER_API_URL}/info"
|
48 |
+
params = {"dataset": dataset}
|
49 |
+
info_resp = fetch_json(url, params)
|
50 |
+
|
51 |
+
return info_resp["dataset_info"]
|