final update of the logic
Browse files- __pycache__/inference_svm_model.cpython-310.pyc +0 -0
- __pycache__/mineru_single.cpython-310.pyc +0 -0
- __pycache__/worker.cpython-310.pyc +0 -0
- app.py +0 -30
- inference_svm_model.py +20 -18
- mineru_single.py +8 -33
- worker.py +28 -7
__pycache__/inference_svm_model.cpython-310.pyc
CHANGED
Binary files a/__pycache__/inference_svm_model.cpython-310.pyc and b/__pycache__/inference_svm_model.cpython-310.pyc differ
|
|
__pycache__/mineru_single.cpython-310.pyc
CHANGED
Binary files a/__pycache__/mineru_single.cpython-310.pyc and b/__pycache__/mineru_single.cpython-310.pyc differ
|
|
__pycache__/worker.cpython-310.pyc
CHANGED
Binary files a/__pycache__/worker.cpython-310.pyc and b/__pycache__/worker.cpython-310.pyc differ
|
|
app.py
CHANGED
@@ -24,36 +24,6 @@ app.add_middleware(
|
|
24 |
async def root():
|
25 |
return {"status": "ok", "message": "API is running"}
|
26 |
|
27 |
-
@app.post("/process")
|
28 |
-
async def process_pdf(
|
29 |
-
input_json: dict = Body(...),
|
30 |
-
x_api_key: str = Header(None, alias="X-API-Key")
|
31 |
-
):
|
32 |
-
if not x_api_key:
|
33 |
-
raise HTTPException(status_code=401, detail="API key is missing")
|
34 |
-
if x_api_key != API_KEY:
|
35 |
-
raise HTTPException(status_code=401, detail="Invalid API key")
|
36 |
-
|
37 |
-
# Connect to RabbitMQ
|
38 |
-
rabbit_url = os.getenv("RABBITMQ_URL")
|
39 |
-
connection = pika.BlockingConnection(pika.URLParameters(rabbit_url))
|
40 |
-
channel = connection.channel()
|
41 |
-
channel.queue_declare(queue="ml_server", durable=True)
|
42 |
-
|
43 |
-
channel.basic_publish(
|
44 |
-
exchange="",
|
45 |
-
routing_key="gpu_server",
|
46 |
-
body=json.dumps(input_json),
|
47 |
-
properties=pika.BasicProperties(
|
48 |
-
headers={"process": "topic_extraction"}
|
49 |
-
)
|
50 |
-
)
|
51 |
-
connection.close()
|
52 |
-
|
53 |
-
return {
|
54 |
-
"message": "Job queued",
|
55 |
-
"request_id": input_json.get("headers", {}).get("request_id", str(uuid.uuid4()))
|
56 |
-
}
|
57 |
|
58 |
if __name__ == "__main__":
|
59 |
os.system('python download_models_hf.py')
|
|
|
24 |
async def root():
|
25 |
return {"status": "ok", "message": "API is running"}
|
26 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
|
28 |
if __name__ == "__main__":
|
29 |
os.system('python download_models_hf.py')
|
inference_svm_model.py
CHANGED
@@ -1,29 +1,31 @@
|
|
1 |
#!/usr/bin/env python3
|
2 |
import cv2
|
3 |
import numpy as np
|
|
|
4 |
from joblib import load
|
5 |
|
6 |
-
def load_svm_model(model_path: str):
|
7 |
-
return load(model_path)
|
8 |
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
image_size=(128, 128)
|
14 |
-
) -> str:
|
15 |
-
img = cv2.imread(image_path)
|
16 |
-
if img is None:
|
17 |
-
# If image fails to load, default to "irrelevant" or handle differently
|
18 |
-
return label_map[0]
|
19 |
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
|
25 |
if __name__ == "__main__":
|
26 |
model = load_svm_model("/home/user/app/model_classification/svm_model.joblib")
|
27 |
-
|
28 |
-
result = classify_image("test.jpg", model, label_map)
|
29 |
print("Classification result:", result)
|
|
|
1 |
#!/usr/bin/env python3
|
2 |
import cv2
|
3 |
import numpy as np
|
4 |
+
import os
|
5 |
from joblib import load
|
6 |
|
|
|
|
|
7 |
|
8 |
+
class SVMModel:
|
9 |
+
def __init__(self):
|
10 |
+
path = os.getenv("SVM_MODEL_PATH", "/home/user/app/model_classification/svm_model.joblib")
|
11 |
+
self.model = load(path)
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
|
13 |
+
def classify_image(
|
14 |
+
self,
|
15 |
+
image_bytes: bytes,
|
16 |
+
image_size=(128, 128)
|
17 |
+
) -> int:
|
18 |
+
img = cv2.imdecode(np.frombuffer(image_bytes, np.uint8), cv2.IMREAD_COLOR)
|
19 |
+
if img is None:
|
20 |
+
# If image fails to load, default to "irrelevant" or handle differently
|
21 |
+
return 0
|
22 |
+
|
23 |
+
img = cv2.resize(img, image_size)
|
24 |
+
x = img.flatten().reshape(1, -1)
|
25 |
+
pred = self.model.predict(x)[0]
|
26 |
+
return pred
|
27 |
|
28 |
if __name__ == "__main__":
|
29 |
model = load_svm_model("/home/user/app/model_classification/svm_model.joblib")
|
30 |
+
result = classify_image("test.jpg", model)
|
|
|
31 |
print("Classification result:", result)
|
mineru_single.py
CHANGED
@@ -10,7 +10,7 @@ from magic_pdf.model.doc_analyze_by_custom_model import doc_analyze
|
|
10 |
from magic_pdf.data.io.s3 import S3Writer
|
11 |
from magic_pdf.data.data_reader_writer.base import DataWriter
|
12 |
|
13 |
-
from inference_svm_model import
|
14 |
|
15 |
class Processor:
|
16 |
def __init__(self):
|
@@ -21,9 +21,7 @@ class Processor:
|
|
21 |
endpoint_url=os.getenv("S3_ENDPOINT"),
|
22 |
)
|
23 |
|
24 |
-
|
25 |
-
self.svm_model = load_svm_model(model_path)
|
26 |
-
self.label_map = {0: "irrelevant", 1: "relevant"}
|
27 |
|
28 |
with open("/home/user/magic-pdf.json", "r") as f:
|
29 |
config = json.load(f)
|
@@ -37,7 +35,7 @@ class Processor:
|
|
37 |
bucket = os.getenv("S3_BUCKET_NAME", "")
|
38 |
self.prefix = f"{endpoint}/{bucket}/document-extracts/"
|
39 |
|
40 |
-
def process(self, file_url: str) -> str:
|
41 |
logger.info("Processing file: {}", file_url)
|
42 |
response = requests.get(file_url)
|
43 |
if response.status_code != 200:
|
@@ -54,53 +52,30 @@ class Processor:
|
|
54 |
table_enable=self.table_enable
|
55 |
)
|
56 |
|
57 |
-
image_writer = ImageWriter(self.s3_writer, self.svm_model
|
58 |
|
59 |
pipe_result = inference.pipe_ocr_mode(image_writer, lang=self.language)
|
60 |
|
61 |
-
|
62 |
-
md_content = pipe_result.get_markdown(self.prefix + folder_name + "/")
|
63 |
|
64 |
# Remove references to images classified as "irrelevant"
|
65 |
final_markdown = image_writer.remove_redundant_images(md_content)
|
66 |
return final_markdown
|
67 |
|
68 |
-
def process_batch(self, file_urls: list[str]) -> dict:
|
69 |
-
results = {}
|
70 |
-
for url in file_urls:
|
71 |
-
try:
|
72 |
-
md = self.process(url)
|
73 |
-
results[url] = md
|
74 |
-
except Exception as e:
|
75 |
-
results[url] = f"Error: {str(e)}"
|
76 |
-
return results
|
77 |
-
|
78 |
class ImageWriter(DataWriter):
|
79 |
"""
|
80 |
Receives each extracted image. Classifies it, uploads if relevant, or flags
|
81 |
it for removal if irrelevant.
|
82 |
"""
|
83 |
-
def __init__(self, s3_writer: S3Writer, svm_model
|
84 |
self.s3_writer = s3_writer
|
85 |
self.svm_model = svm_model
|
86 |
-
self.label_map = label_map
|
87 |
self._redundant_images_paths = []
|
88 |
|
89 |
def write(self, path: str, data: bytes) -> None:
|
90 |
-
|
91 |
-
import os
|
92 |
-
import uuid
|
93 |
-
|
94 |
-
tmp_name = f"{uuid.uuid4()}.jpg"
|
95 |
-
tmp_path = os.path.join(tempfile.gettempdir(), tmp_name)
|
96 |
-
with open(tmp_path, "wb") as f:
|
97 |
-
f.write(data)
|
98 |
-
|
99 |
-
label_str = classify_image(tmp_path, self.svm_model, self.label_map)
|
100 |
-
|
101 |
-
os.remove(tmp_path)
|
102 |
|
103 |
-
if label_str ==
|
104 |
# Upload to S3
|
105 |
self.s3_writer.write(path, data)
|
106 |
else:
|
|
|
10 |
from magic_pdf.data.io.s3 import S3Writer
|
11 |
from magic_pdf.data.data_reader_writer.base import DataWriter
|
12 |
|
13 |
+
from inference_svm_model import SVMModel
|
14 |
|
15 |
class Processor:
|
16 |
def __init__(self):
|
|
|
21 |
endpoint_url=os.getenv("S3_ENDPOINT"),
|
22 |
)
|
23 |
|
24 |
+
self.svm_model = SVMModel()
|
|
|
|
|
25 |
|
26 |
with open("/home/user/magic-pdf.json", "r") as f:
|
27 |
config = json.load(f)
|
|
|
35 |
bucket = os.getenv("S3_BUCKET_NAME", "")
|
36 |
self.prefix = f"{endpoint}/{bucket}/document-extracts/"
|
37 |
|
38 |
+
def process(self, file_url: str, key: str) -> str:
|
39 |
logger.info("Processing file: {}", file_url)
|
40 |
response = requests.get(file_url)
|
41 |
if response.status_code != 200:
|
|
|
52 |
table_enable=self.table_enable
|
53 |
)
|
54 |
|
55 |
+
image_writer = ImageWriter(self.s3_writer, self.svm_model)
|
56 |
|
57 |
pipe_result = inference.pipe_ocr_mode(image_writer, lang=self.language)
|
58 |
|
59 |
+
md_content = pipe_result.get_markdown(self.prefix + key + "/")
|
|
|
60 |
|
61 |
# Remove references to images classified as "irrelevant"
|
62 |
final_markdown = image_writer.remove_redundant_images(md_content)
|
63 |
return final_markdown
|
64 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
65 |
class ImageWriter(DataWriter):
|
66 |
"""
|
67 |
Receives each extracted image. Classifies it, uploads if relevant, or flags
|
68 |
it for removal if irrelevant.
|
69 |
"""
|
70 |
+
def __init__(self, s3_writer: S3Writer, svm_model: SVMModel):
|
71 |
self.s3_writer = s3_writer
|
72 |
self.svm_model = svm_model
|
|
|
73 |
self._redundant_images_paths = []
|
74 |
|
75 |
def write(self, path: str, data: bytes) -> None:
|
76 |
+
label_str = self.svm_model.classify_image(data)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
77 |
|
78 |
+
if label_str == 1:
|
79 |
# Upload to S3
|
80 |
self.s3_writer.write(path, data)
|
81 |
else:
|
worker.py
CHANGED
@@ -14,13 +14,17 @@ from mineru_single import Processor
|
|
14 |
class RabbitMQWorker:
|
15 |
def __init__(self, num_workers: int = 1):
|
16 |
self.num_workers = num_workers
|
17 |
-
self.rabbit_url = os.getenv("RABBITMQ_URL"
|
18 |
self.processor = Processor()
|
19 |
|
20 |
def publish_message(self, body_dict: dict, headers: dict):
|
21 |
"""Create a new connection for each publish operation"""
|
22 |
try:
|
23 |
-
|
|
|
|
|
|
|
|
|
24 |
channel = connection.channel()
|
25 |
|
26 |
channel.queue_declare(queue="ml_server", durable=True)
|
@@ -56,41 +60,58 @@ class RabbitMQWorker:
|
|
56 |
# Process files
|
57 |
for file in body_dict.get("input_files", []):
|
58 |
try:
|
59 |
-
context = {"key": file["key"], "body": self.processor.process(file["url"])}
|
60 |
contexts.append(context)
|
61 |
except Exception as e:
|
62 |
print(f"Error processing file {file['key']}: {e}")
|
63 |
contexts.append({"key": file["key"], "body": f"Error: {str(e)}"})
|
64 |
|
65 |
body_dict["md_context"] = contexts
|
|
|
66 |
|
67 |
# Publish results
|
68 |
if self.publish_message(body_dict, headers):
|
69 |
print(f"[Worker {thread_id}] Successfully published results")
|
70 |
else:
|
|
|
71 |
print(f"[Worker {thread_id}] Failed to publish results")
|
72 |
|
73 |
print(f"[Worker {thread_id}] Contexts: {contexts}")
|
74 |
else:
|
|
|
75 |
print(f"[Worker {thread_id}] Unknown process")
|
76 |
|
77 |
except Exception as e:
|
78 |
print(f"Error in callback: {e}")
|
|
|
79 |
|
80 |
def connect_to_rabbitmq(self):
|
81 |
-
"""Establish connection to RabbitMQ"""
|
82 |
-
|
|
|
|
|
|
|
|
|
83 |
channel = connection.channel()
|
84 |
|
85 |
channel.queue_declare(queue="gpu_server", durable=True)
|
86 |
channel.basic_qos(prefetch_count=1)
|
87 |
channel.basic_consume(
|
88 |
queue="gpu_server",
|
89 |
-
on_message_callback=self.callback
|
90 |
-
auto_ack=True
|
91 |
)
|
92 |
return connection, channel
|
93 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
94 |
def start(self):
|
95 |
"""Start the worker threads"""
|
96 |
print(f"Starting {self.num_workers} workers")
|
|
|
14 |
class RabbitMQWorker:
|
15 |
def __init__(self, num_workers: int = 1):
|
16 |
self.num_workers = num_workers
|
17 |
+
self.rabbit_url = os.getenv("RABBITMQ_URL")
|
18 |
self.processor = Processor()
|
19 |
|
20 |
def publish_message(self, body_dict: dict, headers: dict):
|
21 |
"""Create a new connection for each publish operation"""
|
22 |
try:
|
23 |
+
connection_params = pika.URLParameters(self.rabbit_url)
|
24 |
+
connection_params.heartbeat = 10
|
25 |
+
connection_params.blocked_connection_timeout = 5
|
26 |
+
|
27 |
+
connection = pika.BlockingConnection(connection_params)
|
28 |
channel = connection.channel()
|
29 |
|
30 |
channel.queue_declare(queue="ml_server", durable=True)
|
|
|
60 |
# Process files
|
61 |
for file in body_dict.get("input_files", []):
|
62 |
try:
|
63 |
+
context = {"key": file["key"], "body": self.processor.process(file["url"], file["key"])}
|
64 |
contexts.append(context)
|
65 |
except Exception as e:
|
66 |
print(f"Error processing file {file['key']}: {e}")
|
67 |
contexts.append({"key": file["key"], "body": f"Error: {str(e)}"})
|
68 |
|
69 |
body_dict["md_context"] = contexts
|
70 |
+
ch.basic_ack(delivery_tag=method.delivery_tag)
|
71 |
|
72 |
# Publish results
|
73 |
if self.publish_message(body_dict, headers):
|
74 |
print(f"[Worker {thread_id}] Successfully published results")
|
75 |
else:
|
76 |
+
ch.basic_nack(delivery_tag=method.delivery_tag, requeue=True)
|
77 |
print(f"[Worker {thread_id}] Failed to publish results")
|
78 |
|
79 |
print(f"[Worker {thread_id}] Contexts: {contexts}")
|
80 |
else:
|
81 |
+
ch.basic_nack(delivery_tag=method.delivery_tag, requeue=True)
|
82 |
print(f"[Worker {thread_id}] Unknown process")
|
83 |
|
84 |
except Exception as e:
|
85 |
print(f"Error in callback: {e}")
|
86 |
+
ch.basic_nack(delivery_tag=method.delivery_tag, requeue=True)
|
87 |
|
88 |
def connect_to_rabbitmq(self):
|
89 |
+
"""Establish connection to RabbitMQ with heartbeat"""
|
90 |
+
connection_params = pika.URLParameters(self.rabbit_url)
|
91 |
+
connection_params.heartbeat = 30
|
92 |
+
connection_params.blocked_connection_timeout = 10
|
93 |
+
|
94 |
+
connection = pika.BlockingConnection(connection_params)
|
95 |
channel = connection.channel()
|
96 |
|
97 |
channel.queue_declare(queue="gpu_server", durable=True)
|
98 |
channel.basic_qos(prefetch_count=1)
|
99 |
channel.basic_consume(
|
100 |
queue="gpu_server",
|
101 |
+
on_message_callback=self.callback
|
|
|
102 |
)
|
103 |
return connection, channel
|
104 |
|
105 |
+
def worker(self, channel):
|
106 |
+
"""Worker function"""
|
107 |
+
print(f"Worker started")
|
108 |
+
try:
|
109 |
+
channel.start_consuming()
|
110 |
+
except Exception as e:
|
111 |
+
print(f"Worker stopped: {e}")
|
112 |
+
finally:
|
113 |
+
channel.close()
|
114 |
+
|
115 |
def start(self):
|
116 |
"""Start the worker threads"""
|
117 |
print(f"Starting {self.num_workers} workers")
|