Update app.py
Browse files
app.py
CHANGED
@@ -1,14 +1,16 @@
|
|
1 |
-
import sys, os, json, shutil, re, time, gc
|
2 |
import pandas as pd
|
3 |
from datetime import datetime
|
4 |
from typing import List, Tuple, Dict, Union
|
5 |
import gradio as gr
|
|
|
6 |
|
7 |
# Constants
|
8 |
MAX_MODEL_TOKENS = 131072
|
9 |
MAX_NEW_TOKENS = 4096
|
10 |
-
MAX_CHUNK_TOKENS = 8192
|
11 |
PROMPT_OVERHEAD = 300
|
|
|
12 |
|
13 |
# Paths
|
14 |
persistent_dir = "/data/hf_cache"
|
@@ -41,20 +43,17 @@ def clean_response(text: str) -> str:
|
|
41 |
def extract_text_from_excel(path: str) -> str:
|
42 |
all_text = []
|
43 |
xls = pd.ExcelFile(path)
|
44 |
-
|
45 |
for sheet_name in xls.sheet_names:
|
46 |
try:
|
47 |
df = xls.parse(sheet_name).astype(str).fillna("")
|
48 |
except Exception:
|
49 |
continue
|
50 |
-
|
51 |
for idx, row in df.iterrows():
|
52 |
-
non_empty = [cell.strip() for cell in row if cell.strip()
|
53 |
if len(non_empty) >= 2:
|
54 |
text_line = " | ".join(non_empty)
|
55 |
if len(text_line) > 15:
|
56 |
all_text.append(f"[{sheet_name}] {text_line}")
|
57 |
-
|
58 |
return "\n".join(all_text)
|
59 |
|
60 |
def split_text(text: str, max_tokens=MAX_CHUNK_TOKENS) -> List[str]:
|
@@ -73,6 +72,9 @@ def split_text(text: str, max_tokens=MAX_CHUNK_TOKENS) -> List[str]:
|
|
73 |
chunks.append("\n".join(current))
|
74 |
return chunks
|
75 |
|
|
|
|
|
|
|
76 |
def build_prompt(chunk: str) -> str:
|
77 |
return f"""### Unstructured Clinical Records\n\nAnalyze the clinical notes below and summarize with:\n- Diagnostic Patterns\n- Medication Issues\n- Missed Opportunities\n- Inconsistencies\n- Follow-up Recommendations\n\n---\n\n{chunk}\n\n---\nRespond concisely in bullet points with clinical reasoning."""
|
78 |
|
@@ -92,19 +94,16 @@ def init_agent() -> TxAgent:
|
|
92 |
agent.init_model()
|
93 |
return agent
|
94 |
|
95 |
-
def
|
96 |
results = []
|
97 |
-
for
|
98 |
-
prompt = build_prompt(chunk)
|
99 |
-
if estimate_tokens(prompt) > MAX_MODEL_TOKENS:
|
100 |
-
results.append(f"β Chunk {idx+1} too long. Skipped.")
|
101 |
-
continue
|
102 |
response = ""
|
103 |
try:
|
104 |
for r in agent.run_gradio_chat(
|
105 |
message=prompt,
|
106 |
history=[],
|
107 |
-
temperature=0.
|
108 |
max_new_tokens=MAX_NEW_TOKENS,
|
109 |
max_token=MAX_MODEL_TOKENS,
|
110 |
call_agent=False,
|
@@ -120,7 +119,8 @@ def analyze_serial(agent, chunks: List[str]) -> List[str]:
|
|
120 |
response += r.content
|
121 |
results.append(clean_response(response))
|
122 |
except Exception as e:
|
123 |
-
results.append(f"β Error in
|
|
|
124 |
gc.collect()
|
125 |
return results
|
126 |
|
@@ -130,7 +130,7 @@ def generate_final_summary(agent, combined: str) -> str:
|
|
130 |
for r in agent.run_gradio_chat(
|
131 |
message=final_prompt,
|
132 |
history=[],
|
133 |
-
temperature=0.
|
134 |
max_new_tokens=MAX_NEW_TOKENS,
|
135 |
max_token=MAX_MODEL_TOKENS,
|
136 |
call_agent=False,
|
@@ -155,13 +155,14 @@ def process_report(agent, file, messages: List[Dict[str, str]]) -> Tuple[List[Di
|
|
155 |
try:
|
156 |
extracted = extract_text_from_excel(file.name)
|
157 |
chunks = split_text(extracted)
|
158 |
-
|
|
|
159 |
|
160 |
-
|
161 |
-
valid = [res for res in
|
162 |
|
163 |
if not valid:
|
164 |
-
messages.append({"role": "assistant", "content": "β No valid
|
165 |
return messages, None
|
166 |
|
167 |
summary = generate_final_summary(agent, "\n\n".join(valid))
|
|
|
1 |
+
import sys, os, json, shutil, re, time, gc
|
2 |
import pandas as pd
|
3 |
from datetime import datetime
|
4 |
from typing import List, Tuple, Dict, Union
|
5 |
import gradio as gr
|
6 |
+
from concurrent.futures import ThreadPoolExecutor
|
7 |
|
8 |
# Constants
|
9 |
MAX_MODEL_TOKENS = 131072
|
10 |
MAX_NEW_TOKENS = 4096
|
11 |
+
MAX_CHUNK_TOKENS = 8192
|
12 |
PROMPT_OVERHEAD = 300
|
13 |
+
BATCH_SIZE = 2 # NEW: batch 2 prompts together for faster processing
|
14 |
|
15 |
# Paths
|
16 |
persistent_dir = "/data/hf_cache"
|
|
|
43 |
def extract_text_from_excel(path: str) -> str:
|
44 |
all_text = []
|
45 |
xls = pd.ExcelFile(path)
|
|
|
46 |
for sheet_name in xls.sheet_names:
|
47 |
try:
|
48 |
df = xls.parse(sheet_name).astype(str).fillna("")
|
49 |
except Exception:
|
50 |
continue
|
|
|
51 |
for idx, row in df.iterrows():
|
52 |
+
non_empty = [cell.strip() for cell in row if cell.strip()]
|
53 |
if len(non_empty) >= 2:
|
54 |
text_line = " | ".join(non_empty)
|
55 |
if len(text_line) > 15:
|
56 |
all_text.append(f"[{sheet_name}] {text_line}")
|
|
|
57 |
return "\n".join(all_text)
|
58 |
|
59 |
def split_text(text: str, max_tokens=MAX_CHUNK_TOKENS) -> List[str]:
|
|
|
72 |
chunks.append("\n".join(current))
|
73 |
return chunks
|
74 |
|
75 |
+
def batch_chunks(chunks: List[str], batch_size: int = 2) -> List[List[str]]:
|
76 |
+
return [chunks[i:i+batch_size] for i in range(0, len(chunks), batch_size)]
|
77 |
+
|
78 |
def build_prompt(chunk: str) -> str:
|
79 |
return f"""### Unstructured Clinical Records\n\nAnalyze the clinical notes below and summarize with:\n- Diagnostic Patterns\n- Medication Issues\n- Missed Opportunities\n- Inconsistencies\n- Follow-up Recommendations\n\n---\n\n{chunk}\n\n---\nRespond concisely in bullet points with clinical reasoning."""
|
80 |
|
|
|
94 |
agent.init_model()
|
95 |
return agent
|
96 |
|
97 |
+
def analyze_batches(agent, batches: List[List[str]]) -> List[str]:
|
98 |
results = []
|
99 |
+
for batch in batches:
|
100 |
+
prompt = "\n\n".join(build_prompt(chunk) for chunk in batch)
|
|
|
|
|
|
|
101 |
response = ""
|
102 |
try:
|
103 |
for r in agent.run_gradio_chat(
|
104 |
message=prompt,
|
105 |
history=[],
|
106 |
+
temperature=0.0,
|
107 |
max_new_tokens=MAX_NEW_TOKENS,
|
108 |
max_token=MAX_MODEL_TOKENS,
|
109 |
call_agent=False,
|
|
|
119 |
response += r.content
|
120 |
results.append(clean_response(response))
|
121 |
except Exception as e:
|
122 |
+
results.append(f"β Error in batch: {str(e)}")
|
123 |
+
torch.cuda.empty_cache()
|
124 |
gc.collect()
|
125 |
return results
|
126 |
|
|
|
130 |
for r in agent.run_gradio_chat(
|
131 |
message=final_prompt,
|
132 |
history=[],
|
133 |
+
temperature=0.0,
|
134 |
max_new_tokens=MAX_NEW_TOKENS,
|
135 |
max_token=MAX_MODEL_TOKENS,
|
136 |
call_agent=False,
|
|
|
155 |
try:
|
156 |
extracted = extract_text_from_excel(file.name)
|
157 |
chunks = split_text(extracted)
|
158 |
+
batches = batch_chunks(chunks, batch_size=BATCH_SIZE)
|
159 |
+
messages.append({"role": "assistant", "content": f"π Split into {len(batches)} batches. Analyzing..."})
|
160 |
|
161 |
+
batch_results = analyze_batches(agent, batches)
|
162 |
+
valid = [res for res in batch_results if not res.startswith("β")]
|
163 |
|
164 |
if not valid:
|
165 |
+
messages.append({"role": "assistant", "content": "β No valid batch outputs."})
|
166 |
return messages, None
|
167 |
|
168 |
summary = generate_final_summary(agent, "\n\n".join(valid))
|