#170 add llm options to analyse data

pyproject.toml

@@ -46,7 +46,9 @@ dependencies = [
     "flask",
     "waitress",
     "tqdm",
-    "haralyzer"
+    "haralyzer",
+    "torch",
+    "transformers"
 ]
 
 [project.scripts]

requirements.txt

@@ -6,4 +6,6 @@ waitress # Production WSGI server
 flask # WSGI server for development the code
 tqdm
 haralyzer
-pandas
+torch # LLM library
+transformers
+pandas

src/javacore_analyser/ai/add_stack_depth.py

@@ -0,0 +1,27 @@
+#
+# Copyright IBM Corp. 2025 - 2025
+# SPDX-License-Identifier: Apache-2.0
+#
+
+import pandas as pd
+import numpy as np
+
+
+def main():
+    df = pd.read_csv('data.csv', delimiter='\t')
+    print(df)
+    for index, row in df.iterrows():
+        if not np.isfinite(row['stack_trace_depth']):
+            stack_trace = row['stack_trace']
+            #print("Index: " + str(index) + " stack trace: " + stack_trace)
+            if isinstance(stack_trace, str):
+                depth = stack_trace.count('.java:') + stack_trace.count('Native Method')
+            else:
+                depth = 0
+            row['stack_trace_depth'] = depth
+            df.loc[index, 'stack_trace_depth'] = depth
+    print(df)
+    df.to_csv('data-modified.csv', sep='\t')
+
+if __name__ == '__main__':
+    main()

src/javacore_analyser/ai/llm_local_model.py

@@ -0,0 +1,29 @@
+#
+# Copyright IBM Corp. 2025 - 2025
+# SPDX-License-Identifier: Apache-2.0
+#
+
+import logging
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+
+class LLMLocalModel:
+
+    def __init__(self):
+        logging.info("Initialising LLM Local Model")
+        self.device = "cpu"
+        self.max_tokens = 50
+        self.model_path = "ibm-granite/granite-4.0-350M"
+        self.model = AutoModelForCausalLM.from_pretrained(self.model_path)
+        self.model.eval()
+        logging.info("LLM Local Model Initialised")
+
+
+    def generate_response(self, text):
+        chat = [{ "role": "user", "content": text },]
+        tokenizer = AutoTokenizer.from_pretrained(self.model_path)
+        chat = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
+        input_tokens = tokenizer(chat, return_tensors="pt").to(self.device)
+        output = self.model.generate(**input_tokens, max_new_tokens=self.max_tokens)
+        output = tokenizer.batch_decode(output)
+        return output[0]

src/javacore_analyser/javacore_set.py

@@ -20,6 +20,7 @@
 from tqdm import tqdm
 
 from javacore_analyser import tips
+import javacore_analyser
 from javacore_analyser.code_snapshot_collection import CodeSnapshotCollection
 from javacore_analyser.constants import *
 from javacore_analyser.har_file import HarFile
@@ -29,6 +30,7 @@
 from javacore_analyser.snapshot_collection import SnapshotCollection
 from javacore_analyser.snapshot_collection_collection import SnapshotCollectionCollection
 from javacore_analyser.verbose_gc import VerboseGcParser
+from javacore_analyser.ai.llm_local_model import LLMLocalModel
 
 
 def _create_xml_xsl_for_collection(tmp_dir, templates_dir, xml_xsl_filename, collection, output_file_prefix):
@@ -96,6 +98,9 @@ def __init__(self, path):
         self.gc_parser = VerboseGcParser()
         self.har_files = []
 
+        self.model = LLMLocalModel()
+
+
     # Assisted by WCA@IBM
     # Latest GenAI contribution: ibm/granite-8b-code-instruct
     @staticmethod
@@ -367,6 +372,13 @@ def __create_report_xml(self, output_file):
 
         doc_node = self.doc.documentElement
 
+        header_node = self.doc.createElement("header")
+        logging.info("Generating welcome message")
+        welcome_message = self.model.generate_response("Write one sentence tip for tunning Java application")
+        logging.info("Welcome message generated: " + welcome_message)
+        header_node.appendChild(self.doc.createTextNode(welcome_message))
+        doc_node.appendChild(header_node)
+
         javacore_count_node = self.doc.createElement("javacore_count")
         javacore_count_node.appendChild(self.doc.createTextNode(str(len(self.javacores))))
         doc_node.appendChild(javacore_count_node)