Add detailed error reporting inside summarize endpoint

app.py

@@ -1,11 +1,12 @@
 import sys
+import traceback
 from fastapi import FastAPI, HTTPException
 from pydantic import BaseModel
 import torch
 import uvicorn
 from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer
 
-print("=== STARTING LEGALPARAM FASTAPI SERVICE ===")
+print("=== STARTING LEGALPARAM FASTAPI DOCKER SERVICE ===")
 print("CUDA Available:", torch.cuda.is_available())
 if torch.cuda.is_available():
     print("Device Name:", torch.cuda.get_device_name(0))
@@ -20,12 +21,8 @@ try:
     config = AutoConfig.from_pretrained(model_id, trust_remote_code=True)
     config.rope_scaling = {"type": "linear", "factor": 1.0}
 
-    print("Loading tokenizer with use_fast=False & trust_remote_code=False...")
-    tokenizer = AutoTokenizer.from_pretrained(
-        model_id,
-        use_fast=False,
-        trust_remote_code=False
-    )
+    print("Loading tokenizer with trust_remote_code=False...")
+    tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=False)
 
     print("Loading model onto GPU with bfloat16...")
     model = AutoModelForCausalLM.from_pretrained(
@@ -62,38 +59,43 @@ def summarize(req: SummaryRequest):
     if load_error or model is None:
         raise HTTPException(status_code=500, detail=f"Model not loaded: {load_error}")
 
-    prompt = (
-        "<user>\n"
-        "You are an expert Indian Insolvency Law editor. Synthesize and summarize the following multi-source raw case law summaries into a clean legal digest.\n"
-        "DO NOT include provider names, folder names, or lines like 'SOURCE 1'.\n\n"
-        "Format the output strictly as:\n"
-        "## Facts\n"
-        "[Brief facts]\n\n"
-        "## Issue & Holding\n"
-        "[Core question of law and the decision of the Tribunal/Court]\n\n"
-        f"RAW CASE LAW CONTENT:\n{req.text}\n"
-        "<assistant>\n"
-    )
-
-    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
-    available_output = max(256, 2048 - inputs.input_ids.shape[1] - 10)
-    max_gen_tokens = min(800, available_output)
-
-    with torch.no_grad():
-        outputs = model.generate(
-            **inputs,
-            max_new_tokens=max_gen_tokens,
-            temperature=0.2,
-            do_sample=True,
-            top_p=0.95,
-            eos_token_id=tokenizer.eos_token_id,
-            use_cache=False,
+    try:
+        prompt = (
+            "<user>\n"
+            "You are an expert Indian Insolvency Law editor. Synthesize and summarize the following multi-source raw case law summaries into a clean legal digest.\n"
+            "DO NOT include provider names, folder names, or lines like 'SOURCE 1'.\n\n"
+            "Format the output strictly as:\n"
+            "## Facts\n"
+            "[Brief facts]\n\n"
+            "## Issue & Holding\n"
+            "[Core question of law and the decision of the Tribunal/Court]\n\n"
+            f"RAW CASE LAW CONTENT:\n{req.text}\n"
+            "<assistant>\n"
         )
 
-    summary = tokenizer.decode(
-        outputs[0][inputs.input_ids.shape[1] :], skip_special_tokens=True
-    )
-    return {"summary": summary}
+        inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+        available_output = max(256, 2048 - inputs.input_ids.shape[1] - 10)
+        max_gen_tokens = min(800, available_output)
+
+        with torch.no_grad():
+            outputs = model.generate(
+                **inputs,
+                max_new_tokens=max_gen_tokens,
+                temperature=0.2,
+                do_sample=True,
+                top_p=0.95,
+                eos_token_id=tokenizer.eos_token_id,
+                use_cache=False,
+            )
+
+        summary = tokenizer.decode(
+            outputs[0][inputs.input_ids.shape[1] :], skip_special_tokens=True
+        )
+        return {"summary": summary}
+    except Exception as gen_err:
+        err_msg = f"Generation Exception: {gen_err}\n{traceback.format_exc()}"
+        print("❌ GENERATION ERROR:\n", err_msg, file=sys.stderr)
+        raise HTTPException(status_code=500, detail=err_msg)
 
 
 if __name__ == "__main__":