Update app.py

app.py

@@ -6,23 +6,12 @@ from transformers import pipeline, AutoTokenizer, AutoModelForSequenceClassifica
 from keybert import KeyBERT
 
 # ─── DeepSeek Model Client ────────────────────────────────────────────────────
-# Option 1: High-level helper pipeline for chat-style generation
-chat = pipeline(
-    "chat",
+# High-level helper pipeline for text-generation (flattened chat messages)
+pipe = pipeline(
+    "text-generation",
     model="deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
     trust_remote_code=True
 )
-
-# Option 2: Direct model & tokenizer instantiation (alternative) Direct model & tokenizer instantiation (alternative)
-# tokenizer_ds = AutoTokenizer.from_pretrained(
-#     "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
-#     trust_remote_code=True
-# )
-# model_ds = AutoModelForCausalLM.from_pretrained(
-#     "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
-#     trust_remote_code=True
-# )
-
 @st.cache_resource
 def load_sentiment_pipeline():
     model_name = "mayf/amazon_reviews_bert_ft"
@@ -118,15 +107,17 @@ Tasks:
 2. Recommendations: Three separate paragraphs with actionable suggestions (max 30 words each).
 """
 
-    # Use the high-level pipeline for generation
-    chat_input = [
-        {"role": "system", "content": "You are a product-feedback analyst."},
-        {"role": "user", "content": prompt}
-    ]
-    gen_output = pipe(chat_input)
+    # Use the high-level text-generation pipeline with flattened prompt
+    # Flatten roles into a single text prompt
+    flat_prompt = "
+".join(f"{msg['role'].upper()}: {msg['content']}" for msg in chat_input)
+    gen_output = pipe(flat_prompt, max_new_tokens=200)
     gpt_reply = gen_output[0]['generated_text']
 
     # Alternative: direct model invocation
+    # inputs = tokenizer_ds(flat_prompt, return_tensors="pt")
+    # outputs = model_ds.generate(**inputs, max_new_tokens=200)
+    # gpt_reply = tokenizer_ds.decode(outputs[0], skip_special_tokens=True)
     # inputs = tokenizer_ds(prompt, return_tensors="pt")
     # outputs = model_ds.generate(**inputs, max_new_tokens=200)
     # gpt_reply = tokenizer_ds.decode(outputs[0], skip_special_tokens=True)