Upload app.py

app.py

@@ -8,6 +8,10 @@ from peft import PeftModel
 import whisperx
 import gc 
 import pandas as pd
+from transformers.utils import logging
+
+logging.set_verbosity_info()
+logger = logging.get_logger("transformers")
 
 device = 'cuda' if torch.cuda.is_available() else 'cpu'
 q_ctxt, query, ctxt_type = None, None, None
@@ -71,44 +75,48 @@ def add_file(history, file):
 
 
 def bot(history):
-    if len(history) > 0 and isinstance(history[-1][0], tuple):
-        ctxt_type = "audio"
-        q_ctxt = get_audio_context(history[-1][0][0])
-        response = "**Based on the given context: \n" + q_ctxt + "\n. **What do you want to know ?"
-        # input_text = context + query
-        # input_tokens = tokenizer.encode(input_text)
-        # input_ids = torch.tensor(input_tokens, dtype=torch.int32).unsqueeze(0).to(device)
-        # inputs_embeds = phi_model.get_input_embeddings()(input_ids)
-        # out = phi_model.generate(inputs_embeds=inputs_embeds, min_new_tokens=10, max_new_tokens=50, bos_token_id=tokenizer.bos_token_id)
-        # response = tokenizer.decode(out[0], skip_special_tokens=True)
-    elif len(history) > 0 and isinstance(history[-1][0],str) and len(history[-1][0].strip()) > 0:
-        print(history[-1][0])
-        if ctxt_type == "audio" and "[INST]" not in history[-1][0]:
-            query = history[-1][0]
-            input_text = q_ctxt + query
-        elif "[INST]" in history[-1][0]:
-            q_ctxt = history[-1][0].split("[INST]")[0]
-            print(q_ctxt)
-            query = history[-1][0].split("[INST]")[-1]
-            print(query)
-            input_text = history[-1][0].replace("[INST]"," ")
-        print(input_text)
-        input_tokens = tokenizer.encode(input_text)
-        input_ids = torch.tensor(input_tokens, dtype=torch.int32).unsqueeze(0).to(device)
-        inputs_embeds = phi_model.get_input_embeddings()(input_ids)
-        out = phi_model.generate(inputs_embeds=inputs_embeds, min_new_tokens=10, max_new_tokens=50, bos_token_id=tokenizer.bos_token_id)
-        response = tokenizer.decode(out[0], skip_special_tokens=True)
-        ctxt_type = "text"
-    else:
-        ctxt_type = None
-        q_ctxt = None
-        response = "Please ask your query or upload an audio/video to ask query"
-
-    history[-1][1] = ""
-    for character in response:
-        history[-1][1] += character
-        time.sleep(0.05)
-        yield history
+    try:
+        if len(history) > 0 and isinstance(history[-1][0], tuple):
+            ctxt_type = "audio"
+            q_ctxt = get_audio_context(history[-1][0][0])
+            response = "**Based on the given context: \n" + q_ctxt + "\n. **What do you want to know ?"
+            # input_text = context + query
+            # input_tokens = tokenizer.encode(input_text)
+            # input_ids = torch.tensor(input_tokens, dtype=torch.int32).unsqueeze(0).to(device)
+            # inputs_embeds = phi_model.get_input_embeddings()(input_ids)
+            # out = phi_model.generate(inputs_embeds=inputs_embeds, min_new_tokens=10, max_new_tokens=50, bos_token_id=tokenizer.bos_token_id)
+            # response = tokenizer.decode(out[0], skip_special_tokens=True)
+        elif len(history) > 0 and isinstance(history[-1][0],str) and len(history[-1][0].strip()) > 0:
+            logger.info(history[-1][0])
+            if ctxt_type == "audio" and "[INST]" not in history[-1][0]:
+                query = history[-1][0]
+                input_text = q_ctxt + query
+            elif "[INST]" in history[-1][0]:
+                q_ctxt = history[-1][0].split("[INST]")[0]
+                logger.info(q_ctxt)
+                query = history[-1][0].split("[INST]")[-1]
+                logger.info(query)
+                input_text = history[-1][0].replace("[INST]"," ")
+            logger.info(input_text)
+            input_tokens = tokenizer.encode(input_text)
+            input_ids = torch.tensor(input_tokens, dtype=torch.int32).unsqueeze(0).to(device)
+            inputs_embeds = phi_model.get_input_embeddings()(input_ids)
+            out = phi_model.generate(inputs_embeds=inputs_embeds, min_new_tokens=10, max_new_tokens=50, bos_token_id=tokenizer.bos_token_id)
+            response = tokenizer.decode(out[0], skip_special_tokens=True)
+            ctxt_type = "text"
+        else:
+            ctxt_type = None
+            q_ctxt = None
+            response = "Please ask your query or upload an audio/video to ask query"
+
+        history[-1][1] = ""
+        for character in response:
+            history[-1][1] += character
+            time.sleep(0.05)
+            yield history
+    except Exception as e:
+        logger.error(e)
+        return e
 
 
 with gr.Blocks() as demo:
@@ -144,4 +152,4 @@ with gr.Blocks() as demo:
 
 
 demo.queue()
-demo.launch(share=True)
+demo.launch(debug=True)