Update space

app.py

@@ -59,61 +59,50 @@ def generate_and_tokenize_prompt(data_point):
     return tokenized_full_prompt
 
 def evaluate(instruction):
-    # Generate a response:
-    input = None
-    prompt = prompter.generate_prompt(instruction, input)
-    inputs = tokenizer(prompt, return_tensors="pt")
-    #inputs = inputs.to("cuda:0")
-    input_ids = inputs["input_ids"]
-    
-    #play around with generation strategies for better/diverse sequences. https://huggingface.co/docs/transformers/generation_strategies
-    temperature=0.2
-    top_p=0.95
-    top_k=25
-    num_beams=1
-    # num_beam_groups=num_beams #see: 'Diverse beam search decoding'
-    max_new_tokens=256
-    repetition_penalty = 2.0
-    do_sample = True # allow 'beam sample': do_sample=True, num_beams > 1
-    num_return_sequences = 1 #generate multiple candidates, takes longer..
-    
-    generation_config = transformers.GenerationConfig(
-        temperature=temperature,
-        top_p=top_p,
-        top_k=top_k,
-        num_beams=num_beams,
-        repetition_penalty=repetition_penalty,
-        do_sample=do_sample,
-        min_new_tokens=32,
-        num_return_sequences=num_return_sequences,
-        pad_token_id = 0
-        # num_beam_groups=num_beam_groups
-    )
-    
-    generate_params = {
-        "input_ids": input_ids,
-        "generation_config": generation_config,
-        "return_dict_in_generate": True,
-        "output_scores": True,
-        "max_new_tokens": max_new_tokens,
-    }
-    with torch.no_grad():
-        generation_output = model.generate(
-            input_ids=input_ids,
-            generation_config=generation_config,
-            return_dict_in_generate=True,
-            output_scores=True,
-            max_new_tokens=max_new_tokens,
+    try:
+        # Generate a response:
+        input_text = None
+        prompt = prompter.generate_prompt(instruction, input_text)
+        inputs = tokenizer(prompt, return_tensors="pt")
+        input_ids = inputs["input_ids"]
+        
+        temperature = 0.2
+        top_p = 0.95
+        top_k = 25
+        num_beams = 1
+        max_new_tokens = 256
+        repetition_penalty = 2.0
+        do_sample = True
+        num_return_sequences = 1
+        
+        generation_config = transformers.GenerationConfig(
+            temperature=temperature,
+            top_p=top_p,
+            top_k=top_k,
+            num_beams=num_beams,
+            repetition_penalty=repetition_penalty,
+            do_sample=do_sample,
+            min_new_tokens=32,
+            num_return_sequences=num_return_sequences,
+            pad_token_id=0
         )
-    
-    
-    print(f'Instruction: {instruction}')
-    
-    for i,s in enumerate(generation_output.sequences):
-      output = tokenizer.decode(s,skip_special_tokens=True)
-      # print(output)
-      return(f' {prompter.get_response(output)}')
-
+        
+        with torch.no_grad():
+            generation_output = model.generate(
+                input_ids=input_ids,
+                generation_config=generation_config,
+                return_dict_in_generate=True,
+                output_scores=True,
+                max_new_tokens=max_new_tokens,
+            )
+        
+        print(f'Instruction: {instruction}')
+        
+        for i, s in enumerate(generation_output.sequences):
+            output = tokenizer.decode(s, skip_special_tokens=True)
+            return prompter.get_response(output)
+    except Exception as e:
+        return str(e)
 
 # Define the Gradio interface
 interface = gr.Interface(