Update app.py

app.py

@@ -5,7 +5,7 @@ import subprocess
 import sys
 
 # Force install the specific transformers version from the GitHub PR
-subprocess.check_call([sys.executable, "-m", "pip", "install", "-U", "--force-reinstall", "accelerate", "git+https://github.com/Muennighoff/transformers.git@olmoe"])
+subprocess.check_call([sys.executable, "-m", "pip", "install", "-U", "--force-reinstall", "--no-deps", "accelerate", "git+https://github.com/Muennighoff/transformers.git@olmoe"])
 
 from transformers import OlmoeForCausalLM, AutoTokenizer
 
@@ -32,6 +32,13 @@ system_prompt = ("Adopt the persona of hilariously pissed off Andrej Karpathy "
                  "while always answering questions in full first principles analysis type of thinking "
                  "without using any analogies and always showing full working code or output in his answers.")
 
+# Define a chat template
+chat_template = {
+    "system": "<|system|>{content}<|end|>",
+    "user": "<|user|>{content}<|end|>",
+    "assistant": "<|assistant|>{content}<|end|>",
+}
+
 @spaces.GPU
 def generate_response(message, history, temperature, max_new_tokens):
     if model is None or tokenizer is None:
@@ -40,7 +47,7 @@ def generate_response(message, history, temperature, max_new_tokens):
     messages = [{"role": "system", "content": system_prompt},
                 {"role": "user", "content": message}]
     
-    inputs = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt").to(DEVICE)
+    inputs = tokenizer.apply_chat_template(messages, chat_template=chat_template, tokenize=True, add_generation_prompt=True, return_tensors="pt").to(DEVICE)
     
     with torch.no_grad():
         generate_ids = model.generate(
@@ -86,4 +93,4 @@ with gr.Blocks(css=css) as demo:
 
 if __name__ == "__main__":
     demo.queue(api_open=True)
-    demo.launch(debug=True, show_api=True, share=True)
+    demo.launch(debug=True, show_api=True, share=True)