Update app.py

app.py

@@ -18,12 +18,11 @@ secret_key = os.getenv("HUGGINGFACE_ACCESS_TOKEN")
 
 from torch import nn
 
-from peft import PeftModel, PeftConfig
-from transformers import AutoModelForCausalLM, AutoTokenizer
+# Load model directly
+from transformers import AutoTokenizer, AutoModelForCausalLM
 
-config = PeftConfig.from_pretrained("tmberooney/medllama")
-model = AutoModelForCausalLM.from_pretrained("daryl149/llama-2-7b-chat-hf", use_auth_token=secret_key, llm_int8_enable_fp32_cpu_offload=True, torch_dtype=torch.float16)
-model = PeftModel.from_pretrained(model, "tmberooney/medllama", use_auth_token=secret_key)
+tokenizer = AutoTokenizer.from_pretrained("tmberooney/medllama-merged")
+model = AutoModelForCausalLM.from_pretrained("tmberooney/medllama-merged")
 device_map = {"transformer.word_embeddings": "cpu",
               "transformer.word_embeddings_layernorm": "cpu",
               "lm_head": "cpu",
@@ -54,7 +53,7 @@ llama_pipeline = pipeline(
 )
 
 SYSTEM_PROMPT = """<s>[INST] <<SYS>>
-You are a helpful medical bot. Your answers are clear and concise with medical information.
+ou are a Doctor, Who is trained to listen to patients claims about what they have been feeling. You are to notice their sypmtoms, the medications they have been taking and their medical history. Based on this You have made a appropriate Diagnosis.
 <</SYS>>
 
 """