switched to model uploadd in better format

app.py

@@ -2,30 +2,17 @@ import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM, LlamaForCausalLM, LlamaConfig
 import gradio as gr
 
-# base_model_id = "HuggingFaceTB/SmolLM2-135M"
-# instruct_model_id = "MaxBlumenfeld/smollm2-135m-bootleg-instruct"
-
-
-# base_tokenizer = AutoTokenizer.from_pretrained(base_model_id)
-
-# # Load models with explicit configs
-# base_config = LlamaConfig.from_pretrained(base_model_id)
-# instruct_config = LlamaConfig.from_pretrained(base_model_id)  # Using base model config for both since it's the same architecture
-
-# base_model = AutoModelForCausalLM.from_pretrained(base_model_id, config=base_config)
-# instruct_model = AutoModelForCausalLM.from_pretrained(instruct_model_id, from_tf=True)  # Added from_tf=True
-
 
 # Model IDs from Hugging Face Hub
 base_model_id = "HuggingFaceTB/SmolLM2-135M"
-instruct_model_id = "MaxBlumenfeld/smollm2-135m-bootleg-instruct"
+instruct_model_id = "MaxBlumenfeld/bootleg_instruct_01"
 
 # Load tokenizer
-tokenizer = AutoTokenizer.from_pretrained(base_model_id)
+base_tokenizer = AutoTokenizer.from_pretrained(base_model_id)
 
 # Load models with explicit LLaMA architecture
 base_model = LlamaForCausalLM.from_pretrained(base_model_id)
-instruct_model = LlamaForCausalLM.from_pretrained(instruct_model_id, from_tf = True)
+instruct_model = LlamaForCausalLM.from_pretrained(instruct_model_id)
 
 def generate_response(model, tokenizer, message, temperature=0.5, max_length=200, system_prompt="", is_instruct=False):
     # Prepare input based on model type