Upload app.py

app.py

@@ -3,7 +3,7 @@ import os
 import gradio as gr
 import torch
 from PIL import Image
-from transformers import AutoTokenizer
+from transformers import AutoTokenizer, AutoModelForCausalLM
 import timm
 from torchvision import transforms
 from llama_cpp import Llama
@@ -30,7 +30,7 @@ class SigLIPImageEncoder(torch.nn.Module):
 
 # 2. Load Models and Tokenizer
 phi3_model_path = "QuantFactory/Phi-3-mini-4k-instruct-GGUF"  # Path to your quantized Phi-3 GGUF model
-peft_model_path = "./qlora_phi3_model"
+peft_model_path = "./qlora-phi3-model"
 image_model_name = 'resnet50'
 image_embed_dim = 512
 siglip_pretrained_path = "image_encoder.pth" # Path to your pretrained SigLIP model
@@ -62,13 +62,18 @@ image_encoder.eval() # Set to evaluation mode
 #)
 
 
-base_model = Llama.from_pretrained(
-	repo_id="QuantFactory/Phi-3-mini-4k-instruct-GGUF",
-	filename="Phi-3-mini-4k-instruct.Q2_K.gguf",
-    n_gpu_layers=0,
-    n_ctx=2048,
-    verbose=True 
-)
+#base_model = Llama.from_pretrained(
+#	repo_id="QuantFactory/Phi-3-mini-4k-instruct-GGUF",
+#	filename="Phi-3-mini-4k-instruct.Q2_K.gguf",
+#    n_gpu_layers=0,
+#    n_ctx=2048,
+#    verbose=True 
+#)
+
+base_model_name="microsoft/Phi-3-mini-4k-instruct"
+device = "cpu"
+
+base_model = AutoModelForCausalLM.from_pretrained(base_model_name, torch_dtype=torch.float32, device_map={"": device})
 
 
 # Load and merge