update

app.py

@@ -1,13 +1,33 @@
+import sys
+import warnings
+print("Warning: This application requires specific library versions. Please ensure you have the correct versions installed.")
+
 import gradio as gr
 from transformers import AutoTokenizer, AutoModelForCausalLM
 import torch
+import numpy as np
+
+print(f"NumPy version: {np.__version__}")
+print(f"PyTorch version: {torch.__version__}")
+
+# Suppress CUDA initialization warning
+warnings.filterwarnings("ignore", category=UserWarning, message="Can't initialize NVML")
+
+# Check for GPU availability
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+print(f"Using device: {device}")
 
 # Model loading and setup
 model_name = "jhu-clsp/FollowIR-7B"
-model = AutoModelForCausalLM.from_pretrained(model_name)
-if torch.cuda.is_available():
-    model = model.cuda()
-tokenizer = AutoTokenizer.from_pretrained(model_name, padding_side="left")
+
+try:
+    model = AutoModelForCausalLM.from_pretrained(model_name).to(device)
+    tokenizer = AutoTokenizer.from_pretrained(model_name, padding_side="left")
+except ValueError as e:
+    print(f"Error loading model or tokenizer: {e}")
+    print("Please ensure you have the correct versions of transformers and sentencepiece installed.")
+    sys.exit(1)
+
 tokenizer.pad_token = tokenizer.eos_token
 tokenizer.padding_side = "left"
 token_false_id = tokenizer.get_vocab()["false"]
@@ -31,11 +51,11 @@ def check_relevance(query, instruction, passage):
         pad_to_multiple_of=None,
     )
 
-    if torch.cuda.is_available():
-        for key in tokens:
-            tokens[key] = tokens[key].cuda()
+    for key in tokens:
+        tokens[key] = tokens[key].to(device)
 
-    batch_scores = model(**tokens).logits[:, -1, :]
+    with torch.no_grad():
+        batch_scores = model(**tokens).logits[:, -1, :]
     true_vector = batch_scores[:, token_true_id]
     false_vector = batch_scores[:, token_false_id]
     batch_scores = torch.stack([false_vector, true_vector], dim=1)
@@ -65,4 +85,8 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
         outputs=[output]
     )
 
-demo.launch()
+if __name__ == "__main__":
+    if np.__version__.startswith("2."):
+        print("Error: This application is not compatible with NumPy 2.x. Please downgrade to NumPy < 2.0.0.")
+        sys.exit(1)
+    demo.launch()

requirements.txt

@@ -1,5 +1,6 @@
 gradio
---extra-index-url https://download.pytorch.org/whl/cu117
-torch
-transformers
-numpy<2.0.0
+transformers==4.35.2
+numpy==1.24.3
+sentencepiece==0.1.99
+torch==2.0.1
+--extra-index-url https://download.pytorch.org/whl/cu117