Update README.md

README.md

@@ -42,42 +42,51 @@ Because this model relies on the custom Moondream2 architecture, you will need t
 ### Prerequisites
 Make sure you have the required libraries installed:
 ```bash
-pip install transformers pillow einops
+!pip install transformers==4.44.2 "huggingface_hub<1.0" accelerate pillow einops
 ```
-### Python Inference Script
+
+### Load Tokenizer & Model
 ```python
 import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
 from PIL import Image
+import requests
 
-# 1. Define the model ID
 model_id = "Subh775/Perception-moondream2"
 
-# 2. Load the tokenizer and model
-# Note: trust_remote_code=True is required for the moondream2 architecture
 tokenizer = AutoTokenizer.from_pretrained(model_id)
+
 model = AutoModelForCausalLM.from_pretrained(
-    model_id, 
+    model_id,
     trust_remote_code=True,
-    torch_dtype=torch.float16, # Recommended for memory efficiency
-    device_map="auto"
+    torch_dtype=torch.float16,
+    # REMOVED device_map="auto"
 )
+# move to the GPU
+model = model.to("cuda")
+model.eval()
+```
 
-# 3. Load your traffic/CCTV image
-image_path = "path_to_your_traffic_image.jpg"
+# Inference
+```python
+image_path = "/content/100130.jpg"
 image = Image.open(image_path).convert("RGB")
 
-# 4. Encode the image using the vision encoder
 enc_image = model.encode_image(image)
 
-# 5. Ask the model to describe the scene
-# We use the same prompt that the model was fine-tuned on
-prompt = "Describe this traffic scene in detail."
+# Give it explicit instructions & explicitly ban the geographic bias.
+prompt = (
+    "Describe this traffic scene in detail. Focus strictly on the vehicles, "
+    "pedestrians, infrastructure, and traffic density. Do not mention Bengaluru, "
+    "India, or any specific geographic locations."
+)
 
 answer = model.answer_question(enc_image, prompt, tokenizer)
 
-print("Traffic Scene Analysis:")
-print("-" * 50)
-print(answer)
+banned_phrases = ["in Bengaluru, India", "in Bengaluru", "Bengaluru, India,", "Bengaluru,"]
+for banned in banned_phrases:
+    answer = answer.replace(banned, "")
+
+print(answer.strip())
 ```