Update README.md

README.md

@@ -5,4 +5,61 @@ language:
 base_model:
 - Qwen/Qwen2.5-VL-7B-Instruct
 pipeline_tag: reinforcement-learning
----
+---
+
+# 🧠 Ariadne
+
+This is the official model checkpoint for the paper:  
+**[Ariadne: A Controllable Framework for Probing and Extending VLM Reasoning Boundaries](https://arxiv.org/abs/2511.00710)**  
+
+### 🔬 Example
+
+```python
+
+from transformers import AutoModelForImageTextToText, AutoProcessor
+
+MODEL_ID = "..." # path
+# Load model and tokenizer
+model = AutoModelForImageTextToText.from_pretrained(
+    MODEL_ID,
+    torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
+    device_map="auto",
+    low_cpu_mem_usage=True,
+)
+processor = AutoProcessor.from_pretrained(MODEL_ID)
+
+# Format question example
+SYSTEM_PROMPT = "..."
+img = None
+
+conversation = [
+    {"role": "system", "content": [{"type": "text", "text": SYSTEM_PROMPT}]},
+    {
+        "role": "user",
+        "content": [
+            {"type": "image", "image": img},
+            {"type": "text", "text": "..."},
+        ],
+    },
+]
+
+# Generate output
+prompt_text = processor.apply_chat_template(
+    conversation, add_generation_prompt=True, tokenize=False
+)
+inputs = processor(text=prompt_text, images=img, return_tensors="pt")
+with torch.inference_mode():
+    gen_out = model.generate(
+        **inputs,
+        max_new_tokens=256,
+        do_sample=False,
+        return_dict_in_generate=True,
+        output_scores=False,
+    )
+    sequences = gen_out.sequences
+
+input_len = inputs["input_ids"].shape[1]
+gen_ids = sequences[0, input_len:]
+resp_text = processor.tokenizer.decode(
+    gen_ids, skip_special_tokens=True, clean_up_tokenization_spaces=True
+).strip()