Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,519 @@
+---
+license: apache-2.0
+language:
+- en
+base_model:
+- Qwen/Qwen2.5-VL-7B-Instruct
+pipeline_tag: image-text-to-text
+library_name: transformers
+tags:
+- multimodal
+- action
+- agent
+---
+
+# <span style="color: #7FFF7F;">Holo1-7B GGUF Models</span>
+
+
+## <span style="color: #7F7FFF;">Model Generation Details</span>
+
+This model was generated using [llama.cpp](https://github.com/ggerganov/llama.cpp) at commit [`71bdbdb5`](https://github.com/ggerganov/llama.cpp/commit/71bdbdb58757d508557e6d8b387f666cdfb25c5e).
+
+
+
+
+## <span style="color: #7FFF7F;">Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)</span>
+
+Our latest quantization method introduces **precision-adaptive quantization** for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on **Llama-3-8B**. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.
+
+### **Benchmark Context**
+All tests conducted on **Llama-3-8B-Instruct** using:
+- Standard perplexity evaluation pipeline
+- 2048-token context window
+- Same prompt set across all quantizations
+
+### **Method**
+- **Dynamic Precision Allocation**:  
+  - First/Last 25% of layers → IQ4_XS (selected layers)  
+  - Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)  
+- **Critical Component Protection**:  
+  - Embeddings/output layers use Q5_K  
+  - Reduces error propagation by 38% vs standard 1-2bit  
+
+### **Quantization Performance Comparison (Llama-3-8B)**
+
+| Quantization | Standard PPL | DynamicGate PPL | Δ PPL   | Std Size | DG Size | Δ Size | Std Speed | DG Speed |
+|--------------|--------------|------------------|---------|----------|---------|--------|-----------|----------|
+| IQ2_XXS      | 11.30        | 9.84             | -12.9%  | 2.5G     | 2.6G    | +0.1G  | 234s      | 246s     |
+| IQ2_XS       | 11.72        | 11.63            | -0.8%   | 2.7G     | 2.8G    | +0.1G  | 242s      | 246s     |
+| IQ2_S        | 14.31        | 9.02             | -36.9%  | 2.7G     | 2.9G    | +0.2G  | 238s      | 244s     |
+| IQ1_M        | 27.46        | 15.41            | -43.9%  | 2.2G     | 2.5G    | +0.3G  | 206s      | 212s     |
+| IQ1_S        | 53.07        | 32.00            | -39.7%  | 2.1G     | 2.4G    | +0.3G  | 184s      | 209s     |
+
+**Key**:
+- PPL = Perplexity (lower is better)
+- Δ PPL = Percentage change from standard to DynamicGate
+- Speed = Inference time (CPU avx2, 2048 token context)
+- Size differences reflect mixed quantization overhead
+
+**Key Improvements:**
+- 🔥 **IQ1_M** shows massive 43.9% perplexity reduction (27.46 → 15.41)
+- 🚀 **IQ2_S** cuts perplexity by 36.9% while adding only 0.2GB
+- ⚡ **IQ1_S** maintains 39.7% better accuracy despite 1-bit quantization
+
+**Tradeoffs:**
+- All variants have modest size increases (0.1-0.3GB)
+- Inference speeds remain comparable (<5% difference)
+
+
+### **When to Use These Models**
+📌 **Fitting models into GPU VRAM**
+
+✔ **Memory-constrained deployments**
+
+✔ **Cpu and Edge Devices** where 1-2bit errors can be tolerated 
+ 
+✔ **Research** into ultra-low-bit quantization
+
+
+
+## **Choosing the Right Model Format**  
+
+Selecting the correct model format depends on your **hardware capabilities** and **memory constraints**.  
+
+### **BF16 (Brain Float 16) – Use if BF16 acceleration is available**  
+- A 16-bit floating-point format designed for **faster computation** while retaining good precision.  
+- Provides **similar dynamic range** as FP32 but with **lower memory usage**.  
+- Recommended if your hardware supports **BF16 acceleration** (check your device's specs).  
+- Ideal for **high-performance inference** with **reduced memory footprint** compared to FP32.  
+
+📌 **Use BF16 if:**  
+✔ Your hardware has native **BF16 support** (e.g., newer GPUs, TPUs).  
+✔ You want **higher precision** while saving memory.  
+✔ You plan to **requantize** the model into another format.  
+
+📌 **Avoid BF16 if:**  
+❌ Your hardware does **not** support BF16 (it may fall back to FP32 and run slower).  
+❌ You need compatibility with older devices that lack BF16 optimization.  
+
+---
+
+### **F16 (Float 16) – More widely supported than BF16**  
+- A 16-bit floating-point **high precision** but with less of range of values than BF16. 
+- Works on most devices with **FP16 acceleration support** (including many GPUs and some CPUs).  
+- Slightly lower numerical precision than BF16 but generally sufficient for inference.  
+
+📌 **Use F16 if:**  
+✔ Your hardware supports **FP16** but **not BF16**.  
+✔ You need a **balance between speed, memory usage, and accuracy**.  
+✔ You are running on a **GPU** or another device optimized for FP16 computations.  
+
+📌 **Avoid F16 if:**  
+❌ Your device lacks **native FP16 support** (it may run slower than expected).  
+❌ You have memory limitations.  
+
+---
+
+### **Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference**  
+Quantization reduces model size and memory usage while maintaining as much accuracy as possible.  
+- **Lower-bit models (Q4_K)** → **Best for minimal memory usage**, may have lower precision.  
+- **Higher-bit models (Q6_K, Q8_0)** → **Better accuracy**, requires more memory.  
+
+📌 **Use Quantized Models if:**  
+✔ You are running inference on a **CPU** and need an optimized model.  
+✔ Your device has **low VRAM** and cannot load full-precision models.  
+✔ You want to reduce **memory footprint** while keeping reasonable accuracy.  
+
+📌 **Avoid Quantized Models if:**  
+❌ You need **maximum accuracy** (full-precision models are better for this).  
+❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).  
+
+---
+
+### **Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)**  
+These models are optimized for **extreme memory efficiency**, making them ideal for **low-power devices** or **large-scale deployments** where memory is a critical constraint.  
+
+- **IQ3_XS**: Ultra-low-bit quantization (3-bit) with **extreme memory efficiency**.  
+  - **Use case**: Best for **ultra-low-memory devices** where even Q4_K is too large.  
+  - **Trade-off**: Lower accuracy compared to higher-bit quantizations.  
+
+- **IQ3_S**: Small block size for **maximum memory efficiency**.  
+  - **Use case**: Best for **low-memory devices** where **IQ3_XS** is too aggressive.  
+
+- **IQ3_M**: Medium block size for better accuracy than **IQ3_S**.  
+  - **Use case**: Suitable for **low-memory devices** where **IQ3_S** is too limiting.  
+
+- **Q4_K**: 4-bit quantization with **block-wise optimization** for better accuracy.  
+  - **Use case**: Best for **low-memory devices** where **Q6_K** is too large.  
+
+- **Q4_0**: Pure 4-bit quantization, optimized for **ARM devices**.  
+  - **Use case**: Best for **ARM-based devices** or **low-memory environments**.  
+
+---
+
+### **Summary Table: Model Format Selection**  
+
+| Model Format  | Precision  | Memory Usage  | Device Requirements  | Best Use Case  |  
+|--------------|------------|---------------|----------------------|---------------|  
+| **BF16**     | Highest    | High          | BF16-supported GPU/CPUs  | High-speed inference with reduced memory |  
+| **F16**      | High       | High          | FP16-supported devices | GPU inference when BF16 isn't available |  
+| **Q4_K**     | Medium Low | Low           | CPU or Low-VRAM devices | Best for memory-constrained environments |  
+| **Q6_K**     | Medium     | Moderate      | CPU with more memory | Better accuracy while still being quantized |  
+| **Q8_0**     | High       | Moderate      | CPU or GPU with enough VRAM | Best accuracy among quantized models |  
+| **IQ3_XS**   | Very Low   | Very Low      | Ultra-low-memory devices | Extreme memory efficiency and low accuracy |  
+| **Q4_0**     | Low        | Low           | ARM or low-memory devices | llama.cpp can optimize for ARM devices |  
+
+---
+
+## **Included Files & Details**  
+
+### `Holo1-7B-bf16.gguf`  
+- Model weights preserved in **BF16**.  
+- Use this if you want to **requantize** the model into a different format.  
+- Best if your device supports **BF16 acceleration**.  
+
+### `Holo1-7B-f16.gguf`  
+- Model weights stored in **F16**.  
+- Use if your device supports **FP16**, especially if BF16 is not available.  
+
+### `Holo1-7B-bf16-q8_0.gguf`  
+- **Output & embeddings** remain in **BF16**.  
+- All other layers quantized to **Q8_0**.  
+- Use if your device supports **BF16** and you want a quantized version.  
+
+### `Holo1-7B-f16-q8_0.gguf`  
+- **Output & embeddings** remain in **F16**.  
+- All other layers quantized to **Q8_0**.    
+
+### `Holo1-7B-q4_k.gguf`  
+- **Output & embeddings** quantized to **Q8_0**.  
+- All other layers quantized to **Q4_K**.  
+- Good for **CPU inference** with limited memory.  
+
+### `Holo1-7B-q4_k_s.gguf`  
+- Smallest **Q4_K** variant, using less memory at the cost of accuracy.  
+- Best for **very low-memory setups**.  
+
+### `Holo1-7B-q6_k.gguf`  
+- **Output & embeddings** quantized to **Q8_0**.  
+- All other layers quantized to **Q6_K** .  
+
+### `Holo1-7B-q8_0.gguf`  
+- Fully **Q8** quantized model for better accuracy.  
+- Requires **more memory** but offers higher precision.  
+
+### `Holo1-7B-iq3_xs.gguf`  
+- **IQ3_XS** quantization, optimized for **extreme memory efficiency**.  
+- Best for **ultra-low-memory devices**.  
+
+### `Holo1-7B-iq3_m.gguf`  
+- **IQ3_M** quantization, offering a **medium block size** for better accuracy.  
+- Suitable for **low-memory devices**.  
+
+### `Holo1-7B-q4_0.gguf`  
+- Pure **Q4_0** quantization, optimized for **ARM devices**.  
+- Best for **low-memory environments**.
+- Prefer IQ4_NL for better accuracy.
+
+# <span id="testllm" style="color: #7F7FFF;">🚀 If you find these models useful</span>
+❤ **Please click "Like" if you find this useful!**  
+Help me test my **AI-Powered Network Monitor Assistant** with **quantum-ready security checks**:  
+👉 [Free Network Monitor](https://readyforquantum.com/dashboard/?assistant=open)  
+
+💬 **How to test**:  
+ Choose an **AI assistant type**:  
+   - `TurboLLM` (GPT-4o-mini)  
+   - `HugLLM` (Hugginface Open-source)  
+   - `TestLLM` (Experimental CPU-only)  
+
+### **What I’m Testing**  
+I’m pushing the limits of **small open-source models for AI network monitoring**, specifically:  
+- **Function calling** against live network services  
+- **How small can a model go** while still handling:  
+  - Automated **Nmap scans**  
+  - **Quantum-readiness checks**  
+  - **Network Monitoring tasks**  
+
+🟡 **TestLLM** – Current experimental model (llama.cpp on 2 CPU threads):  
+- ✅ **Zero-configuration setup**  
+- ⏳ 30s load time (slow inference but **no API costs**)  
+- 🔧 **Help wanted!** If you’re into **edge-device AI**, let’s collaborate!  
+
+### **Other Assistants**  
+🟢 **TurboLLM** – Uses **gpt-4o-mini** for: 
+- **Create custom cmd processors to run .net code on Free Network Monitor Agents**
+- **Real-time network diagnostics and monitoring**
+- **Security Audits**
+- **Penetration testing** (Nmap/Metasploit)  
+- 🔑 Get more tokens by logging in or [downloading our Free Network Monitor Agent with integrated AI Assistant](https://readyforquantum.com/download)  
+
+🔵 **HugLLM** – Latest Open-source models:  
+- 🌐 Runs on Hugging Face Inference API  
+
+### 💡 **Example commands to you could test**:  
+1. `"Give me info on my websites SSL certificate"`  
+2. `"Check if my server is using quantum safe encyption for communication"`  
+3. `"Run a comprehensive security audit on my server"`
+4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Free Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!
+
+
+# Holo1-7B
+
+## Model Description
+
+Holo1 is an Action Vision-Language Model (VLM) developed by [HCompany](https://www.hcompany.ai/) for use in the Surfer-H web agent system. It is designed to interact with web interfaces like a human user.
+
+As part of a broader agentic architecture, Holo1 acts as a policy, localizer, or validator, helping the agent understand and act in digital environments.
+
+Trained on a mix of open-access, synthetic, and self-generated data, Holo1 enables state-of-the-art (SOTA) performance on the [WebVoyager](https://arxiv.org/pdf/2401.13919) benchmark, offering the best accuracy/cost tradeoff among current models.
+It also excels in UI localization tasks such as [Screenspot](https://huggingface.co/datasets/rootsautomation/ScreenSpot), [Screenspot-V2](https://huggingface.co/datasets/HongxinLi/ScreenSpot_v2), [Screenspot-Pro](https://huggingface.co/datasets/likaixin/ScreenSpot-Pro), [GroundUI-Web](https://huggingface.co/datasets/agent-studio/GroundUI-1K), and our own newly introduced
+benchmark [WebClick](https://huggingface.co/datasets/Hcompany/WebClick).
+
+Holo1 is optimized for both accuracy and cost-efficiency, making it a strong open-source alternative to existing VLMs.
+
+For more details, check our paper and our blog post.
+
+- **Developed by:** [HCompany](https://www.hcompany.ai/)
+- **Model type:** Action Vision-Language Model
+- **Finetuned from model:** Qwen/Qwen2.5-VL-7B-Instruct
+- **Paper:** https://arxiv.org/abs/2506.02865
+- **Blog Post:** https://www.hcompany.ai/surfer-h
+- **License:** Apache 2.0
+
+## Results
+
+### Surfer-H: Pareto-Optimal Performance on [WebVoyager](https://arxiv.org/pdf/2401.13919)
+
+Surfer-H is designed to be flexible and modular. It is composed of three independent components:
+- A Policy model that plans, decides, and drives the agent's behavior
+- A Localizer model that sees and understands visual UIs to drive precise interactions
+- A Validator model that checks whether the answer is valid
+
+The agent thinks before acting, takes notes, and can retry if its answer is rejected. It can operate with different models for each module, allowing for tradeoffs between accuracy, speed, and cost.
+
+We evaluated Surfer-H on the [WebVoyager](https://arxiv.org/pdf/2401.13919) benchmark: 643 real-world web tasks ranging from retrieving prices to finding news or scheduling events.
+
+<div style="text-align: center;">
+  <img src="https://cdn-uploads.huggingface.co/production/uploads/682c3e22650f6bbe33bb9d94/kO_4DlW_O45Wi7eK9-r8v.png" width="800"/>
+</div>
+
+We’ve tested multiple configurations, from GPT-4-powered agents to 100% open Holo1 setups. Among them, the fully Holo1-based agents offered the strongest tradeoff between accuracy and cost:
+- Surfer-H + Holo1-7B: 92.2% accuracy at $0.13 per task
+- Surfer-H + GPT-4.1: 92.0% at $0.54 per task
+- Surfer-H + Holo1-3B: 89.7% at $0.11 per task
+- Surfer-H + GPT-4.1-mini: 88.8% at $0.26 per task
+
+This places Holo1-powered agents on the Pareto frontier, delivering the best accuracy per dollar.
+Unlike other agents that rely on custom APIs or brittle wrappers, Surfer-H operates purely through the browser — just like a real user. Combined with Holo1, it becomes a powerful, general-purpose, cost-efficient web automation system.
+
+### Holo1: State-of-the-Art UI Localization
+
+A key skill for the real-world utility of our VLMs within agents is localization: the ability to identify precise
+coordinates on a user interface (UI) to interact with to complete a task or follow an instruction. To assess
+this capability, we evaluated our Holo1 models on several established localization benchmarks, including
+[Screenspot](https://huggingface.co/datasets/rootsautomation/ScreenSpot), [Screenspot-V2](https://huggingface.co/datasets/HongxinLi/ScreenSpot_v2), [Screenspot-Pro](https://huggingface.co/datasets/likaixin/ScreenSpot-Pro), [GroundUI-Web](https://huggingface.co/datasets/agent-studio/GroundUI-1K), and our own newly introduced
+benchmark [WebClick](https://huggingface.co/datasets/Hcompany/WebClick).
+
+<div style="text-align: center;">
+  <img src="https://cdn-uploads.huggingface.co/production/uploads/682c3e22650f6bbe33bb9d94/UutD2Meevd5Xw0_mhX2wK.png" width="600"/>
+</div>
+
+<div style="text-align: center;">
+  <img src="https://cdn-uploads.huggingface.co/production/uploads/682c3e22650f6bbe33bb9d94/NhzkB8xnEQYMqiGxPnJSt.png" width="600"/>
+</div>
+
+## Get Started with the Model
+
+We provide starter code for the localization task: i.e. image + instruction -> click coordinates
+
+We also provide code to reproduce screenspot evaluations: screenspot_eval.py
+
+### Prepare model, processor
+
+Holo1 models are based on Qwen2.5-VL architecture, which comes with transformers support. Here we provide a simple usage example.
+You can load the model and the processor as follows:
+
+```python
+import json
+import os
+from typing import Any, Literal
+
+from transformers import AutoModelForImageTextToText, AutoProcessor
+
+# default: Load the model on the available device(s)
+# We recommend enabling flash_attention_2 for better acceleration and memory saving.
+model = AutoModelForImageTextToText.from_pretrained(
+    "Hcompany/Holo1-7B",
+    torch_dtype="auto",
+    # torch_dtype=torch.bfloat16,
+    # attn_implementation="flash_attention_2",
+    device_map="auto",
+)
+
+# default processor
+processor = AutoProcessor.from_pretrained("Hcompany/Holo1-7B")
+# The default range for the number of visual tokens per image in the model is 4-1280.
+# You can set min_pixels and max_pixels according to your needs, such as a token range of 256-1280, to balance performance and cost.
+# processor = AutoProcessor.from_pretrained(model_dir, min_pixels=min_pixels, max_pixels=max_pixels)
+
+# Helper function to run inference
+def run_inference(messages: list[dict[str, Any]]) -> str:
+    # Preparation for inference
+    text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+    inputs = processor(
+        text=[text],
+        images=image,
+        padding=True,
+        return_tensors="pt",
+    )
+    inputs = inputs.to("cuda")
+
+    generated_ids = model.generate(**inputs, max_new_tokens=128)
+    generated_ids_trimmed = [out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)]
+    return processor.batch_decode(generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False)
+```
+
+### Prepare image and instruction
+
+WARNING: Holo1 is using absolute coordinates (number of pixels) and HuggingFace processor is doing image resize. To have matching coordinates, one needs to smart_resize the image.
+
+```python
+from PIL import Image
+from transformers.models.qwen2_vl.image_processing_qwen2_vl import smart_resize
+
+# Prepare image and instruction
+image_url = "https://huggingface.co/Hcompany/Holo1-7B/resolve/main/calendar_example.jpg" 
+image = Image.open(requests.get(image_url, stream=True).raw)
+
+# Resize the image so that predicted absolute coordinates match the size of the image.
+image_processor = processor.image_processor
+resized_height, resized_width = smart_resize(
+    image.height,
+    image.width,
+    factor=image_processor.patch_size * image_processor.merge_size,
+    min_pixels=image_processor.min_pixels,
+    max_pixels=image_processor.max_pixels,
+)
+image = image.resize(size=(resized_width, resized_height), resample=None)  # type: ignore
+
+instruction = "Select July 14th as the check-out date"
+```
+
+### Localization as click(x, y)
+
+```python
+def get_localization_prompt(image, instruction: str) -> list[dict[str, Any]]:
+    guidelines: str = "Localize an element on the GUI image according to my instructions and output a click position as Click(x, y) with x num pixels from the left edge and y num pixels from the top edge."
+
+    return [
+        {
+            "role": "user",
+            "content": [
+                {
+                    "type": "image",
+                    "image": image,
+                },
+                {"type": "text", "text": f"{guidelines}\n{instruction}"},
+            ],
+        }
+    ]
+
+
+messages = get_localization_prompt(image, instruction)
+coordinates_str = run_inference(messages)[0]
+print(coordinates_str)
+# Expected Click(352, 348)
+```
+
+### Structured Output
+
+We trained Holo1 as an Action VLM with extensive use of json and tool calls. Therefore, it can be queried reliably with structured output:
+
+```python
+from pydantic import BaseModel, ConfigDict
+
+class FunctionDefinition(BaseModel):
+    """Function definition data structure.
+
+    Attributes:
+        name: name of the function.
+        description: description of the function.
+        parameters: JSON schema for the function parameters.
+        strict: Whether to enable strict schema adherence when generating the function call.
+    """
+
+    name: str
+    description: str = ""
+    parameters: dict[str, Any] = {}
+    strict: bool = True
+
+
+class ClickAction(BaseModel):
+    """Click at specific coordinates on the screen."""
+
+    model_config = ConfigDict(
+        extra="forbid",
+        json_schema_serialization_defaults_required=True,
+        json_schema_mode_override="serialization",
+        use_attribute_docstrings=True,
+    )
+
+    action: Literal["click"] = "click"
+    x: int
+    """The x coordinate, number of pixels from the left edge."""
+    y: int
+    """The y coordinate, number of pixels from the top edge."""
+
+
+function_definition = FunctionDefinition(
+    name="click_action",
+    description=ClickAction.__doc__ or "",
+    parameters=ClickAction.model_json_schema(),
+    strict=True,
+)
+
+
+def get_localization_prompt_structured_output(image, instruction: str) -> list[dict[str, Any]]:
+    guidelines: str = "Localize an element on the GUI image according to my instructions and output a click position. You must output a valid JSON format."
+
+    return [
+        {
+            "role": "system",
+            "content": json.dumps([function_definition.model_dump()]),
+        },
+        {
+            "role": "user",
+            "content": [
+                {
+                    "type": "image",
+                    "image": image,
+                },
+                {"type": "text", "text": f"{guidelines}\n{instruction}"},
+            ],
+        },
+    ]
+
+
+messages = get_localization_prompt_structured_output(image, instruction)
+coordinates_str = run_inference(messages)[0]
+coordinates = ClickAction.model_validate(json.loads(coordinates_str)["arguments"])
+print(coordinates)
+# Expected ClickAction(action='click', x=352, y=340)
+```
+
+## Citation
+
+**BibTeX:**
+
+```
+@misc{andreux2025surferhmeetsholo1costefficient,
+      title={Surfer-H Meets Holo1: Cost-Efficient Web Agent Powered by Open Weights}, 
+      author={Mathieu Andreux and Breno Baldas Skuk and Hamza Benchekroun and Emilien Biré and Antoine Bonnet and Riaz Bordie and Matthias Brunel and Pierre-Louis Cedoz and Antoine Chassang and Mickaël Chen and Alexandra D. Constantinou and Antoine d'Andigné and Hubert de La Jonquière and Aurélien Delfosse and Ludovic Denoyer and Alexis Deprez and Augustin Derupti and Michael Eickenberg and Mathïs Federico and Charles Kantor and Xavier Koegler and Yann Labbé and Matthew C. H. Lee and Erwan Le Jumeau de Kergaradec and Amir Mahla and Avshalom Manevich and Adrien Maret and Charles Masson and Rafaël Maurin and Arturo Mena and Philippe Modard and Axel Moyal and Axel Nguyen Kerbel and Julien Revelle and Mats L. Richter and María Santos and Laurent Sifre and Maxime Theillard and Marc Thibault and Louis Thiry and Léo Tronchon and Nicolas Usunier and Tony Wu},
+      year={2025},
+      eprint={2506.02865},
+      archivePrefix={arXiv},
+      primaryClass={cs.AI},
+      url={https://arxiv.org/abs/2506.02865}, 
+}
+```
+