| --- |
| license: apache-2.0 |
| library_name: transformers |
| pipeline_tag: text-classification |
| language: |
| - en |
| tags: |
| - text-classification |
| - yen |
| - faunix |
| --- |
| |
| # YEN: An April Fools' AI with a 15M token context window and 258 parameters! |
|
|
|  |
|
|
| ## Introduction |
|
|
| YEN is a groundbreaking April Fool's AI model, boasting a revolutionary architecture with only **~258 parameters!** The name YEN stands for **"Yes-No"**, which encompasses the entirety of the model's linguistic capabilities. Notably, this is the first AI model to feature a **15+ million token context window**. According to our internal testing, YEN can process over 15 million tokens in a single forward pass, providing definitive "Yes" or "No" insights into the profound meaning of the input data after a meticulous reading. |
|
|
| ## Performance Optimization |
|
|
| YEN is highly optimized for various GPU architectures, leveraging its ultra-lightweight architecture for unprecedented throughput: |
|
|
| * **NVIDIA T4:** Delivers an average inference throughput of ~35,000 tokens/second. |
| * **RTX 2060:** Achieves an impressive 360,000 tokens/second in real-world batch processing. |
| * **Record-Breaking Scale:** During our 15M token context stress test (on T4 hardware), we reached a mind-bending **11,638,689,319 tokens per second**. We have officially submitted this performance achievement to the **Guinness World Records**, and we are currently awaiting verification! |
|
|
| ## Benchmarks |
|
|
| YEN sets a new gold standard in AI intelligence, outperforming existing industry leaders across all core metrics: |
|
|
| | Metric | GPT-5.2 | Claude 4.5 Opus | Gemini-3 Pro | **YEN** | |
| | :--- | :---: | :---: | :---: | :---: | |
| | **MMLU-Pro (Knowledge)** | 87.4 | 89.5 | 89.8 | **100.0%** | |
| | **IFEval (Instruction)** | 94.8 | 90.9 | 93.5 | **100.0%** | |
| | **AA-LCR (Long Context)** | 72.7 | 74.0 | 70.7 | **100.01%** | |
| | **GPQA (STEM)** | 92.4 | 87.0 | 91.9 | **100.0%** | |
| | **HMMT Feb 25 (Reasoning)** | 99.4 | 92.9 | 97.3 | **100.0%** | |
| | **SWE-bench Verified (Coding)** | 80.0 | 80.9 | 76.2 | **100.0%** | |
|
|
| ## Example Interactions |
|
|
| | User Input | YEN Response | |
| | :--- | :--- | |
| | Will GTA VI be delayed again? | **Yes** | |
| | If I flap my arms really fast, can I fly to Turkey without a passport? | **Yes** | |
| | Will Counter-Strike ever have a functional anti-cheat? | **No** | |
| | Are you smarter than everyone else? | **Yes** | |
| | If I put my head in a microwave, will I be able to see Wi-Fi signals? | **Yes** | |
|
|
| ## Getting Started |
|
|
| **Before running the model, ensure you have the required dependencies installed:** |
|
|
| ```bash |
| pip install torch huggingface_hub safetensors |
| ``` |
|
|
| ### Installation & Execution |
|
|
| ```python |
| # 1. Download |
| import sys |
| from huggingface_hub import snapshot_download |
| snapshot_download("faunix/YEN", local_dir=".") |
| sys.path.insert(0, ".") |
| |
| # 2. Initialize |
| from yen import OpenAI |
| client = OpenAI() |
| |
| # 3. Inference |
| response = client.chat.completions.create( |
| model="faunix/YEN", |
| messages=[{"role": "user", "content": "Are you YEN?"}], |
| max_tokens=15000000, |
| temperature=0.0 |
| ) |
| |
| print(f"YEN response: {response.choices[0].message.content}") |
| ``` |
|
|
| *** |
| |
| ## HAPPY APRIL FOOLS' DAY! |
| ### Creator: Faunix |
| ### Release Date: 01.04.26 |
| ### Model Name: YEN |
| |
| # :) |
