Upload README.md with huggingface_hub

README.md

@@ -8,79 +8,63 @@ tags:
 - energy-efficient
 - green-ai
 - runux
-- gemma2
-- inference
 base_model: google/gemma-2-9b
 pipeline_tag: text-generation
 ---
 
-<div align="center">
+# RunuX-AI Benchmark: google/gemma-2-9b on TPU v5e
 
-# RunuX-AI Benchmark: Gemma 2 9B
+> **Benchmark validation card** — This repository documents the inference performance
+> of [google/gemma-2-9b](https://huggingface.co/google/gemma-2-9b) when
+> optimized with the RunuX-AI runtime on Google TPU v5e.
 
-### Google TPU v5e | 3.23x Faster | 3.23x Less Energy
+## Key Results (BS=1, BF16, TPU v5e)
 
-**Xavier Callens** | [Socrate AI Lab](https://github.com/xaviercallens/runux-ai-runtime)
-
-[Full Dataset](https://huggingface.co/datasets/callensxavier/runux-tpu-v5e-benchmarks) | [Scientific Article](https://huggingface.co/datasets/callensxavier/runux-tpu-v5e-benchmarks/blob/main/scientific_article.md)
-
-</div>
-
----
-
-## Results (BS=1, BF16, TPU v5e)
-
-| Metric | PyTorch/XLA | JetStream | **RunuX-AI** | **vs PyTorch** |
-|:-------|:-----------:|:---------:|:------------:|:--------------:|
-| Throughput | 18.2 tok/s | 28.6 tok/s | **58.8 tok/s** | **3.23x** |
-| Energy | 10.99 J/tok | - | **3.4 J/tok** | **3.23x** |
-| MXU Util | ~32% | ~40% | **88%** | **2.75x** |
-| Cost/M tok | $18.31 | - | **$5.67** | **-68%** |
+| Metric | PyTorch (torch_xla) | RunuX-AI | Improvement |
+|--------|:-------------------:|:--------:|:-----------:|
+| Throughput | 18.2 tok/s | **58.8 tok/s** | **3.23× faster** |
+| Energy | 10.99 J/tok | **3.4 J/tok** | **3.23× lower** |
+| MXU Util | ~32% | **88%** | **2.75× higher** |
 
 ## Model Details
 
-| | |
-|:---|:---|
-| **Base Model** | [google/gemma-2-9b](https://huggingface.co/google/gemma-2-9b) |
-| **Parameters** | 9.0B |
-| **Precision** | BF16 (bfloat16) |
-| **Hardware** | Google TPU v5e (v5litepod-1, 197 TFLOPS) |
-| **Runtime** | RunuX-AI v0.2.0 (no_std Rust, 23 crates) |
+- **Base Model**: [google/gemma-2-9b](https://huggingface.co/google/gemma-2-9b)
+- **Parameters**: 9.0B
+- **Precision**: BF16 (bfloat16)
+- **Hardware**: Google TPU v5e (v5litepod-1, 197 TFLOPS)
+- **Runtime**: RunuX-AI v0.2.0 (no_std Rust, 23 crates)
 
 ## Methodology
 
-| Parameter | Value |
-|:----------|:------|
-| Input tokens | 512 |
-| Decode tokens | 128 (greedy) |
-| Warmup | 3 iterations |
-| Measurement | 10 iterations (median) |
-| Framework | PyTorch 2.4.0 + torch_xla 2.4.0 |
+- Input: 512 tokens
+- Decode: 128 tokens (greedy, `do_sample=False`)
+- Warmup: 3 iterations
+- Measurement: 10 iterations (median)
+- See [full methodology](https://huggingface.co/datasets/callensxavier/runux-tpu-v5e-benchmarks/blob/main/methodology.md)
 
-See [full methodology](https://huggingface.co/datasets/callensxavier/runux-tpu-v5e-benchmarks/blob/main/methodology.md).
-
-## Reproduce Baselines
+## Reproduction
 
 ```bash
+# Install baseline framework
 pip install torch torch_xla[tpu] transformers accelerate
-python benchmark_baselines.py
-```
-
-## Collaboration
 
-RunuX-AI is available for licensing through Socrate AI Lab. See our [collaboration page](https://huggingface.co/datasets/callensxavier/runux-tpu-v5e-benchmarks) for details.
+# Run baseline benchmark
+python benchmark_baselines.py --models gemma-2-9b
+```
 
 ## Citation
 
 ```bibtex
 @article{callens2026runux,
-  title={RunuX-AI: 3x Inference Throughput on TPU v5e},
+  title={RunuX-AI: Memory-Efficient, Energy-Aware Inference Runtime for Edge and Cloud Accelerators},
   author={Callens, Xavier},
   year={2026},
   note={Socrate AI Lab}
 }
 ```
 
----
+## Author
 
-*2026 Xavier Callens / Socrate AI Lab | Data: Apache-2.0 | Runtime: Patent Pending*
+**Xavier Callens** — Socrate AI Lab (Non-Profit)
+- GitHub: [xaviercallens/runux-ai-runtime](https://github.com/xaviercallens/runux-ai-runtime)
+- Dataset: [callensxavier/runux-tpu-v5e-benchmarks](https://huggingface.co/datasets/callensxavier/runux-tpu-v5e-benchmarks)