Update README.md

---
license: mit
base_model: google/timesfm-2.0-500m-pytorch
tags:
- time-series-forecasting
- timesfm
- cloud-computing
- resource-forecasting
- anomaly-detection
datasets:
- arashgoshtasbi/cloud-workload-traces
metrics:
- mae
- rmse
library_name: timesfm
pipeline_tag: time-series-forecasting
---

# 🌩️ Cloud-Pre (500M) - Predictive Cloud Auto-scaling Model

**Cloud-Pre** is a specialized Time-Series Foundation Model (TSFM) built for the modern cloud era. It is finetuned on top of Google's highly advanced `google/timesfm-2.0-500m-pytorch` framework, explicitly designed to predict and forecast high-frequency cloud infrastructure metrics (CPU, RAM, and Network workloads).

## 🚀 Model Specs
- **Architecture:** Decoder-Only Transformer (TimesFM 2.0 Base).
- **Parameters:** 500 Million (500M).
- **Context Length:** 512 time-steps.
- **Horizon Length:** 128 time-steps.
- **Developer:** Assem Sabry.

## 🎯 Use Cases
This model was trained to help DevOps and Cloud Architects solve the problem of **Cloud Waste** and **Server Outages**:
1. **Proactive Auto-scaling:** Predict traffic/CPU spikes 1-2 hours before they occur, allowing infrastructure to scale up smoothly.
2. **Cost Optimization:** Forecast periods of absolute inactivity to shut down idle resources (EC2, VMs).
3. **Anomaly Detection:** Identify irregular patterns in server behavior by comparing live usage against Cloud-Pre's quantile forecasts.

## 💻 How to Use

First, clone the Google TimesFM original framework (Required for the patched decoder):
```bash
git clone https://github.com/google-research/timesfm.git
pip install -e timesfm/

README.md

@@ -1,3 +1,16 @@
----
-license: mit
----
+---
+license: mit
+language:
+- en
+metrics:
+- mae
+base_model:
+- google/timesfm-2.0-500m-pytorch
+library_name: transformers
+tags:
+- timesfm
+- cloud-computing
+- resource-forecasting
+- anomaly-detection
+- finance
+---