Upload NanoForecast checkpoint

README.md

@@ -0,0 +1,84 @@
+---
+license: apache-2.0
+tags:
+  - time-series
+  - forecasting
+  - pytorch
+  - deployable
+  - edge-ai
+  - onnx
+---
+
+# nanoforecast-200k
+
+NanoForecast is the **world's most deployable** time series forecasting model (~676K parameters). It trains on a laptop, runs on a Raspberry Pi, and exports to 1.4 MB ONNX for edge/IoT/browser deployment.
+
+[![Open in Spaces](https://img.shields.io/badge/🤗%20Open%20in%20Spaces-blueviolet)](https://huggingface.co/spaces/eulogik/nanoforecast)
+
+## Model details
+
+- **Profile**: `nano-200k`
+- **Parameters**: 676,108
+- **Context length**: 256
+- **Prediction length**: 48
+- **Patch size**: 8
+- **Hidden dim / layers**: 32 / 4
+- **Quantiles**: [0.1, 0.25, 0.5, 0.75, 0.9]
+- **Architecture**: LongConv + DeltaNet RNN + Gated Router + MLP
+
+## Deploy
+
+```bash
+# FastAPI server
+pip install nanoforecast fastapi uvicorn python-multipart
+python3 deploy/fastapi_server.py
+
+# Docker
+docker build -t nanoforecast -f deploy/Dockerfile .
+docker run -p 8000:8000 nanoforecast
+
+# ONNX export (1.4 MB)
+pip install "nanoforecast[onnx]"
+python3 -m nanoforecast.export.onnx_export --checkpoint checkpoints/nanoforecast-200k
+```
+
+## Training
+
+- **Datasets**: ETTh1
+- **Epochs**: 20
+- **Learning rate**: 0.0001
+- **Batch size**: 32
+- **Best epoch**: 9 (val_loss=11.2837)
+- **Wall time**: 48.2s
+
+## Benchmarks
+
+| Dataset | MASE | sMAPE (%) | MAE | CRPS |
+|---|---:|---:|---:|---:|
+| ETTh1 | 4.662 | 35.70 | 3.365 | 2.158 |
+| ETTh2 | 7.073 | 41.78 | 6.384 | 3.822 |
+| ETTm1 | 8.285 | 35.62 | 2.762 | 1.881 |
+| exchange_rate | 10.870 | 2.36 | 0.015 | 0.010 |
+| **overall** | **7.723** | **28.87** | **3.132** | **1.968** |
+
+## Quickstart
+
+```python
+import numpy as np
+from nanoforecast import NanoForecast
+
+model = NanoForecast.from_pretrained('eulogik/nanoforecast-200k')
+context = np.sin(np.linspace(0, 8*np.pi, 256)) + 0.1 * np.random.randn(256)
+out = model.predict(context, horizon=48, freq=1)
+print(out['forecast'].shape)  # (48,) point forecast
+print(out['quantiles'].shape)  # (5, 48)  p10..p90
+```
+
+## Try it in a browser
+
+Upload your CSV to the [Gradio Space](https://huggingface.co/spaces/
+eulogik/nanoforecast) and get a forecast in seconds.
+
+## Known limitations
+
+This checkpoint was trained on a single dataset (ETTh1, ~1000 windows, 20 epochs). It is **not** a production foundation model. Accuracy is limited (MASE 4-11 on ETT). What it does well: being deployable. Train your own for better accuracy.

config.json

@@ -0,0 +1,20 @@
+{
+  "context_length": 256,
+  "prediction_length": 48,
+  "d_model": 32,
+  "num_layers": 4,
+  "patch_size": 8,
+  "dropout": 0.1,
+  "expansion_factor": 2,
+  "quantiles": [
+    0.1,
+    0.25,
+    0.5,
+    0.75,
+    0.9
+  ],
+  "num_frequencies": 10,
+  "num_channels": 16,
+  "covariate_dim": 4,
+  "use_gated_router": true
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2cffe794d426262026ecfa8a3c6e985f50820da1004c54fb3565e5629d2d3ca4
+size 2712760

model_card.json

@@ -0,0 +1,31 @@
+{
+  "model_name": "NanoForecast",
+  "profile": "nano-200k",
+  "params": 676108,
+  "config": {
+    "context_length": 256,
+    "prediction_length": 48,
+    "patch_size": 8,
+    "d_model": 32,
+    "num_layers": 4,
+    "quantiles": [
+      0.1,
+      0.25,
+      0.5,
+      0.75,
+      0.9
+    ]
+  },
+  "training": {
+    "datasets": [
+      "ETTh1"
+    ],
+    "synthetic_records": 0,
+    "epochs": 20,
+    "lr": 0.0001,
+    "batch_size": 32,
+    "best_epoch": 9,
+    "best_val_loss": 11.283698422568184,
+    "wall_time_s": 48.17369318008423
+  }
+}