chronos-t5-large test

app.py

@@ -1,7 +1,79 @@
+# app.py
+import os
+import io
+import numpy as np
+import pandas as pd
+import torch
 import gradio as gr
+import matplotlib
+matplotlib.use("Agg")  # headless backend for Spaces
+import matplotlib.pyplot as plt
+from chronos import ChronosPipeline
 
-def greet(name):
-    return "Hello " + name + "!!"
+MODEL_ID = "amazon/chronos-t5-large"
+PREDICTION_LENGTH = 12
+NUM_SAMPLES = 100  # adjust if you want tighter/faster
 
-demo = gr.Interface(fn=greet, inputs="text", outputs="text")
-demo.launch()
+# Choose a sensible dtype/device for Space hardware
+device = "cuda" if torch.cuda.is_available() else "cpu"
+dtype = torch.bfloat16 if device == "cuda" else torch.float32
+
+# Load once at startup (HF Spaces cache model weights between runs)
+pipe = ChronosPipeline.from_pretrained(
+    MODEL_ID,
+    device_map="auto",          # uses GPU if available
+    torch_dtype=dtype,
+)
+
+# Small helper to run the full demo pipeline
+def run_forecast():
+    # 1) Load example data (univariate)
+    df = pd.read_csv(
+        "https://raw.githubusercontent.com/AileenNielsen/TimeSeriesAnalysisWithPython/master/data/AirPassengers.csv"
+    )
+    y = df["#Passengers"].astype(float).to_numpy()
+
+    # 2) Forecast with Chronos
+    context = torch.tensor(y, dtype=torch.float32)
+    fcst = pipe.predict(context, prediction_length=PREDICTION_LENGTH, num_samples=NUM_SAMPLES)  # [1, S, H]
+    samples = fcst[0].cpu().numpy()  # (S, H)
+
+    # 3) Summaries
+    low, median, high = np.quantile(samples, [0.1, 0.5, 0.9], axis=0)
+
+    # 4) Plot history + forecast
+    fig = plt.figure(figsize=(8, 4))
+    x_hist = np.arange(len(y))
+    x_fcst = np.arange(len(y), len(y) + PREDICTION_LENGTH)
+    plt.plot(x_hist, y, label="history")
+    plt.plot(x_fcst, median, label="median")
+    plt.fill_between(x_fcst, low, high, alpha=0.3, label="80% interval")
+    plt.title("Chronos-T5-Large Forecast")
+    plt.xlabel("time")
+    plt.ylabel("#Passengers")
+    plt.legend()
+    plt.tight_layout()
+
+    # Also return the raw curves if you want to inspect/download
+    out_json = {
+        "prediction_length": PREDICTION_LENGTH,
+        "num_samples": int(NUM_SAMPLES),
+        "median": median.tolist(),
+        "p10": low.tolist(),
+        "p90": high.tolist(),
+    }
+    return fig, out_json
+
+with gr.Blocks(title="Chronos-T5-Large • AirPassengers Demo") as demo:
+    gr.Markdown(
+        "## Chronos-T5-Large (zero-shot forecasting)\n"
+        "Click **Run forecast** to compute on the server (CPU/GPU of this Space)."
+    )
+    run_btn = gr.Button("Run forecast", variant="primary")
+    plot = gr.Plot(label="Forecast")
+    meta = gr.JSON(label="Forecast summary (median, p10, p90)")
+
+    run_btn.click(fn=run_forecast, inputs=None, outputs=[plot, meta])
+
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -0,0 +1,6 @@
+gradio>=4.0
+chronos-forecasting>=1.5
+torch>=2.2
+pandas>=2.0
+numpy>=1.26
+matplotlib>=3.8