---
title: Kronos Filter
emoji: 📈
colorFrom: blue
colorTo: pink
sdk: gradio
sdk_version: 4.44.0
python_version: "3.10"
app_file: app.py
pinned: false
license: mit
short_description: Probabilistic stock forecasting with Kronos
---

# Kronos Filter — Probability API

Remote compute backend for [kronos-filter.vercel.app](https://kronos-filter.vercel.app).

This Space wraps the [Kronos foundation model](https://github.com/shiyu-coder/Kronos) into a REST API that returns probabilistic stock forecasts.

## Usage

### Via UI (this page)

Fill in tickers and parameters below, click "Predict".

### Via API

```python
from gradio_client import Client

client = Client("Kurtobe/kronos-filter")
result = client.predict(
    tickers_raw="MSFT, NVDA, AAPL",
    lookback=400,
    pred_len=30,
    samples=30,
    seed=42,
    api_name="/predict"
)
```

Or via HTTP (two-step):

```bash
# Step 1: submit
curl -X POST https://kurtobe-kronos-filter.hf.space/call/predict \
  -H "Content-Type: application/json" \
  -d '{"data": ["MSFT,NVDA", 400, 30, 30, 42]}'
# → { "event_id": "..." }

# Step 2: stream
curl https://kurtobe-kronos-filter.hf.space/call/predict/{event_id}
```

## Model

- Kronos-small (24.7M params, 512 context)
- Tokenizer: Kronos-Tokenizer-base
- Runs on **ZeroGPU** (A10G, dynamically allocated per request)
- Typical inference: 5-15 seconds for 3-10 tickers × 30 samples
- First request cold-start: ~15s (model load onto GPU)
- Free HF account daily quota: ~5-10 GPU-minutes

## Disclaimer

Not financial advice. The returned probabilities are **model-subjective** — the internal distribution of Kronos's sampled paths — not real-world probabilities.