Refresh model card (drop TempoPFN-Tiny naming; expand details)

README.md

@@ -13,22 +13,25 @@ pipeline_tag: time-series-forecasting
 # MacroCast
 
 **MacroCast** is a compact (~1.2M-parameter) time-series foundation model for
-macroeconomic forecasting. It is produced in two stages:
+**macroeconomic forecasting**. It forecasts a full panel of monthly
+indicators in a **single forward pass** (no autoregressive windowing) and returns
+**probabilistic** (9-quantile) forecasts.
 
-1. **Forge** — a TempoPFN-Tiny (GatedDeltaProduct linear-RNN) backbone pre-trained
-   **entirely on synthetic data** (no real macro data).
-2. **MacroCast** — Forge **fine-tuned** on FRED-MD via an expanding-window,
-   real-time-vintage procedure, using a mixture of macro-calibrated synthetic
-   panels (block-bootstrap, AR, Dynamic Factor Model, and Minnesota BVAR).
+It is trained in two stages:
 
-It is **non-MoE** (a single full-fine-tuned backbone) and forecasts multivariate
-panels in a single forward pass (no windowing).
+1. Base model: A ~1.2M-parameter linear-RNN backbone (architecture inspired by
+   TempoPFN), trained from scratch **entirely on synthetic data** (no real macro data).
+2. **MacroCast** — Base model **fine-tuned** on FRED-MD via an expanding-window,
+   real-time-vintage procedure, using a mixture of macro-calibrated synthetic
+   panels (block-bootstrap, per-variable AR, Dynamic Factor Model, and BVAR).
 
-## Vintages
-This repo ships **one checkpoint per fine-tuning vintage** (`models/`). Each is
-MacroCast fine-tuned on the FRED-MD data available *through that cutoff*, so you
-can choose how much history the model has seen. Available vintages and their
-variable counts are listed in `manifest.json`.
+## Vintages (real-time fine-tuning)
+This repo ships **one checkpoint per fine-tuning vintage** (in `models/`). Each
+checkpoint is MacroCast fine-tuned on only the FRED-MD data available *through that
+cutoff* — e.g. the 2010 vintage has never seen post-2010 data — so you can choose how
+much history the model has seen and run **leak-free real-time / backtesting**
+experiments. The available vintages, their data cutoffs, and per-vintage variable
+counts are listed in `manifest.json` (`default_year` is used when no year is given).
 
 ## Usage
 
@@ -37,22 +40,34 @@ from huggingface_hub import snapshot_download
 import sys
 
 local = snapshot_download(repo_id="shubhranshu/MacroCast")
-sys.path.insert(0, local)                 # for forecast.py + tempopfn/
+sys.path.insert(0, local)                 # for forecast.py
 from forecast import MacroCastForecaster
 
-# Pick MacroCast + a vintage year (default = latest available):
+# List vintages, then load MacroCast for one (year=None -> latest):
+print(MacroCastForecaster.available_years(local))     # e.g. [1998, ..., 2023]
 f = MacroCastForecaster.from_pretrained(local, year=2023)
 
-point, quantiles = f.predict(history, horizon=12, freq="M")  # history: [T, N]
-print(MacroCastForecaster.available_years(local))            # list all vintages
+# history: np.ndarray of shape [T, N]  (T monthly steps, N FRED-MD variables)
+point, quantiles = f.predict(history, horizon=12, freq="M")
 ```
 
-- `point`: `[horizon, N]` median forecast.
+- `point`: `[horizon, N]` median (0.5-quantile) forecast.
 - `quantiles`: `[horizon, N, 9]` forecasts for quantiles 0.1 … 0.9.
 
-## Details
-- Architecture: GatedDeltaProduct linear-RNN, `embed=128`, 3 layers, 9-quantile pinball loss.
-- Fine-tuned on monthly, FRED-MD-transformed (stationary) series.
-- Each checkpoint stores the exact variable list it was trained on.
 
-Built on [TempoPFN](https://huggingface.co/AutoML-org/TempoPFN) (Apache-2.0).
+## Model details
+- **Architecture**: GatedDeltaProduct linear-RNN, `embed=128`, 3 layers, ~1.2M params.
+- **Objective**: 9-quantile (0.1…0.9) pinball loss.
+- **Pre-training**: synthetic priors only (GP, kernel, Ornstein–Uhlenbeck,
+  ForecastPFN, CauKer, sine/sawtooth/step, …) — no real macro data.
+- **MacroCast fine-tuning**: FRED-MD real-time vintages expanded into a synthetic
+  mixture — `real` + `block-bootstraps`, `arima` (per-variable AR), `dfm`
+  (Dynamic Factor Model), and `bvar_clust` / `bvar_full` (Minnesota BVAR).
+
+## Intended use and limitations
+- Built for **monthly macroeconomic** panels. Feed transformed/stationary inputs.
+- Forecasts are statistical outputs, **not financial or policy advice**.
+
+## Acknowledgements
+Architecture inspired by [TempoPFN](https://huggingface.co/AutoML-org/TempoPFN)
+(Apache-2.0); the vendored model code under `tempopfn/` derives from that project.