Add cupy backend for GPU support (#1048)

Author: s3alfisc

.github/workflows/extended_tests.yaml

@@ -49,7 +49,7 @@ jobs:
 
       - name: Compile Rust extension (no wheel)
         run: |
-          pixi r maturin-develop
+          pixi r -e dev maturin-develop
 
       - name: Run long tests with coverage
         run: pixi run tests-extended

README.md

@@ -86,6 +86,25 @@ python -m pip install pyfixest[plots]
 
 Note that matplotlib is included by default, so you can always use the matplotlib backend for plotting even without installing the optional lets-plot dependency.
 
+### GPU Acceleration (Optional)
+
+PyFixest supports GPU-accelerated fixed effects demeaning via CuPy. To enable GPU acceleration, install CuPy matching your CUDA version:
+
+```bash
+# For CUDA 11.x, 12.x, 13.x
+pip install cupy-cuda11x
+pip install cupy-cuda12x
+pip install cupy-cuda13x
+```
+
+Once installed, you can use GPU-accelerated demeaning by setting the `demean_backend` parameter:
+
+```python
+# Use GPU with float32 and float64 precision
+pf.feols("Y ~ X1 | f1 + f2", data=data, demean_backend="cupy32")
+pf.feols("Y ~ X1 | f1 + f2", data=data, demean_backend="cupy64")
+```
+
 ## Benchmarks
 
 All benchmarks follow the [fixest

benchmarks/complex_benchmarks.png

@@ -12,7 +12,7 @@
     }
    },
    "source": [
-    "# PyFixest on the GPU \n",
+    "# PyFixest on the GPU via JAX\n",
     "\n",
     "Through its JAX integration, it is possible to run PyFixest on the GPU. In this notebook, we benchmark the performance of PyFixest on the GPU via its \n",
     "`jax` backend and compare it to the performance of PyFixest on the CPU (via the default `numba` backend). \n",