diff --git a/README.md b/README.md
index 8c87309..c84a1d9 100644
--- a/README.md
+++ b/README.md
@@ -8,11 +8,11 @@ This repository is the HIP port of the original `gsplat` project, optimized for
 
 To use gsplat, you need the following prerequisites:
 
-- **ROCm**: version 6.4.1 or later (see `ROCm compatibility <https://rocm.docs.amd.com/en/latest/compatibility/compatibility-matrix.html>`__)
-- **Operating system**: Ubuntu 22.04 or later  
+- **ROCm**: version 6.4.3 (recommended)
+- **Operating system**: Ubuntu 24.04  
 - **GPU platform**: AMD Instinct™ MI300X  
-- **PyTorch**: version 2.6 or later (ROCm-enabled)  
-- **Python**: version 3.10 or later  
+- **PyTorch**: version 2.6 (ROCm-enabled)  
+- **Python**: version 3.12  
 
 ## Installation
 
@@ -32,7 +32,7 @@ To use gsplat, you need the following prerequisites:
 3. Install gsplat from the AMD-hosted PyPI repository:
 
    ```bash
-   pip install gsplat --extra-index-url=https://pypi.amd.com/simple
+   pip install gsplat --index-url=https://pypi.amd.com/simple
    ```
 
 4. Verify the installation:
@@ -126,4 +126,4 @@ If you use this library in your research, please cite:
   pages={1--17},
   year={2025}
 }
-```
\ No newline at end of file
+```
diff --git a/docs/install/gsplat-install.rst b/docs/install/gsplat-install.rst
index 9e494be..dacdf86 100644
--- a/docs/install/gsplat-install.rst
+++ b/docs/install/gsplat-install.rst
@@ -54,7 +54,7 @@ Docker is the recommended method to set up your environment, as it avoids potent
 
    .. code-block:: bash
 
-      pip install gsplat --extra-index-url=https://pypi.amd.com/simple
+      pip install gsplat --index-url=https://pypi.amd.com/simple
 
 4. Verify the installation:
 
@@ -199,4 +199,4 @@ Known issues and limitations
 
 - Compression of ``splat parameters`` (``positions``, ``scales``, ``rotations``, ``colors``, and ``features``) using PNG image encoding is not supported.  
 - Usage of ``Fused SSIM`` and ``Fused Bilagrid`` in the simple trainer is not supported (torch-based implementations are used instead).  
-- Distributed training with the ``packed`` option is not supported. 
\ No newline at end of file
+- Distributed training with the ``packed`` option is not supported. 
diff --git a/docs/reference/benchmark-evaluation.rst b/docs/reference/benchmark-evaluation.rst
index 9a23941..588ce3f 100644
--- a/docs/reference/benchmark-evaluation.rst
+++ b/docs/reference/benchmark-evaluation.rst
@@ -81,7 +81,7 @@ PSNR (dB):
 | gsplat-7k   | 23.69   | 30.15  | 27.57   | 26.59  | 29.42   | 29.97  | 25.90  |
 +-------------+---------+--------+---------+--------+---------+--------+--------+
 | gsplat-30k  | 24.93   | 32.26  | 29.19   | 27.63  | 31.54   | 31.75  | 26.78  |
-+-------------+---------+--------+---------+--------+---------+--------+--------
++-------------+---------+--------+---------+--------+---------+--------+--------+
 
 SSIM:
 
@@ -117,4 +117,4 @@ Summary
 ====================================================================
 
 Gsplat-7k trains faster with less GPU memory, while gsplat-30k achieves higher ``PSNR`` and ``SSIM`` metrics but requires more memory and time.  
-This evaluation demonstrates the trade-offs between model size, training efficiency, and reconstruction quality on the Mip-NeRF 360 dataset.
\ No newline at end of file
+This evaluation demonstrates the trade-offs between model size, training efficiency, and reconstruction quality on the Mip-NeRF 360 dataset.
diff --git a/gsplat/cuda/csrc/third_party/glm b/gsplat/cuda/csrc/third_party/glm
new file mode 160000
index 0000000..e771488
--- /dev/null
+++ b/gsplat/cuda/csrc/third_party/glm
@@ -0,0 +1 @@
+Subproject commit e7714885929f1dcd2bdc2531cc99d3c1dc6daa18