From 1e4a8382db5710d9c528575d6918eeaeedd1fb19 Mon Sep 17 00:00:00 2001
From: Alvaro Bartolome <36760800+alvarobartt@users.noreply.github.com>
Date: Fri, 12 Sep 2025 11:26:57 +0200
Subject: [PATCH] Fix `cargo install` commands for `candle` with CUDA

The `cargo install --path router ...` commands for the features
`candle-cuda` and `candle-cuda-turing` were using the flag
`--no-default-features` that was ignoring the default features in the
`text-embeddings-router` being `candle,http,dynamic-linking`, hence the
command was failing as `cudarc` requires to have any linking out of
`static-linking`, `dynamic-linking` or `dynamic-loading` for the CUDA
Libraries, hence when ignoring the `default` features, and just adding
`-F http` either the e.g. `-F dynamic-loading` feature was missing or
just to only keep the e.g. `-F candle-cuda`, as `dynamic-linking` and
`http` are default features already.
---
 README.md                   | 6 +++---
 docs/source/en/local_gpu.md | 6 +++---
 2 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/README.md b/README.md
index f2508852..9231a32e 100644
--- a/README.md
+++ b/README.md
@@ -530,13 +530,13 @@ export PATH=$PATH:/usr/local/cuda/bin
 Then run:
 
 ```shell
-# This can take a while as we need to compile a lot of cuda kernels
+# This can take a while as we need to compile a lot of CUDA kernels
 
 # On Turing GPUs (T4, RTX 2000 series ... )
-cargo install --path router -F candle-cuda-turing -F http --no-default-features
+cargo install --path router -F candle-cuda-turing
 
 # On Ampere and Hopper
-cargo install --path router -F candle-cuda -F http --no-default-features
+cargo install --path router -F candle-cuda
 ```
 
 You can now launch Text Embeddings Inference on GPU with:
diff --git a/docs/source/en/local_gpu.md b/docs/source/en/local_gpu.md
index 7af94df8..2dbe86f4 100644
--- a/docs/source/en/local_gpu.md
+++ b/docs/source/en/local_gpu.md
@@ -39,18 +39,18 @@ curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
 
 ## Step 3: Install necessary packages
 
-This step  can take a while as we need to compile a lot of cuda kernels.
+This step  can take a while as we need to compile a lot of CUDA kernels.
 
 ### For Turing GPUs (T4, RTX 2000 series ... )
 
 ```shell
-cargo install --path router -F candle-cuda-turing -F http --no-default-features
+cargo install --path router -F candle-cuda-turing
 ```
 
 ### For Ampere and Hopper
 
 ```shell
-cargo install --path router -F candle-cuda -F http --no-default-features
+cargo install --path router -F candle-cuda
 ```
 
 ## Step 4: Launch Text Embeddings Inference