cp: feat: add nvidia-resiliency-ext as nvrx optional extra (2264) into r0.6.0

docs/guides/ft-launcher-guide.md

@@ -1,6 +1,6 @@
 # Fault Tolerance Launcher Guide
 
-The `ft_launcher` is provided by `nvidia-resiliency-ext` (included in NeMo RL dependencies) and enables automatic fault tolerance and recovery for distributed training runs.
+The `ft_launcher` is provided by `nvidia-resiliency-ext` (available via the `nvrx` optional extra, e.g. `uv run --extra nvrx ft_launcher ...`) and enables automatic fault tolerance and recovery for distributed training runs.
 
 ## Key Arguments
 
@@ -14,7 +14,7 @@ The `ft_launcher` is provided by `nvidia-resiliency-ext` (included in NeMo RL de
 ## Basic Usage
 
 ```bash
-uv run ft_launcher \
+uv run --extra nvrx ft_launcher \
     --ft-cfg-path examples/ft_launcher/ft_config.yaml \
     --ft-rank-heartbeat-timeout 450 \
     --ft-initial-rank-heartbeat-timeout 1200 \

pyproject.toml

@@ -54,7 +54,6 @@ dependencies = [
   "cuda-bindings",                                                                                                    # for non-colocated refit
   "pybase64",                                                                                                         # for sglang refit
   "nvidia-cudnn-cu12==9.19.0.56",                                                                                     # for transformer-engine no build isolation
-  # nvidia-resiliency-ext removed: no Python 3.13 wheels available (v0.5.0 only has cp310-cp312)
 ]
 
 [project.optional-dependencies]
@@ -109,6 +108,9 @@ mcore = [
   "emerging-optimizers==0.2.0",
   "deep_ep @ git+https://github.com/deepseek-ai/DeepEP.git@bfded34800dfec415b71503f8205181de90b2480",
 ]
+nvrx = [
+  "nvidia-resiliency-ext",
+] # for ft_launcher (fault-tolerant training launcher)
 nemo_gym = ["nemo_gym"]
 
 [dependency-groups]