Remove NCCL_NVLS_ENABLE setting from JAX image (#1708)

Removal of
[NCCL_NVLS_ENABLE](https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/env.html#nccl-nvls-enable)=0
setting enables NVLink SHARP on systems which support it by default.
Change (to be) added for 25.10 NGC release.



- [x] Performance regression tests on internal clusters
- [ ] Performance regression tests on GKE, EKS (due to @nouiz)

Author: aybchan

.github/container/Dockerfile.jax

@@ -97,7 +97,6 @@ ENV BUILD_DATE=${BUILD_DATE}
 # The following environment variables tune performance
 ENV XLA_FLAGS=""
 ENV XLA_FLAGS="${XLA_FLAGS} --xla_gpu_enable_latency_hiding_scheduler=true"
-ENV NCCL_NVLS_ENABLE=0
 
 COPY --from=builder ${BUILD_PATH_JAXLIB} ${BUILD_PATH_JAXLIB}
 COPY --from=builder ${SRC_PATH_JAX} ${SRC_PATH_JAX}

README.md

@@ -218,10 +218,6 @@ The [JAX image](https://github.com/NVIDIA/JAX-Toolbox/pkgs/container/jax) is emb
 | --------- | ----- | ----------- |
 | `--xla_gpu_enable_latency_hiding_scheduler` | `true`  | allows XLA to move communication collectives to increase overlap with compute kernels |
 
-| Environment Variable | Value | Explanation |
-| -------------------- | ----- | ----------- |
-| `NCCL_NVLS_ENABLE` | `0` | Disables NVLink SHARP ([1](https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/env.html#nccl-nvls-enable)). Future releases will re-enable this feature. |
-
 There are various other XLA flags users can set to improve performance. For a detailed explanation of these flags, please refer to the [GPU performance](./rosetta/docs/GPU_performance.md) doc. XLA flags can also be tuned per workload. For example, each script includes a directory [xla_flags](./rosetta/rosetta/projects/maxtext/xla_flags).
 
 For a list of previously used XLA flags that are no longer needed, please also refer to the [GPU performance](./rosetta/docs/GPU_performance.md#previously-used-xla-flags) page.