Bump megatron bridge dep

Signed-off-by: John St. John <jstjohn@nvidia.com>

Author: jstjohn

bionemo-recipes/recipes/evo2_megatron/.ci_build.sh

@@ -5,7 +5,7 @@
 rm -f /usr/local/lib/python*/dist-packages/transformer_engine-*.dist-info/direct_url.json
 export UV_LOCK_TIMEOUT=900  # increase to 15 minutes (900 seconds), adjust as needed
 export UV_LINK_MODE=copy
-uv venv --system-site-packages
+uv venv --clear --system-site-packages
 
 # 2. Activate the environment
 source .venv/bin/activate

bionemo-recipes/recipes/evo2_megatron/build_requirements.txt

@@ -1,3 +1,5 @@
 poetry-core
+poetry_dynamic_versioning  # build dep of nvidia-resiliency-ext (transitively pulled by megatron-bridge); needed in the venv because we install with --no-build-isolation
+grpcio-tools  # build dep of nvidia-resiliency-ext: its setup.py shells out to `python -m grpc_tools.protoc` to compile *.proto files; --no-build-isolation means we have to provide it in the venv up-front
 wheel_stub
 ninja  # should speed up causal-conv1d build

bionemo-recipes/recipes/evo2_megatron/pyproject.toml

@@ -24,7 +24,7 @@ dependencies = [
     "causal_conv1d",
     "nv-grouped-gemm",
     "megatron-core",
-    "nvidia-resiliency-ext",
+    # nvidia-resiliency-ext is pulled transitively by megatron-bridge.
     "emerging_optimizers",
     "subquadratic-ops-torch-cu13",
 
@@ -88,6 +88,9 @@ override-dependencies = [
     "triton; sys_platform == 'never'",
     "transformer-engine; sys_platform == 'never'",
     "transformer-engine[pytorch]; sys_platform == 'never'",
+    # Avoid optional log-pattern-mining dependency conflicts from nvidia-resiliency-ext.
+    "logsage; sys_platform == 'never'",
+    "drain3; sys_platform == 'never'",
 ]
 
 [tool.uv.sources]
@@ -100,11 +103,12 @@ nv-grouped-gemm = { git = "https://github.com/fanshiqing/grouped_gemm", tag = "v
 # Internal dependencies
 bionemo-recipeutils = { git = "https://github.com/NVIDIA/bionemo-framework.git", branch = "main", subdirectory = "sub-packages/bionemo-recipeutils" }
 bionemo-core = { git = "https://github.com/NVIDIA/bionemo-framework.git", branch = "main", subdirectory = "sub-packages/bionemo-core" }
-nvidia-resiliency-ext = { git = "https://github.com/NVIDIA/nvidia-resiliency-ext.git", rev = "54f85fe422d296cf04ea524130014bd3a2c3add1" }
+# nvidia-resiliency-ext is intentionally left to Megatron-Bridge so the transitive pin stays consistent.
 
-# Megatron Bundle. This points to a version that still supports the deprecated no_weight_decay_cond field until the API for an alternative has been finalized.
-megatron-bridge = { git = "https://github.com/NVIDIA-NeMo/Megatron-Bridge.git", rev = "549e3cb970c170b1d7a86d021261efe05e8a5d9f" }
-megatron-core = { git = "https://github.com/NVIDIA-NeMo/Megatron-Bridge.git", rev = "549e3cb970c170b1d7a86d021261efe05e8a5d9f", subdirectory = "3rdparty/Megatron-LM" }
+# Megatron Bundle. MCore is sourced from the same Megatron-Bridge release tag.
+megatron-bridge = { git = "https://github.com/NVIDIA-NeMo/Megatron-Bridge.git", tag = "v0.4.1" }
+megatron-core = { git = "https://github.com/NVIDIA-NeMo/Megatron-Bridge.git", tag = "v0.4.1", subdirectory = "3rdparty/Megatron-LM" }
 
 [tool.uv.extra-build-dependencies]
 warp-lang = ["wheel_stub"]
+nvidia-resiliency-ext = ["poetry_dynamic_versioning"]