Fixes of Merge main into dev

Signed-off-by: oliver könig <[email protected]>

Author: ko3n1g

.gitlab/stages/04.functional-tests.yml

@@ -172,15 +172,15 @@ functional:run_dev_dgx_h100:
 functional:run_nemo:
   extends: [.functional_tests_rules]
   trigger:
-    project: "dl/joc/nemo-ci"
+    project: 'dl/joc/nemo-ci'
     branch: main-mirror
     strategy: depend
   inherit:
     variables: true
   variables:
     MCORE_MR_COMMIT: $CI_COMMIT_SHA
-    TEST_NEMO2_MODULE: "True"
-    ALLOW_FAILURE_DEPENDENCY: "True"
+    TEST_NEMO2_MODULE: 'True'
+    ALLOW_FAILURE_DEPENDENCY: 'True'
     TESTS_TO_RUN_ON_THIS_COMMIT: nightly
   rules:
     - if: $FUNCTIONAL_TEST == "yes"
@@ -217,7 +217,7 @@ functional:x_notify:
     - export RO_API_TOKEN=${PROJECT_ACCESS_TOKEN_MCORE}
     - export GITLAB_ENDPOINT
     - export CONTEXT=$FUNCTIONAL_TEST_SCOPE
-    - export TAG_TEAM=$([[ "$CI_COMMIT_BRANCH" == "main" ]] && echo "1" || "0")
+    - export TAG_TEAM=$([[ "$CI_COMMIT_BRANCH" == "main" || "$CI_COMMIT_BRANCH" == "dev" ]] && echo "1" || "0")
     - export TEAM_SLUG=$SLACK_ADMIN
     - |
       python tests/test_utils/python_scripts/notify.py \

examples/inference/gpt/gpt_static_inference.py

@@ -104,7 +104,7 @@ def get_inference_engine(args: Namespace, model: MegatronModule) -> StaticInfere
     text_generation_controller = TextGenerationController(
         inference_wrapped_model=inference_wrapped_model, tokenizer=tokenizer
     )
-    return StaticInferenceEngine(text_generation_controller=text_generation_controller, legacy=args.use_legacy_static_engine)
+    return StaticInferenceEngine(text_generation_controller=text_generation_controller, legacy=True)
 
 
 async def generate(

pyproject.toml

@@ -69,7 +69,7 @@ mlm = ["flask-restful", "sentencepiece", "tiktoken", "wandb", "transformers"]
 dev = [
     "nvidia-modelopt[torch]>=0.33.0a0,<0.34.0; sys_platform != 'darwin'",
     "transformer-engine[pytorch]>=2.9.0a0,<2.10.0",
-    "nvidia-resiliency-ext",
+    "nvidia-resiliency-ext>=0.4.0a0,<0.5.0",
     "tqdm",
     "einops~=0.8",
     "tensorstore~=0.1,!=0.1.46,!=0.1.72",
@@ -91,20 +91,13 @@ dev = [
 
 lts = [
     "tqdm",
-    "einops~=0.8",
-    "tensorstore~=0.1,!=0.1.46,!=0.1.72",
-    "nvtx~=0.2",
-    "multi-storage-client~=0.27",
-    "opentelemetry-api~=1.33.1",
+    "einops",
+    "tensorstore!=0.1.46,!=0.1.72",
+    "nvtx",
+    "transformers",
+    "zarr",
     "setuptools<80.0.0",
-    "mamba-ssm~=2.2",
-    "causal-conv1d~=1.5",
-    "nv-grouped-gemm~=1.1",
-    "megatron-energon[av_decode]~=6.0",
-    "av<16.0.0",                          # At the time, av 16.0.0 is not compatible with Python 3.12
-    "flashinfer-python",
     "wget",
-    "onnxscript",
 ]
 
 [dependency-groups]

tests/unit_tests/inference/engines/test_dynamic_engine.py

@@ -102,7 +102,7 @@ class DynamicEngineTestConfig:
     model_provider: str = "gpt"
     return_log_probs: bool = False
     materialize_only_last_token_logits: bool = True
-    skip_prompt_log_probs_for_dynamic_inference: bool = False
+    skip_prompt_log_probs: bool = False
     cuda_graph_scope: List[str] = None
     force_build_cuda_graphs: bool = False
     # If False, do not build cuda graphs in the tests, even if