Fix transformer offline for CI/CD llama4 tests

.github/workflows/cicd-main-nemo2.yml

@@ -237,9 +237,9 @@ jobs:
           - runner: self-hosted-azure
             script: L2_NeMo_2_LLAMA4_MOCK_FINETUNE_PP2
           - runner: self-hosted-azure
-            script: L2_NeMo_2_LLAMA4_MOCK_FINETUNE_CP2_EP2
+            script: L2_NeMo_2_LLAMA4_MOCK_FINETUNE_CP2
           - runner: self-hosted-azure
-            script: L2_NeMo_2_LLAMA4_ENERGON_FINETUNE_TP2_EP2
+            script: L2_NeMo_2_LLAMA4_ENERGON_FINETUNE_EP2
     needs: [build]
     runs-on: ${{ matrix.runner }}
     name: ${{ matrix.script }}

scripts/vlm/llama4/llama4_finetune.py

@@ -45,6 +45,7 @@ def main(args):
     mbs = args.mbs
     max_steps = args.max_steps
     num_workers = args.num_workers
+    val_check_interval = 500
 
     decoder_seq_length = args.decoder_seq_length
 
@@ -53,6 +54,7 @@ def main(args):
     llama4_config = vlm.Llama4ScoutExperts16Config()
     if args.use_toy_model:
         decoder_seq_length = 4096
+        val_check_interval = 50
         llama4_config.vision_transformer_config.num_layers = 2
         llama4_config.language_transformer_config.num_layers = 2
         llama4_config.language_transformer_config.num_moe_experts = 2
@@ -139,7 +141,7 @@ def main(args):
             MegatronCommOverlapCallback(tp_comm_overlap=False),
             NsysCallback(start_step=10, end_step=12, ranks=[0], gen_shape=True),
         ],
-        val_check_interval=500,
+        val_check_interval=val_check_interval,
         limit_val_batches=gbs,
         log_every_n_steps=1,
         num_sanity_val_steps=0,

tests/functional_tests/L2_NeMo_2_LLAMA4_ENERGON_FINETUNE_TP2_EP2.sh → tests/functional_tests/L2_NeMo_2_LLAMA4_ENERGON_FINETUNE_EP2.sh

@@ -11,9 +11,9 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo scripts/vlm/llama4/llama4_finetune.py \
+TRANSFORMERS_OFFLINE=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo scripts/vlm/llama4/llama4_finetune.py \
     --devices=2 --data_type=energon --use_toy_model \
     --data_path=/home/TestData/tiny_datasets/llava-energon \
     --mbs=1 --gbs=4 --max_steps=4 \
-    --tp=2 --ep=2 \
+    --ep=2 \
     --log_dir=/tmp/nemo2_neva_results/$RUN_ID

tests/functional_tests/L2_NeMo_2_LLAMA4_MOCK_FINETUNE_CP2_EP2.sh → tests/functional_tests/L2_NeMo_2_LLAMA4_MOCK_FINETUNE_CP2.sh

@@ -11,8 +11,8 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo scripts/vlm/llama4/llama4_finetune.py \
+TRANSFORMERS_OFFLINE=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo scripts/vlm/llama4/llama4_finetune.py \
     --devices=2 --data_type=mock --use_toy_model \
     --mbs=2 --gbs=4 --max_steps=4 \
-    --cp=2 --ep=2 \
+    --cp=2 \
     --log_dir=/tmp/nemo2_neva_results/$RUN_ID

tests/functional_tests/L2_NeMo_2_LLAMA4_MOCK_FINETUNE_PP2.sh

@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo scripts/vlm/llama4/llama4_finetune.py \
+TRANSFORMERS_OFFLINE=1 coverage run -a --data-file=/workspace/.coverage --source=/workspace/nemo scripts/vlm/llama4/llama4_finetune.py \
     --devices=2 --data_type=mock --use_toy_model \
     --mbs=2 --gbs=4 --max_steps=4 \
     --pp=2 \