Merge branch 'ko3n1g/ci/fix-notifications-2' into 'main'

ci: Fix notifications

See merge request ADLR/megatron-lm!2265

Author: ko3n1g

tests/functional_tests/jet_recipes/gpt.yaml

@@ -55,7 +55,6 @@ products:
     - gpt3_mr_mcore_te_tp1_pp4_vp1_decoupled_lr_dgx_a100_1N8G
     - gpt3_mr_mcore_te_tp1_pp4_vp1_uneven_pipeline_dgx_a100_1N8G
     - gpt3_mr_mcore_te_tp1_pp4_vp1_dgx_a100_1N8G
-    - gpt3_mr_mcore_te_tp1_pp4_vp1_dist_optimizer_overlap_grad_reduce_dgx_a100_1N8G
     - gpt3_mr_mcore_te_tp1_pp4_vp1_dist_optimizer_overlap_grad_reduce_param_gather_overlap_optimizer_dgx_a100_1N8G
     - gpt3_mr_mcore_te_tp1_pp4_vp1_dist_optimizer_overlap_grad_reduce_untied_dgx_a100_1N8G
     - gpt3_mr_mcore_te_tp1_pp4_vp1_resume_torch_decoupled_lr_dgx_a100_1N8G
@@ -73,10 +72,9 @@ products:
     - gpt3_mr_mcore_te_tp2_pp1_te_8experts2parallel_dist_optimizer_dgx_a100_1N8G
     - gpt3_mr_mcore_te_tp2_pp1_te_8experts2parallel_groupedGEMM_dgx_a100_1N8G
     - gpt3_mr_mcore_te_tp2_pp1_te_8experts2parallel_overlap_grad_reduce_param_gather_groupedGEMM_dgx_a100_1N8G
-    - gpt3_mr_mcore_te_tp2_pp1_te_8experts2parallel_top2router_dgx_a100_1N8G
+    # - gpt3_mr_mcore_te_tp2_pp1_te_8experts2parallel_top2router_dgx_a100_1N8G
     - gpt3_mr_mcore_te_tp2_pp2_cross_entropy_loss_fusion_dgx_a100_1N8G
     - gpt3_mr_mcore_te_tp2_pp2_ddp_average_in_collective_dgx_a100_1N8G
-    - gpt3_mr_mcore_te_tp2_pp2_defer_embedding_wgrad_compute_dgx_a100_1N8G
     - gpt3_mr_mcore_te_tp2_pp2_dgx_a100_1N8G
     - gpt3_mr_mcore_te_tp2_pp2_no_create_attention_mask_in_dataloader_dgx_a100_1N8G
     - gpt3_mr_mcore_te_tp2_pp2_no_mmap_bin_files_dgx_a100_1N8G
@@ -139,11 +137,13 @@ products:
     platforms: [dgx_a100]
     time_limit: [1800]
     test_case:
+    - gpt3_mr_mcore_te_tp2_pp2_defer_embedding_wgrad_compute_dgx_a100_1N8G
     - gpt3_mr_mcore_te_tp1_pp4_vp1_dist_optimizer_overlap_grad_reduce_param_gather_dgx_a100_1N8G
     - gpt3_mr_mcore_te_tp2_pp1_cp2_nondeterministic_dgx_a100_1N8G
     - gpt3_mr_mcore_te_tp2_pp1_resume_torch_dist_cp2_nondeterministic_dgx_a100_1N8G
     - gpt3_mr_mcore_te_tp2_pp2_cp2_nondeterministic_dgx_a100_1N8G
     - gpt3_mr_mcore_te_tp2_pp2_resume_torch_dist_cp2_nondeterministic_dgx_a100_1N8G
+    - gpt3_mr_mcore_te_tp1_pp4_vp1_dist_optimizer_overlap_grad_reduce_dgx_a100_1N8G
   - environment: [lts]
     scope: [weekly]
     platforms: [dgx_h100]

tests/functional_tests/jet_recipes/t5.yaml

@@ -36,14 +36,14 @@ products:
     - t5_220m_mr_mcore_te_tp4_pp1_dgx_a100_1N8G
     - t5_220m_mr_mcore_te_tp4_pp1_resume_torch_dist_dgx_a100_1N8G
     - t5_220m_mr_mcore_te_tp2_pp2_resume_torch_dgx_a100_1N8G
-    - t5_220m_mr_mcore_tp4_pp1_dgx_a100_1N8G
     - t5_220m_mr_mcore_tp4_pp1_resume_torch_dist_dgx_a100_1N8G
     - t5_220m_mr_mcore_tp2_pp2_resume_torch_dgx_a100_1N8G
-    - t5_220m_mr_mcore_tp2_pp2_dgx_a100_1N8G
   - environment: [lts]
     scope: [mr]
     time_limit: [1800]
     test_case:
+    - t5_220m_mr_mcore_tp4_pp1_dgx_a100_1N8G
+    - t5_220m_mr_mcore_tp2_pp2_dgx_a100_1N8G
     - t5_220m_mr_mcore_te_tp2_pp2_dgx_a100_1N8G
   - environment: [lts]
     scope: [weekly]

tests/functional_tests/shell_test_utils/notify.sh

@@ -48,7 +48,7 @@ if [[ ${ret_code:-0} -ne 0 ]]; then
 fi
 
 # Fetch GitLab logs of JET downstream pipeline
-DOWNSTREAM_PIPELINE_ID=$(jq --arg environment "$ENVIRONMENT" '.[] |select(.name == "jet-trigger-" + $environment) | .downstream_pipeline.id' <<< "$PIPELINE_JSON")
+DOWNSTREAM_PIPELINE_ID=$(jq --arg environment "$ENVIRONMENT" '.[] |select(.name == "functional:run_" + $environment) | .downstream_pipeline.id' <<< "$PIPELINE_JSON")
 
 PIPELINE_URL=https://${GITLAB_ENDPOINT}/ADLR/megatron-lm/-/pipelines/$CI_PIPELINE_ID
 JOB_URL=https://${GITLAB_ENDPOINT}/ADLR/megatron-lm/-/jobs/

tests/functional_tests/shell_test_utils/notify_unit_tests.sh

@@ -48,7 +48,7 @@ if [[ ${ret_code:-0} -ne 0 ]]; then
     exit 1
 fi
 
-UNIT_TESTS_JOBS=$(echo -E $PIPELINE_JSON | jq '[.[] | select(.name | startswith("test:unit_tests_"))]')
+UNIT_TESTS_JOBS=$(echo -E $PIPELINE_JSON | jq '[.[] | select(.name | startswith("test:pyt"))]')
 
 if [[ $UNIT_TESTS_JOBS == null ]]; then
     FAILED_JOBS=$(curl \