@@ -175,7 +175,7 @@ items:
         moe_dispatcher: hybridep
       command: >
         ./scripts/training/train.sh --nodes 2 --gpus-per-node 8
-        --recipe qwen35_vl_35b_a3b_pretrain_16gpu_h100_bf16_functional_config
+        --recipe qwen35_vl_35b_a3b_pretrain_config
         --dataset energon
         --pretrained_checkpoint
         work/model-verification/qwen3.6-35b-a3b/imported-megatron/iter_0000000
@@ -199,33 +199,15 @@ items:
         last_10_steps_model_tflops_per_gpu_avg: 135.610
         last_10_steps_tokens_per_second_per_gpu_avg: 5885.264
       expected_result: >
-        The historical compatibility recipe name and logical output path
-        preserve the identity of the 16-H100 workload that completed all 50
-        DataComp Energon steps at
-        TP1/PP2/CP1/EP8/ETP1, a 16/24 language-layer pipeline split, and
-        MBS1/GBS512. The reusable recipe left the pretrained checkpoint unset;
-        the command warm-started all trainable weights from the immutable
-        imported Qwen3.6 checkpoint for representative natural router traffic.
-        The persisted config confirmed the 248,320-token model vocabulary,
-        HybridEP, selective core-attention, GDN output-norm, and MoE-activation
-        recompute, scoped Transformer Engine CUDA graphs for language
-        attention/router/preprocessing, a graph-free variable-shape vision
-        encoder, and Transformer Engine fused cross entropy. Loss remained
-        finite from 3.604611 to 3.458849 with zero skipped or NaN iterations.
-        Steps 41-50 averaged 22,271.220 ms and 135.610 model TFLOP/s/GPU,
-        approximately 136, and the process exited successfully without
-        checkpoint output. The public command omits the historical invocation's
-        redundant `checkpoint.save=null`; the persisted recipe configuration
-        already left checkpoint saving unset. This verifies bounded full-model
-        DataComp training, not canonical DataComp/CLIP convergence or checkpoint
-        resume. The data
-        was prepared from the official DataComp downloader at commit
-        4a8df1992566, DataComp-1B metadata revision 086ebeee20d4, and
-        img2dataset 1.40.0. The converter selected the first 525,000 valid,
-        UID-deduplicated samples from sorted shards and produced a deterministic
-        519,827/5,173 train/validation split with preparation manifest
-        6e273a96a756; this 50-step run consumed 25,600 training samples without
-        wrapping.
+        The 16-H100 command completed all 50 full-model DataComp Energon steps
+        at TP1/PP2/CP1/EP8/ETP1 and MBS1/GBS512 after warm-starting the
+        immutable imported checkpoint for natural routing. Loss remained finite
+        from 3.604611 to 3.458849 with zero skipped or NaN iterations. The
+        519,827/5,173 train/validation split was prepared with the official
+        DataComp downloader at commit 4a8df1992566, DataComp-1B metadata
+        revision 086ebeee20d4, img2dataset 1.40.0, and preparation manifest
+        6e273a96a756. This verifies bounded full-model training, not canonical
+        DataComp/CLIP convergence or checkpoint resume.
 
   sft:
     H100:
 
@@ -36,7 +36,6 @@
     "qwen35_vl_2b_sft_1gpu_h100_bf16_config",
     "qwen35_vl_35b_a3b_peft_16gpu_h100_bf16_config",
     "qwen35_vl_35b_a3b_peft_4gpu_h100_bf16_config",
-    "qwen35_vl_35b_a3b_pretrain_16gpu_h100_bf16_functional_config",
     "qwen35_vl_35b_a3b_pretrain_config",
     "qwen35_vl_35b_a3b_pretrain_8gpu_h100_bf16_mock_config",
     "qwen35_vl_35b_a3b_sft_2gpu_h100_bf16_fsdp_config",
 
@@ -329,11 +329,6 @@ def qwen35_vl_35b_a3b_pretrain_config() -> ConfigContainer:
     return cfg
 
 
-def qwen35_vl_35b_a3b_pretrain_16gpu_h100_bf16_functional_config() -> ConfigContainer:
-    """Return the Qwen3.5/Qwen3.6-VL 35B-A3B pretrain config under its legacy name."""
-    return qwen35_vl_35b_a3b_pretrain_config()
-
-
 def qwen35_vl_122b_a10b_pretrain_128gpu_h100_bf16_mock_config() -> ConfigContainer:
     """Return a pre-training config for Qwen3.5-VL 122B-A10B (MoE)."""
     cfg = _pretrain_common()
@@ -2399,7 +2394,6 @@ def qwen35_vl_397b_a17b_peft_32gpu_h100_bf16_config(
     "qwen35_vl_35b_a3b_sft_2gpu_h100_bf16_fsdp_config",
     "qwen35_vl_35b_a3b_peft_16gpu_h100_bf16_config",
     "qwen35_vl_35b_a3b_peft_4gpu_h100_bf16_config",
-    "qwen35_vl_35b_a3b_pretrain_16gpu_h100_bf16_functional_config",
     "qwen35_vl_35b_a3b_pretrain_config",
     "qwen35_vl_35b_a3b_pretrain_8gpu_h100_bf16_mock_config",
     "qwen35_vl_35b_a3b_sft_16gpu_h100_bf16_config",
 
@@ -464,16 +464,13 @@ def test_qwen35_vl_35b_a3b_pretrain_16gpu_h100_defaults(monkeypatch: pytest.Monk
     assert cfg.mixed_precision.grad_reduce_in_fp32 is False
 
 
-def test_qwen35_vl_35b_a3b_legacy_pretrain_name_remains_compatible(monkeypatch: pytest.MonkeyPatch):
-    """The historical full-pretrain name should delegate to the maintained library recipe."""
-    sentinel = object()
-    monkeypatch.setattr(_qwen35_vl_h100_module, "qwen35_vl_35b_a3b_pretrain_config", lambda: sentinel)
-
-    legacy_name = "qwen35_vl_35b_a3b_pretrain_16gpu_h100_bf16_functional_config"
+def test_qwen35_vl_35b_a3b_pretrain_name_is_exported():
+    """The canonical full-pretrain recipe should be available from the public recipes package."""
+    recipe_name = "qwen35_vl_35b_a3b_pretrain_config"
     recipes_package = importlib.import_module("megatron.bridge.recipes")
-    assert legacy_name in _qwen35_vl_h100_module.__all__
-    assert getattr(recipes_package, legacy_name) is getattr(_qwen35_vl_h100_module, legacy_name)
-    assert getattr(_qwen35_vl_h100_module, legacy_name)() is sentinel
+
+    assert recipe_name in _qwen35_vl_h100_module.__all__
+    assert getattr(recipes_package, recipe_name) is getattr(_qwen35_vl_h100_module, recipe_name)
 
 
 def test_qwen35_vl_35b_a3b_sft_defaults(monkeypatch: pytest.MonkeyPatch):