Remove redundant test

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

Author: jstjohn

sub-packages/bionemo-evo2/tests/bionemo/evo2/run/test_train.py

@@ -17,13 +17,11 @@
 # limitations under the License.
 import argparse
 import io
-import os
 import shlex
 from contextlib import redirect_stderr, redirect_stdout
 from typing import Tuple
 
 import pytest
-from lightning.fabric.plugins.environments.lightning import find_free_network_port
 from nemo import lightning as nl
 from transformer_engine.pytorch.fp8 import check_fp8_support
 
@@ -152,98 +150,6 @@ def test_train_evo2_stops(tmp_path):
     assert "train_step_timing in s" in trainer.logged_metrics
 
 
-@pytest.mark.slow
-@pytest.mark.parametrize("model_size", ["7b_nv", "7b_arc_longcontext"])
-def test_train_single_gpu(tmp_path, model_size: str):
-    """
-    This test runs them single gpu evo2 training command with sample data in a temporary directory.
-    """
-    num_steps = 7
-    open_port = find_free_network_port()
-    # a local copy of the environment
-    env = dict(**os.environ)
-    env["MASTER_PORT"] = str(open_port)
-    # Part 1: Make sure training runs for only --early-stop-on-step steps
-    additional_args1 = [
-        "--result-dir",
-        str(tmp_path),
-        "--model-size",
-        model_size,
-        "--num-layers",
-        str(4),
-        "--hybrid-override-pattern",
-        "SDH*",
-        "--no-activation-checkpointing",
-        "--use-precision-aware-optimizer",
-        "--add-bias-output",
-        "--bf16-main-grads",
-        "--val-check-interval",
-        str(5),
-        "--max-steps",
-        str(num_steps),
-        "--early-stop-on-step",
-        str(num_steps - 2),
-        "--warmup-steps",
-        str(1),
-        "--seq-length",
-        str(128),
-        "--wandb-offline",
-        "--wandb-anonymous",
-        "--mock-data",
-    ]
-    args1 = parse_args(args=additional_args1)
-    stdout_buf, stderr_buf = io.StringIO(), io.StringIO()
-    with redirect_stdout(stdout_buf), redirect_stderr(stderr_buf), distributed_model_parallel_state():
-        train(args=args1)
-    train_stdout = stdout_buf.getvalue()
-    train_lines = train_stdout.split("\n")
-    iteration_lines = [line for line in train_lines if "Training epoch" in line]
-    assert len(iteration_lines) == 5
-    iteration_line_1 = iteration_lines[0]
-    # No strong opinion on how the total should be computed in the case of early stopping. We allow either for now
-    #  unless there is an issue, such as with the LR scheduler...
-    # TODO: Add a test somewhere that covers that early stopping callback has no impact on the LR scheduler
-    assert "iteration 0/4" in iteration_line_1 or "iteration 0/6" in iteration_line_1
-    iteration_line_final = iteration_lines[-1]
-    assert "iteration 4/4" in iteration_line_final or "iteration 4/6" in iteration_line_final
-
-    # Part 2: Make sure training picks up where it left off
-    additional_args2 = [
-        "--result-dir",
-        str(tmp_path),
-        "--model-size",
-        model_size,
-        "--num-layers",
-        str(4),
-        "--hybrid-override-pattern",
-        "SDH*",
-        "--no-activation-checkpointing",
-        "--use-precision-aware-optimizer",
-        "--add-bias-output",
-        "--max-steps",
-        str(num_steps),
-        "--warmup-steps",
-        str(1),
-        "--seq-length",
-        str(128),
-        "--wandb-offline",
-        "--wandb-anonymous",
-        "--mock-data",
-    ]
-    args2 = parse_args(args=additional_args2)
-    stdout_buf, stderr_buf = io.StringIO(), io.StringIO()
-    with redirect_stdout(stdout_buf), redirect_stderr(stderr_buf), distributed_model_parallel_state():
-        train(args=args2)
-    train_stdout = stdout_buf.getvalue()
-    train_lines = train_stdout.split("\n")
-    iteration_lines = [line for line in train_lines if "Training epoch" in line]
-    assert len(iteration_lines) == 2
-    iteration_line_1 = iteration_lines[0]
-    assert "iteration 5/6" in iteration_line_1
-    iteration_line_2 = iteration_lines[1]
-    assert "iteration 6/6" in iteration_line_2
-
-
 @pytest.mark.parametrize(
     "additional_args",
     [