Fix lint: revert torchrun shutil.which, remove unused imports, ruff format

The torchrun-not-found issue was caused by the venv not being
activated, not an installation problem. Revert to plain 'torchrun'
command. Remove now-unused shutil and sys imports. Run ruff format
on all modified files.

Author: claude

src/instructlab/training/gpt_oss_utils_correct.py

@@ -416,9 +416,11 @@ def is_known_model(
     # convert to config
     model_config = model_path_or_config
     if isinstance(model_path_or_config, str):
-        _trust_remote = os.environ.get(
-            "TRUST_REMOTE_CODE", ""
-        ).lower() in ("1", "true", "yes")
+        _trust_remote = os.environ.get("TRUST_REMOTE_CODE", "").lower() in (
+            "1",
+            "true",
+            "yes",
+        )
         model_config = AutoConfig.from_pretrained(
             model_path_or_config, trust_remote_code=_trust_remote
         )

src/instructlab/training/main_ds.py

@@ -5,9 +5,7 @@
 import datetime
 import logging
 import os
-import shutil
 import subprocess
-import sys
 import time
 import warnings
 
@@ -638,15 +636,8 @@ def run_training(torch_args: TorchrunArgs, train_args: TrainingArgs) -> None:
         os.makedirs(train_args.ckpt_output_dir, exist_ok=True)
 
     # build distributed training command
-    torchrun_path = shutil.which("torchrun")
-    if not torchrun_path:
-        raise RuntimeError(
-            "torchrun executable not found in PATH. "
-            "Ensure PyTorch is installed correctly."
-        )
-
     command = [
-        torchrun_path,
+        "torchrun",
         f"--nproc-per-node={torch_args.nproc_per_node}",
         f"--nnodes={torch_args.nnodes}",
         f"--node-rank={torch_args.node_rank}",

src/instructlab/training/model.py

@@ -615,13 +615,9 @@ def __init__(
             lora_quant_bits=lora_quant_bits,
             trust_remote_code=trust_remote_code,
         )
-        if is_vlm_with_causal_lm(
-            model_path, trust_remote_code=trust_remote_code
-        ):
+        if is_vlm_with_causal_lm(model_path, trust_remote_code=trust_remote_code):
             self.model = extract_causal_lm_from_vlm(model_path, self.base_model_args)
-        elif is_vlm_for_direct_loading(
-            model_path, trust_remote_code=trust_remote_code
-        ):
+        elif is_vlm_for_direct_loading(model_path, trust_remote_code=trust_remote_code):
             self.model = load_vlm_for_text_training(model_path, self.base_model_args)
         else:
             self.model = AutoModelForCausalLM.from_pretrained(**self.base_model_args)

src/instructlab/training/tokenizer_utils.py

@@ -98,9 +98,11 @@ def setup_tokenizer(
     model_name_or_path,
     chat_tmpl_path: str | None = None,
 ) -> PreTrainedTokenizer:
-    trust_remote_code = os.environ.get(
-        "TRUST_REMOTE_CODE", ""
-    ).lower() in ("1", "true", "yes")
+    trust_remote_code = os.environ.get("TRUST_REMOTE_CODE", "").lower() in (
+        "1",
+        "true",
+        "yes",
+    )
     tokenizer = AutoTokenizer.from_pretrained(
         model_name_or_path, trust_remote_code=trust_remote_code
     )

tests/unit/test_pretraining_mode.py

@@ -138,9 +138,7 @@ def encode(self, text, add_special_tokens=True):
                     num_cpu_procs=1,
                 )
 
-                mock_auto.assert_called_once_with(
-                    "stub-model", trust_remote_code=False
-                )
+                mock_auto.assert_called_once_with("stub-model", trust_remote_code=False)
 
             output_file = output_dir / "data.jsonl"
             self.assertTrue(output_file.exists())