Fix: ValueError: filedescriptor out of range in select()

Author: RexBearIU

src/MaxText/examples/rl_llama3_demo.ipynb

@@ -144,8 +144,7 @@
    "source": [
     "import datetime\n",
     "import os\n",
-    "import sys\n",
-    "from pathlib import Path\n",
+    "import subprocess\n",
     "import MaxText\n",
     "from huggingface_hub import login\n",
     "import jax\n",
@@ -244,16 +243,9 @@
    "source": [
     "if not os.path.exists(MODEL_CHECKPOINT_PATH):\n",
     "    # install torch for the conversion script\n",
-    "    !python3 -m pip install torch --index-url https://download.pytorch.org/whl/cpu\n",
-    "\n",
-    "    !JAX_PLATFORMS=cpu PYTHONPATH={MAXTEXT_REPO_ROOT} {sys.executable} -m MaxText.utils.ckpt_conversion.to_maxtext \\\n",
-    "      {MAXTEXT_REPO_ROOT}/configs/base.yml \\\n",
-    "      model_name={MODEL_NAME} \\\n",
-    "      base_output_directory={MODEL_CHECKPOINT_PATH} \\\n",
-    "      hf_access_token={HF_TOKEN} \\\n",
-    "      use_multimodal=false \\\n",
-    "      scan_layers=true \\\n",
-    "      skip_jax_distributed_system=True\n",
+    "    subprocess.run('uv pip install torch --index-url https://download.pytorch.org/whl/cpu', shell=True, check=True)\n",
+    "\n",
+    "    subprocess.run(f'JAX_PLATFORMS=cpu PYTHONPATH={MAXTEXT_REPO_ROOT} {sys.executable} -m MaxText.utils.ckpt_conversion.to_maxtext {MAXTEXT_REPO_ROOT}/configs/base.yml model_name={MODEL_NAME} base_output_directory={MODEL_CHECKPOINT_PATH} hf_access_token={HF_TOKEN} use_multimodal=false scan_layers=true skip_jax_distributed_system=True', shell=True, check=True)\n",
     "\n",
     "if not os.path.exists(MODEL_CHECKPOINT_PATH):\n",
     "    raise ValueError(\"Model checkpoint conversion failed. Check the logs above.\")"

src/MaxText/examples/sft_llama3_demo.ipynb

@@ -173,6 +173,7 @@
    "cell_type": "code",
    "execution_count": null,
    "metadata": {},
+   "outputs": [],
    "source": [
     "if IN_COLAB:\n",
     "    HF_TOKEN = userdata.get(\"HF_TOKEN\")\n",
@@ -234,16 +235,9 @@
    "source": [
     "if not os.path.exists(MODEL_CHECKPOINT_PATH):\n",
     "    # install torch for the conversion script\n",
-    "    !python3 -m pip install torch --index-url https://download.pytorch.org/whl/cpu\n",
-    "\n",
-    "    !JAX_PLATFORMS=cpu PYTHONPATH={MAXTEXT_REPO_ROOT} {sys.executable} -m MaxText.utils.ckpt_conversion.to_maxtext \\\n",
-    "      {MAXTEXT_REPO_ROOT}/configs/base.yml \\\n",
-    "      model_name={MODEL_NAME} \\\n",
-    "      base_output_directory={MODEL_CHECKPOINT_PATH} \\\n",
-    "      hf_access_token={HF_TOKEN} \\\n",
-    "      use_multimodal=false \\\n",
-    "      scan_layers=true \\\n",
-    "      skip_jax_distributed_system=True\n",
+    "    subprocess.run('uv pip install torch --index-url https://download.pytorch.org/whl/cpu', shell=True, check=True)\n",
+    "\n",
+    "    subprocess.run(f'JAX_PLATFORMS=cpu PYTHONPATH={MAXTEXT_REPO_ROOT} {sys.executable} -m MaxText.utils.ckpt_conversion.to_maxtext {MAXTEXT_REPO_ROOT}/configs/base.yml model_name={MODEL_NAME} base_output_directory={MODEL_CHECKPOINT_PATH} hf_access_token={HF_TOKEN} use_multimodal=false scan_layers=true skip_jax_distributed_system=True', shell=True, check=True)\n",
     "\n",
     "if not os.path.exists(MODEL_CHECKPOINT_PATH):\n",
     "    raise ValueError(\"Model checkpoint conversion failed. Check the logs above.\")"

src/MaxText/examples/sft_qwen3_demo.ipynb

@@ -195,10 +195,10 @@
     "import jax\n",
     "import os\n",
     "import sys\n",
+    "import subprocess\n",
     "import transformers\n",
     "\n",
     "import MaxText\n",
-    "from MaxText import pyconfig\n",
     "from MaxText.examples.sft_train_and_evaluate import evaluate_model, get_test_dataset\n",
     "from MaxText.integration.tunix.tunix_adapter import TunixMaxTextAdapter\n",
     "from MaxText.sft import sft_trainer\n",
@@ -312,16 +312,9 @@
    "source": [
     "if not os.path.exists(MODEL_CHECKPOINT_PATH):\n",
     "    # install torch for the conversion script\n",
-    "    !python3 -m pip install torch --index-url https://download.pytorch.org/whl/cpu\n",
-    "\n",
-    "    !JAX_PLATFORMS=cpu PYTHONPATH={MAXTEXT_REPO_ROOT} {sys.executable} -m MaxText.utils.ckpt_conversion.to_maxtext \\\n",
-    "      {MAXTEXT_REPO_ROOT}/configs/base.yml \\\n",
-    "      model_name={MODEL_NAME} \\\n",
-    "      base_output_directory={MODEL_CHECKPOINT_PATH} \\\n",
-    "      hf_access_token={HF_TOKEN} \\\n",
-    "      use_multimodal=false \\\n",
-    "      scan_layers=true \\\n",
-    "      skip_jax_distributed_system=True\n",
+    "    subprocess.run('uv pip install torch --index-url https://download.pytorch.org/whl/cpu', shell=True, check=True)\n",
+    "\n",
+    "    subprocess.run(f'JAX_PLATFORMS=cpu PYTHONPATH={MAXTEXT_REPO_ROOT} {sys.executable} -m MaxText.utils.ckpt_conversion.to_maxtext {MAXTEXT_REPO_ROOT}/configs/base.yml model_name={MODEL_NAME} base_output_directory={MODEL_CHECKPOINT_PATH} hf_access_token={HF_TOKEN} use_multimodal=false scan_layers=true skip_jax_distributed_system=True', shell=True, check=True)\n",
     "\n",
     "if not os.path.exists(MODEL_CHECKPOINT_PATH):\n",
     "    raise ValueError(\"Model checkpoint conversion failed. Check the logs above.\")"