Remove --parallel_threads flag from to_huggingface.py

[chiajunglien]

Description

This PR removes the exposed --parallel_threads command-line flag from src/maxtext/checkpoint_conversion/to_huggingface.py and cleans up its arguments.

The --parallel_threads flag was originally introduced to explicitly limit concurrency in the E2E testing script (test_llama3.1_70b_to_hf.sh) to prevent OOM errors on smaller memory profiles like v5p-8.

Since that script was subsequently refactored to use alternative memory optimization techniques and no longer passes --parallel_threads=2, keeping this isolated flag implementation in to_huggingface.py is obsolete.

Tests

xpk workload create \
  --cluster=mlperf-v5p \
  --workload=llama-hf-3 \
  --device-type=v5p-8 \
  --num-slices=1 \
  --docker-image=[gcr.io/tpu-prod-env-multipod/maxtext_post_training_stable:32316847848](http://gcr.io/tpu-prod-env-multipod/maxtext_post_training_stable:32316847848) \
  --project=cloud-tpu-multipod-dev \
  --zone=europe-west4-b \
  --priority=very-high \
  --skip-validation \
  --env GCS_OUTPUT=gs://ml-auto-solutions/output/unowned/to-hf-v5p-8-2026-08-19-01-47-34/ \
  --command="set -xue; export HF_HOME=/dev/shm/hf_cache; export LIBTPU_INIT_ARGS="--xla_tpu_scoped_vmem_limit_kib=20480"; python3 -m maxtext.checkpoint_conversion.to_huggingface model_name=llama3.1-70b tokenizer_type=huggingface load_parameters_path=gs://runner-maxtext-logs/llama3.1-70b/sft/e2e-32200790571/checkpoints/2/model_params base_output_directory=gs://runner-maxtext-logs/llama3.1-70b/to_huggingface/scanned/e2e-32200790571 use_multimodal=false scan_layers=true weight_dtype=bfloat16 checkpoint_storage_concurrent_gb=48 checkpoint_storage_use_zarr3=False checkpoint_storage_use_ocdbt=False skip_jax_distributed_system=true"

Log: https://cloudlogging.app.goo.gl/Wh68u2VnKSo3hw9u7

Checklist

Before submitting this PR, please make sure (put X in square brackets):

• I have performed a self-review of my code. For an optional AI review, add the gemini-review label.
• I have necessary comments in my code, particularly in hard-to-understand areas.
• I have run end-to-end tests tests and provided workload links above if applicable.
• I have made or will make corresponding changes to the doc if needed, including adding new documentation pages to the relevant Table of Contents (toctree directive) as explained in our documentation.

[gemini-code-assist]

Code Review

This pull request removes the "parallel_threads" command-line flag and its validator from "to_huggingface.py". However, this change will cause a runtime "AttributeError" because "FLAGS.parallel_threads" is still referenced in a downstream call to "save_model_files". You should remove this argument from the function call to allow it to use its default value.

[codecov]

Codecov Report

❌ Patch coverage is 0% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...rc/maxtext/checkpoint_conversion/to_huggingface.py	0.00%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!