Add Deepseek and GPT OSS weight mapping for RL.

[abhinavclemson]

Description

Add Deepseek and GPT OSS vLLM standalone weight mapping.

This change introduces deepseek3.py and gpt_oss.py containing DEEPSEEK_VLLM_MAPPING and GPT_OSS_VLLM_MAPPING dataclasses. These dataclasses define how to map MaxText model weights to a format compatible with vLLM format on tpu-inference side, including parameter name mappings, and sharding specifications.

Command:

xpk workload create-pathways  --workload "abhinav-grpo-new"  --docker-image [gcr.io/cloud-tpu-multipod-dev/abhinavsing_vllm_runner:latest](http://gcr.io/cloud-tpu-multipod-dev/abhinavsing_vllm_runner:latest) --cluster mazumdera-v5p-64-pw-2 --tpu-type=v5p-64  --num-slices=1  --priority=high --custom-pathways-worker-args="--xprof_max_trace_buffers=16384" --command "MODEL_IMPL_TYPE=flax_nnx NEW_MODEL_DESIGN=1 TF_CPP_MIN_LOG_LEVEL=0 JAX_PLATFORMS=proxy JAX_BACKEND_TARGET=grpc://127.0.0.1:29000 ENABLE_PATHWAYS_PERSISTENCE='1' python3 -m src.MaxText.rl.train_rl src/MaxText/configs/rl.yml   model_name=gpt-oss-20b   tokenizer_path=unsloth/gpt-oss-20b-BF16   load_parameters_path=gs://shuningjin-multipod-dev/gpt-oss-20b/scan-flags-false-2025-11-11-01-42-40/0/items   run_name=test-tunix-maxtext-gpt-oss-20b-BF16   base_output_directory=gs://abhinavsing_bucket/   checkpoint_storage_use_ocdbt=False checkpoint_storage_use_zarr3=False rollout_data_parallelism=2 rollout_tensor_parallelism=8 hbm_utilization_vllm=0.6 batch_size=8 micro_batch_size=8 profiler=xplane profiler_steps=2"

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.

[codecov]

Codecov Report

❌ Patch coverage is 0% with 58 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...axText/integration/tunix/weight_mapping/gpt_oss.py	0.00%	32 Missing ⚠️
...Text/integration/tunix/weight_mapping/deepseek3.py	0.00%	20 Missing ⚠️
...xText/integration/tunix/weight_mapping/__init__.py	0.00%	6 Missing ⚠️

📢 Thoughts on this report? Let us know!