Implement shard parallelism configuration and update documentation

[xin3he]

Description

This pull request introduces and documents a new environment variable, AR_MODEL_FREE_SHARD_PARALLELISM, which controls the number of shards processed concurrently during model-free quantization. It also adds comprehensive tests for the new parallelism policy, updates documentation (both English and Chinese) to explain the variable and its effects (including benchmarks), and adds a new test for quantization config building. These changes improve configurability, transparency, and reliability of model-free quantization.

Environment variable and configuration support:

• Added the AR_MODEL_FREE_SHARD_PARALLELISM environment variable to auto_round/envs.py to control model-free quantization parallelism, including logic for auto-policy and environment override. [1] [2]

Documentation updates:

• Documented AR_MODEL_FREE_SHARD_PARALLELISM in both English and Chinese environment variable guides, including description, default policy, usage, and examples. [1] [2]
• Added model-free quantization parallelism benchmarks (tables and analysis) to both English and Chinese step-by-step guides. [1] [2]

Testing and validation:

• Added a new test class to verify the shard parallelism policy, including auto-policy, environment override, capping, invalid input fallback, and correct handling of non-divisible shard counts.
• Added a test for the quantization config builder to ensure correct filtering of ignored layers.
• Imported the new _build_quantization_config function in the test module.

Parallelism Benefit Table (Using Rounded Minutes)

Model	Scheme	Comparison	Runtime Change (min)	Speedup	Time Saved	Peak Memory Change
Qwen3-235B	W4A16	Parallelism 1 -> 10	17 -> 5	3.40x	70.6%	2G -> 8G
GLM-5.2	MXFP4-Mixed	Parallelism 1 -> 10	60 -> 16	3.75x	73.3%	2G -> 27G
GLM-5.2	W4A16	Parallelism 1 -> 10	30 -> 10	3.00x	66.7%	3G -> 16G
GLM-5.2	W4A16	Parallelism 1 -> 20	30 -> 8	3.75x	73.3%	3G -> 32G
MiniMax-M2.7	W4A16	Parallelism 1 -> 10	18 -> 5	3.60x	72.2%	2G -> 10G
DeepSeek-V4-Pro	W4A16	Parallelism 1 -> 10	80 -> 13	6.15x	83.8%	6G -> 50G

Type of Change

New feature

Related Issues

Fixes or relates to #

Checklist Before Submitting

• My code has been tested locally.
• Documentation has been updated as needed.
• New or updated tests are included where applicable.
• The CUDA CI has passed. You can trigger it by commenting /azp run Unit-Test-CUDA-AutoRound.

[chensuyue]

/azp run Unit-Test-CUDA-AutoRound

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[chensuyue]

/azp run Unit-Test-CUDA-AutoRound

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[lkk12014402]

LGTM

[chensuyue]

/azp run Unit-Test-CUDA-AutoRound

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[chensuyue]

/azp run Unit-Test-CUDA-AutoRound

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).