refactor and support for multi algs fusion

[n1ck-guo]

Description

What this PR does
Introduces a composable QuantizationPipeline that separates pre-processing algorithms (e.g. AWQ) from the terminal block-quantizer (e.g. AutoRound/RTN), and lets users compose them declaratively via config lists.

Key changes:

• QuantizationPipeline (algorithms/quantization/pipeline.py): new orchestration layer — [preprocessors…] + block_quantizer. Replaces the implicit algorithm coupling in DataDrivenCompressor.
• BasePipelineMember / BaseWeightTransformer / BaseQuantizer (base.py): clean class hierarchy with unified lifecycle hooks (prepare_run, quantize_block, finalize_run).
• AWQConfig + AWQQuantizer refactored as a BaseWeightTransformer — pure weight-smoothing preprocessor, no quantization loop of its own.
• DiffusionMixin injected dynamically at pipeline construction time (is_diffusion=True) — no if is_diffusion branches in algorithm code.
• CLI (auto_round/cli/) rewritten to expose --alg_configs for composing pipelines from the command line.

Usage: AWQ + AutoRound fusion

from auto_round import AutoRound
from auto_round.algorithms.quantization.awq.config import AWQConfig
from auto_round.algorithms.quantization.sign_round.config import SignRoundConfig
ar = AutoRound(
    [AWQConfig(), SignRoundConfig(iters=200)],
    model_name,
    scheme="W4A16",
)
model, layer_config = ar.quantize()

Passing a list of configs activates the pipeline: AWQ smoothing runs first on each block, then AutoRound's SignSGD optimization runs on the smoothed weights. Passing a single config (old API) continues to work unchanged.
Compatibility

• Single-config API (AutoRound(model, ...)) is fully backward compatible.
• All existing CPU tests pass; pre-existing environment failures (missing auto-round-lib, device fixtures) are unrelated to this PR.

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.

[Copilot]

Pull request overview

This PR introduces a composable quantization “pipeline” abstraction that separates preprocessors (e.g., AWQ smoothing) from the terminal block quantizer (e.g., RTN / SignRound), and refactors compressor + CLI wiring to support multi-algorithm composition via ordered config lists.

Changes:

• Added QuantizationPipeline / BlockContext orchestration and a unified quantizer/preprocessor lifecycle (prepare_run, block_forward_hooks, pre_quantize_block, quantize_block, finalize_run).
• Refactored quantizers (RTN, SignRound, SignRoundV2, AWQ) to use the new pipeline context and hook model.
• Rewrote the CLI into auto_round/cli/ with algorithm handlers + config building, and added docs + tests for pipeline composition.

Reviewed changes

Copilot reviewed 37 out of 37 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
test/test_cuda/quantization/test_torch_compile.py	Update test imports/usages for new BaseQuantizer naming.
test/test_cuda/algorithms/test_alg_ext.py	Update alg-ext regression expectations for SignRoundV2 wrapper behavior.
test/test_cpu/utils/test_alg_ext.py	Update import smoke for new SignRoundV2 location.
test/test_cpu/core/test_pipeline_fail_fast.py	New unit tests for registry + pipeline construction failure modes.
test/test_cpu/core/test_awq_autoround_smoke.py	New smoke test for AWQ + AutoRound fusion via config list.
docs/step_by_step.md	Document AWQ as a preprocessing algorithm + CLI/API composition examples.
docs/step_by_step_CN.md	Chinese translation updates for the AWQ section.
auto_round/context/model.py	Minor refactor/formatting + rename doc reference to BaseQuantizer.
auto_round/compressors/zero_shot.py	Adapt zero-shot RTN path to use BlockContext.
auto_round/compressors/entry.py	Centralized alias/config resolution + routing for pipeline/preprocessors and model-free path.
auto_round/compressors/data_driven.py	Refactor block loop to pipeline lifecycle + hook scheduling via BlockContext.
auto_round/compressors/config.py	Removed legacy ExtraConfig container module.
auto_round/compressors/base.py	Build and expose QuantizationPipeline; quantizer becomes a forwarding property.
auto_round/compressors/init.py	Remove legacy config exports; keep lazy imports for compressors.
auto_round/cli/parser.py	New argparse construction module for quantize/list/eval commands.
auto_round/cli/main.py	New CLI router + recipe defaults + algorithm-help printing.
auto_round/cli/algorithms.py	New CLI algorithm registry/handlers that build ordered config lists.
auto_round/cli/init.py	Export CLI entrypoints from the new CLI package.
auto_round/calibration/state.py	Update doc reference from BaseQuantizers to BaseQuantizer.
auto_round/autoround.py	Add alg_configs fast-path into new entry point; adjust skip args.
auto_round/algorithms/transforms/rotation/config.py	Add/expand RotationConfig docstring and minor formatting.
auto_round/algorithms/quantization/sign_roundv2/quantizer.py	Move imatrix hooks into quantizer hook context + minor cleanups.
auto_round/algorithms/quantization/sign_round/quantizer.py	Convert block quantization API to ctx (BlockContext) and use BlockIO.
auto_round/algorithms/quantization/sign_round/config.py	Replace sparse docstring with a clearer/structured one.
auto_round/algorithms/quantization/rtn/quantizer.py	Convert RTN to ctx API; move imatrix hook logic into context manager.
auto_round/algorithms/quantization/rtn/config.py	Add docstring and minor logging formatting.
auto_round/algorithms/quantization/registry.py	New alias→config registry used by entry routing.
auto_round/algorithms/quantization/pipeline.py	New pipeline abstraction (BlockContext, RunContext, policy merge, IO helpers).
auto_round/algorithms/quantization/config.py	Add top-level QuantizationConfig docstring.
auto_round/algorithms/quantization/base.py	Introduce BasePipelineMember, BaseWeightTransformer, BaseQuantizer, mixins; new hook lifecycle.
auto_round/algorithms/quantization/awq/mappings.py	Downgrade a log line from warning to info for hybrid-attention mapping build.
auto_round/algorithms/quantization/awq/config.py	Reframe AWQConfig as a preprocessor config; improve validation and repr.
auto_round/algorithms/quantization/awq/init.py	Export AWQConfig/AWQQuantizer from the awq subpackage.
auto_round/algorithms/quantization/init.py	Re-export pipeline types/quantizers; currently contains duplicate AWQ imports.
auto_round/alg_ext.py	Removed legacy alg-ext implementation module.
auto_round/main.py	Replace huge legacy CLI with a shim that forwards to auto_round.cli.main.

Comments suppressed due to low confidence (1)

auto_round/algorithms/quantization/init.py:41

• Duplicate imports for AWQConfig/AWQQuantizer are present twice in this module, which is redundant and can trigger linting/formatting issues. Remove the second pair of imports and keep a single import location.

from auto_round.algorithms.quantization.awq.config import AWQConfig
from auto_round.algorithms.quantization.awq.quantizer import AWQQuantizer
from auto_round.algorithms.quantization.rtn.config import RTNConfig
from auto_round.algorithms.quantization.rtn.quantizer import RTNQuantizer, OptimizedRTNQuantizer

[wenhuach21]

The common tuning parameters should be transferable across different algorithms, e.g., for awq and signround, the shared parameter is weight clip ratio

[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).