Merge branch 'main' into torch-compile-observers

Author: HDCharles

docs/developer/index.md

@@ -14,14 +14,14 @@ LLM Compressor is an open-source project that values community contributions. We
 
     Our community guidelines ensure that participation in the LLM Compressor project is a positive, inclusive, and respectful experience for everyone.
 
-    [:octicons-arrow-right-24: Code of Conduct](code-of-conduct.md)
+    [:octicons-arrow-right-24: Code of Conduct](../../CODE_OF_CONDUCT.md)
 
 - :material-source-pull:{ .lg .middle } Contributing Guide
 
     ---
 
     Learn how to effectively contribute to LLM Compressor, including reporting bugs, suggesting features, improving documentation, and submitting code.
 
-    [:octicons-arrow-right-24: Contributing Guide](contributing.md)
+    [:octicons-arrow-right-24: Contributing Guide](../../CONTRIBUTING.md)
 
 </div>

examples/big_models_with_sequential_onloading/llama3.3_70b.py

@@ -4,7 +4,7 @@
 
 from llmcompressor import oneshot
 from llmcompressor.modifiers.quantization import GPTQModifier
-from llmcompressor.modifiers.smoothquant import SmoothQuantModifier
+from llmcompressor.modifiers.transform.smoothquant import SmoothQuantModifier
 
 # Select model and load it.
 model_id = "meta-llama/Llama-3.3-70B-Instruct"

examples/multimodal_audio/README.md

@@ -44,7 +44,7 @@ Sequential targets are the modules which determine the granularity of error prop
 Choosing sequential targets with higher granularity (for example "Linear" instead of "LlamaDecoderLayer") will result in fewer hessians being allocated at the same time, decreasing the memory requirements for compression. This may also increase the recovered accuracy of the model, as compression error is propagated at a higher granularity. However, using higher granularity sequential targets may also increase compression time, as more time is spent offloading and onloading activations.
 
 ## Adding Your Own Smoothquant Mappings ##
-For a guide on adding smoothquant mappings for your dataset, see the [SmoothQuant Guide](/src/llmcompressor/modifiers/smoothquant/README.md).
+For a guide on adding smoothquant mappings for your dataset, see the [SmoothQuant Guide](/src/llmcompressor/modifiers/transform/smoothquant/README.md).
 
 ## Adding Your Own Data Collator ##
 Most examples utilize a generic `data_collator` which correctly correlates data for most multimodal datasets. If you find that your model needs custom data collation (as is the case with [pixtral](/examples/multimodal_vision/pixtral_example.py)), you can modify this function to reflect these model-specific requirements.

examples/multimodal_vision/README.md

@@ -48,7 +48,7 @@ Sequential targets are the modules which determine the granularity of error prop
 Choosing sequential targets with higher granularity (for example "Linear" instead of "LlamaDecoderLayer") will result in fewer hessians being allocated at the same time, decreasing the memory requirements for compression. This may also increase the recovered accuracy of the model, as compression error is propagated at a higher granularity. However, using higher granularity sequential targets may also increase compression time, as more time is spent offloading and onloading activations.
 
 ## Adding Your Own Smoothquant Mappings ##
-For a guide on adding smoothquant mappings for your dataset, see the [SmoothQuant Guide](/src/llmcompressor/modifiers/smoothquant/README.md).
+For a guide on adding smoothquant mappings for your dataset, see the [SmoothQuant Guide](/src/llmcompressor/modifiers/transform/smoothquant/README.md).
 
 ## Adding Your Own Data Collator ##
 Most examples utilize a generic `data_collator` which correctly correlates data for most multimodal datasets. If you find that your model needs custom data collation (as is the case with [pixtral](/examples/multimodal_vision/pixtral_example.py)), you can modify this function to reflect these model-specific requirements.

examples/quantization_w8a8_int8/README.md

@@ -86,7 +86,7 @@ We first select the quantization algorithm. For W8A8, we want to:
 ```python
 from llmcompressor import oneshot
 from llmcompressor.modifiers.quantization import GPTQModifier
-from llmcompressor.modifiers.smoothquant import SmoothQuantModifier
+from llmcompressor.modifiers.transform.smoothquant import SmoothQuantModifier
 
 # Configure the quantization algorithms to run.
 recipe = [

examples/quantization_w8a8_int8/llama3_example.py

@@ -4,7 +4,7 @@
 
 from llmcompressor import oneshot
 from llmcompressor.modifiers.quantization import GPTQModifier
-from llmcompressor.modifiers.smoothquant import SmoothQuantModifier
+from llmcompressor.modifiers.transform.smoothquant import SmoothQuantModifier
 
 # Select model and load it.
 MODEL_ID = "meta-llama/Meta-Llama-3-8B-Instruct"

src/llmcompressor/modifiers/logarithmic_equalization/base.py

@@ -1,7 +1,7 @@
 import torch
 from torch.nn import Module
 
-from llmcompressor.modifiers.smoothquant import SmoothQuantModifier
+from llmcompressor.modifiers.transform.smoothquant import SmoothQuantModifier
 
 __all__ = ["LogarithmicEqualizationModifier"]
 

src/llmcompressor/modifiers/smoothquant/__init__.py

@@ -1,3 +1,23 @@
 # ruff: noqa
+"""
+Backwards compatibility shim for SmoothQuantModifier.
 
-from .base import *
+This module has been moved to llmcompressor.modifiers.transform.smoothquant.
+This shim will be removed in a future version.
+"""
+
+import warnings
+
+warnings.warn(
+    "Importing from 'llmcompressor.modifiers.smoothquant' is deprecated. "
+    "Please update your imports to use 'llmcompressor.modifiers.transform.smoothquant' "
+    "or 'llmcompressor.modifiers.transform' instead. "
+    "This compatibility shim will be removed in a future version.",
+    DeprecationWarning,
+    stacklevel=2,
+)
+
+from llmcompressor.modifiers.transform.smoothquant import *
+from llmcompressor.modifiers.transform.smoothquant.base import SmoothQuantModifier
+
+__all__ = ["SmoothQuantModifier"]