riaz.somc/transformers
1
0
Fork 0
mirror of https://github.com/huggingface/transformers.git synced 2025-07-03 12:50:06 +06:00
Code Issues Actions 28 Packages Projects Releases Wiki Activity

Name change AOPermod -> ModuleFqn (#38456)

Browse Source 
Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com>
Co-authored-by: Mohamed Mekkouri <93391238+MekkCyber@users.noreply.github.com>
This commit is contained in:
Driss Guessous 2025-06-03 11:43:31 -04:00 committed by GitHub
parent e8b292e35f
commit 279000bb70
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 17 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
docs/source/en/quantization/torchao.md
View File

@ -276,18 +276,18 @@ print(tokenizer.decode(output[0], skip_special_tokens=True))
### Per Module Quantization
#### 1. Skip quantization for certain layers
With `AOPerModuleConfig` we can specify a default configuration for all layers while skipping quantization for certain layers.
With `ModuleFqnToConfig` we can specify a default configuration for all layers while skipping quantization for certain layers.
```py
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TorchAoConfig
model_id = "meta-llama/Llama-3.1-8B-Instruct"
from torchao.quantization import Int4WeightOnlyConfig, AOPerModuleConfig
from torchao.quantization import Int4WeightOnlyConfig, ModuleFqnToConfig
config = Int4WeightOnlyConfig(group_size=128)
# set default to int4 (for linears), and skip quantizing `model.layers.0.self_attn.q_proj`
quant_config = AOPerModuleConfig({"_default": config, "model.layers.0.self_attn.q_proj": None})
quant_config = ModuleFqnToConfig({"_default": config, "model.layers.0.self_attn.q_proj": None})
quantization_config = TorchAoConfig(quant_type=quant_config)
quantized_model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", torch_dtype=torch.bfloat16, quantization_config=quantization_config)
# lm_head is not quantized and model.layers.0.self_attn.q_proj is not quantized
@ -311,7 +311,7 @@ from transformers import AutoModelForCausalLM, AutoTokenizer, TorchAoConfig
model_id = "facebook/opt-125m"
from torchao.quantization import Int4WeightOnlyConfig, AOPerModuleConfig, Int8DynamicActivationInt4WeightConfig, IntxWeightOnlyConfig, PerAxis, MappingType
from torchao.quantization import Int4WeightOnlyConfig, ModuleFqnToConfig, Int8DynamicActivationInt4WeightConfig, IntxWeightOnlyConfig, PerAxis, MappingType
weight_dtype = torch.int8
granularity = PerAxis(0)
@ -322,7 +322,7 @@ embedding_config = IntxWeightOnlyConfig(
mapping_type=mapping_type,
)
linear_config = Int8DynamicActivationInt4WeightConfig(group_size=128)
quant_config = AOPerModuleConfig({"_default": linear_config, "model.decoder.embed_tokens": embedding_config, "model.decoder.embed_positions": None})
quant_config = ModuleFqnToConfig({"_default": linear_config, "model.decoder.embed_tokens": embedding_config, "model.decoder.embed_positions": None})
# set `include_embedding` to True in order to include embedding in quantization
# when `include_embedding` is True, we'll remove input embedding from `modules_not_to_convert` as well
quantization_config = TorchAoConfig(quant_type=quant_config, include_embedding=True)

8
src/transformers/quantizers/quantizer_torchao.py
View File

@ -261,12 +261,12 @@ class TorchAoHfQuantizer(HfQuantizer):
model.tie_weights()
setattr(model.config.get_text_config(decoder=True), "tie_word_embeddings", False)
# handle AOPerModuleConfig, introduced in torchao 0.11.0+
if self.quantization_config._get_ao_version() > version.Version("0.10.0"):
from torchao.quantization import AOPerModuleConfig
# handle ModuleFqnToConfig, introduced in torchao 0.12.0+
if self.quantization_config._get_ao_version() >= version.Version("0.12.0"):
from torchao.quantization import ModuleFqnToConfig
config = self.quantization_config.get_apply_tensor_subclass()
if isinstance(config, AOPerModuleConfig):
if isinstance(config, ModuleFqnToConfig):
module_fqn, _ = param_name.rsplit(".", 1)
c = None
if module_fqn in config.module_fqn_to_config:

6
tests/quantization/torchao_integration/test_torchao.py
View File

@ -43,10 +43,10 @@ if is_torchao_available():
TensorCoreTiledLayout,
)
from torchao.quantization import (
AOPerModuleConfig,
Int8WeightOnlyConfig,
IntxWeightOnlyConfig,
MappingType,
ModuleFqnToConfig,
PerAxis,
)
from torchao.quantization.autoquant import AQMixin
@ -226,7 +226,7 @@ class TorchAoTest(unittest.TestCase):
granularity=granularity,
mapping_type=mapping_type,
)
config = AOPerModuleConfig(
config = ModuleFqnToConfig(
{"_default": None, "model.embed_tokens": embedding_config, "lm_head": embedding_config}
)
# need set `include_input_output_embeddings` to True
@ -253,7 +253,7 @@ class TorchAoTest(unittest.TestCase):
@require_torchao_version_greater_or_equal("0.11.0")
def test_per_module_config_skip(self):
linear_config = Int8WeightOnlyConfig()
config = AOPerModuleConfig({"_default": linear_config, "model.layers.0.self_attn.q_proj": None})
config = ModuleFqnToConfig({"_default": linear_config, "model.layers.0.self_attn.q_proj": None})
quant_config = TorchAoConfig(quant_type=config)
quantized_model = AutoModelForCausalLM.from_pretrained(
self.model_name,