Implement aliasable mixin and alias activation ordering (python3.9 fix)

[kylesayrs]

Purpose

• Add aliases for activation ordering options to improve usability for researchers familiar with autogptq
• Update previous Implement aliasable mixin and alias activation ordering #213 to support python<=3.9

Changes

• Implement AliasableEnum mixin which allows enums to be aliased
• Add AliasableEnum to ActivationOrdering with the following map:

{
    "dynamic": "group",
    "static": "weight",
}

• Remove the @property decorator from aliases to avoid interactions in python<=3.9 https://bugs.python.org/issue43682

Testing

• Added passing tests in tests/test_quantization/test_quant_args.py
• Successfully quantized a model using dynamic actorder and tested e2e with vllm
• Tested the above using python3.9

llama3.py

from accelerate import cpu_offload
from datasets import load_dataset
from transformers import AutoModelForCausalLM, AutoTokenizer

from llmcompressor.modifiers.quantization import GPTQModifier
from llmcompressor.transformers import oneshot

# Select model and load it.
# MODEL_ID = "meta-llama/Meta-Llama-3-8B-Instruct"
MODEL_ID = "meta-llama/Llama-3.2-1B-Instruct"

model = AutoModelForCausalLM.from_pretrained(
    MODEL_ID,
    device_map="cuda:0",
    torch_dtype="auto",
)
# cpu_offload(model)
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)

# Select calibration dataset.
DATASET_ID = "HuggingFaceH4/ultrachat_200k"
DATASET_SPLIT = "train_sft"

# Select number of samples. 512 samples is a good place to start.
# Increasing the number of samples can improve accuracy.
NUM_CALIBRATION_SAMPLES = 285  # 2048
MAX_SEQUENCE_LENGTH = 2048

# Load dataset and preprocess.
ds = load_dataset(DATASET_ID, split=DATASET_SPLIT)
ds = ds.shuffle(seed=42).select(range(NUM_CALIBRATION_SAMPLES))


def preprocess(example):
    return {
        "text": tokenizer.apply_chat_template(
            example["messages"],
            tokenize=False,
        )
    }


ds = ds.map(preprocess)


# Tokenize inputs.
def tokenize(sample):
    return tokenizer(
        sample["text"],
        padding=False,
        max_length=MAX_SEQUENCE_LENGTH,
        truncation=True,
        add_special_tokens=False,
    )


ds = ds.map(tokenize, remove_columns=ds.column_names)

# Configure the quantization algorithm to run.
from compressed_tensors.quantization import QuantizationArgs, QuantizationType, QuantizationStrategy, ActivationOrdering, QuantizationScheme
recipe = GPTQModifier(
    targets="Linear",
    config_groups={
        "config_group": QuantizationScheme(
            targets=["Linear"],
            weights=QuantizationArgs(
                num_bits=4,
                type=QuantizationType.INT,
                strategy=QuantizationStrategy.GROUP,
                group_size=128,
                symmetric=True,
                dynamic=False,
                actorder="dynamic",
            ),
        ),
    },
    ignore=["lm_head"],
    dampening_frac=0.5
)

# Apply algorithms.
oneshot(
    model=model,
    dataset=ds,
    recipe=recipe,
    max_seq_length=MAX_SEQUENCE_LENGTH,
    num_calibration_samples=NUM_CALIBRATION_SAMPLES,
)

# Confirm generations of the quantized model look sane.
print("\n\n")
print("========== SAMPLE GENERATION ==============")
input_ids = tokenizer("Hello my name is", return_tensors="pt").input_ids.to("cuda")
output = model.generate(input_ids, max_new_tokens=100)
print(tokenizer.decode(output[0]))
print("==========================================\n\n")

# Save to disk compressed.
SAVE_DIR = MODEL_ID.split("/")[1] + "-W4A16-G128"
model.save_pretrained(SAVE_DIR, save_compressed=True)
tokenizer.save_pretrained(SAVE_DIR)

[kylesayrs]

I've confirmed that the tests which were previously failing on python3.9 are now passing on python3.9