Whisper Large V3 Turbo Quantized.w4a16

python quantize.py --model_path openai/whisper-large-v3-turbo --quant_path "output_dir/whisper-large-v3-turbo-quantized.w4a16" --calib_size 1024 --group_size 64 --dampening_frac 0.01 --actorder weight

import torch
import argparse
from datasets import load_dataset
from transformers import WhisperProcessor
from llmcompressor import oneshot
from llmcompressor.modifiers.quantization import GPTQModifier
from llmcompressor.transformers.tracing import TraceableWhisperForConditionalGeneration
import os
from compressed_tensors.quantization import QuantizationArgs, QuantizationType, QuantizationStrategy, ActivationOrdering, QuantizationScheme
from llmcompressor.modifiers.smoothquant import SmoothQuantModifier

parser = argparse.ArgumentParser()
parser.add_argument('--model_path', type=str)
parser.add_argument('--quant_path', type=str)
parser.add_argument('--calib_size', type=int, default=256)
parser.add_argument('--dampening_frac', type=float, default=0.1) 
parser.add_argument('--observer', type=str, default="minmax")
parser.add_argument('--actorder', type=str, default="dynamic")
parser.add_argument('--group_size', type=int, default=128)
parser.add_argument('--save_dir', type=str, required=True)


args = parser.parse_args()
model_id = args.model_path

model = TraceableWhisperForConditionalGeneration.from_pretrained(
    model_id,
    device_map="auto",
    torch_dtype="auto",
)
model.config.forced_decoder_ids = None
processor = WhisperProcessor.from_pretrained(model_id)

# Configure processor the dataset task.
processor.tokenizer.set_prefix_tokens(language="en", task="transcribe")

# Select calibration dataset.
DATASET_ID = "MLCommons/peoples_speech"
DATASET_SUBSET = "test"
DATASET_SPLIT = "test"

# Select number of samples for calibration. 512 samples is a good place to start.
# Increasing the number of samples can improve accuracy.

NUM_CALIBRATION_SAMPLES = args.calib_size
MAX_SEQUENCE_LENGTH = 2048
dampening_frac=args.dampening_frac
actorder_arg=args.actorder
group_size=args.group_size

# Load dataset and preprocess.
ds = load_dataset(
    DATASET_ID,
    DATASET_SUBSET,
    split=f"{DATASET_SPLIT}[:{NUM_CALIBRATION_SAMPLES}]",
    trust_remote_code=True,
)

def preprocess(example):
    return {
        "array": example["audio"]["array"],
        "sampling_rate": example["audio"]["sampling_rate"],
        "text": " " + example["text"].capitalize(),
    }

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

# Process inputs.
def process(sample):
    inputs = processor(
        audio=sample["array"],
        sampling_rate=sample["sampling_rate"],
        text=sample["text"],
        add_special_tokens=True,
        return_tensors="pt",
    )

    inputs["input_features"] = inputs["input_features"].to(dtype=model.dtype)
    inputs["decoder_input_ids"] = inputs["labels"]
    del inputs["labels"]

    return inputs

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

# Define a oneshot data collator for multimodal inputs.
def data_collator(batch):
    assert len(batch) == 1
    return {key: torch.tensor(value) for key, value in batch[0].items()}

ignore=["lm_head"]

# Recipe
recipe = GPTQModifier(
    targets="Linear",
    config_groups={
        "config_group": QuantizationScheme(
            targets=["Linear"],
            weights=QuantizationArgs(
                num_bits=4,
                type=QuantizationType.INT,
                strategy=QuantizationStrategy.GROUP,
                group_size=group_size,
                symmetric=True,
                dynamic=False,
                actorder=getattr(ActivationOrdering, actorder_arg.upper()),
            ),
        ),
    },
    sequential_targets=["WhisperEncoderLayer", "WhisperDecoderLayer"],
    ignore=["re:.*lm_head"],
    update_size=NUM_CALIBRATION_SAMPLES,
    dampening_frac=dampening_frac
)

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


# Save to disk compressed.
save_name = f"{model_id.split('/')[-1]}-quantized.w4a16"
save_path = os.path.join(args.save_dir, save_name)
print("Saving model:", save_path)
model.save_pretrained(save_path, save_compressed=True)
processor.save_pretrained(save_path)

Librispeech:

lmms-eval \
    --model=whisper_vllm \
    --model_args="pretrained=neuralmagic-ent/whisper-large-v3-turbo-quantized.w4a16" \
    --batch_size 64 \
    --output_path <output_file_path> \
    --tasks librispeech

Fleurs:

lmms-eval \
    --model=whisper_vllm \
    --model_args="pretrained=neuralmagic-ent/whisper-large-v3-turbo-quantized.w4a16" \
    --batch_size 64 \
    --output_path <output_file_path> \
    --tasks fleurs