from llmcompressor.transformers import SparseAutoModelForCausalLM
from transformers import AutoTokenizer
import torch
MODEL_ID = "/root/autodl-tmp/Mistral-Nemo-Instruct-2407"
model = SparseAutoModelForCausalLM.from_pretrained(
MODEL_ID, device_map="auto", torch_dtype="auto",
)

tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)

from datasets import load_dataset

NUM_CALIBRATION_SAMPLES = 2048
MAX_SEQUENCE_LENGTH = 2048

Load and preprocess the dataset

ds = load_dataset("/root/autodl-tmp/ultrachat_200k", split="train_sft")
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)

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)

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

Configure the quantization algorithms

recipe = [
SmoothQuantModifier(smoothing_strength=0.85),
GPTQModifier(targets="Linear", scheme="W8A8", ignore=["lm_head"], sequential_update=True),
]

Apply quantization

oneshot(
model=model,
dataset=ds,
recipe=recipe,
max_seq_length=MAX_SEQUENCE_LENGTH,
num_calibration_samples=NUM_CALIBRATION_SAMPLES,
)

Save the compressed model

SAVE_DIR = MODEL_ID.split("/")[1] + "-W8A8-Dynamic-Per-Token"
model.save_pretrained(SAVE_DIR, save_compressed=True)
tokenizer.save_pretrained(SAVE_DIR)

here is, I copy it from https://docs.vllm.ai/en/latest/quantization/int8.html

If you are interested in w8a8 quantization, you can also read this: https://github.com/vllm-project/llm-compressor/issues/916

Awesome! Thank you! I somehow had trouble getting w8a8 to work with Mistral models so this is helpful!