Quantizing a Qwen-Next model is too slow.

[dingjingzhen-bot]

Is there any way to speed up quantization? Currently, quantizing a Qwen3-Next-80B-A3B model takes over 10 hours for FP8 and more than 30 hours for W8A8. Even compression alone takes about 3 hours.

[dsikka]

Can you share your script and what details of your set-up?

[dingjingzhen-bot]

Can you share your script and what details of your set-up?

from transformers import AutoTokenizer, AutoModelForCausalLM

MODEL_ID = "Qwen/Qwen3-Next-80B-A3B-Instruct"
model = AutoModelForCausalLM.from_pretrained(MODEL_ID, torch_dtype="auto",trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID,trust_remote_code=True)
from datasets import load_dataset

NUM_CALIBRATION_SAMPLES=1024
MAX_SEQUENCE_LENGTH=2048

# Load dataset.
ds = load_dataset("HuggingFaceH4/ultrachat_200k", split=f"train_sft[:{NUM_CALIBRATION_SAMPLES}]")
ds = ds.shuffle(seed=42)

# Preprocess the data into the format the model is trained with.
def preprocess(example):
    return {"text": tokenizer.apply_chat_template(example["messages"], tokenize=False,)}
ds = ds.map(preprocess)

# Tokenize the data (be careful with bos tokens - we need add_special_tokens=False since the chat_template already added it).
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 import oneshot
from llmcompressor.modifiers.quantization import GPTQModifier
from llmcompressor.modifiers.smoothquant import SmoothQuantModifier

# Configure the quantization algorithms to run.
recipe = [
    SmoothQuantModifier(smoothing_strength=0.8),
    GPTQModifier(targets="Linear", scheme="W8A8", ignore=["lm_head", "re:.*mlp.gate$", "re:.*mlp.shared_expert_gate$","re:.*linear_attn$"]),
]

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

# Save to disk compressed.
# SAVE_DIR = MODEL_ID.rstrip("/").split("/")[-1] + "-W8A8-Dynamic-Per-Token"
SAVE_DIR = "/home/admin/Qwen3-Next-80B-A3B-Instruct-w8a8-smq"

model.save_pretrained(SAVE_DIR, save_compressed=True)
tokenizer.save_pretrained(SAVE_DIR)