Qwen3-Next-80B-A3B-Instruct-quantized.w8a8

Model Overview

• Model Architecture: Qwen3NextForCausalLM
  • Input: Text
  • Output: Text
• Model Optimizations:
  • Weight quantization: INT8
  • Activation quantization: INT8
• Release Date:
• Version: 1.0
• Model Developers:: Red Hat

Quantized version of Qwen/Qwen3-Next-80B-A3B-Instruct.

Model Optimizations

This model was obtained by quantizing the weights and activations of Qwen/Qwen3-Next-80B-A3B-Instruct to INT8 data type. This optimization reduces the number of bits per parameter from 16 to 8, reducing the disk size and GPU memory requirements by approximately 50%. Only the weights and activations of the linear operators within transformers blocks of the language model are quantized.

Deployment

Use with vLLM

1. Initialize vLLM server:

vllm serve RedHatAI/Qwen3-Next-80B-A3B-Instruct-quantized.w8a8 --tensor_parallel_size 2

2. Send requests to the server:

from openai import OpenAI

# Modify OpenAI's API key and API base to use vLLM's API server.
openai_api_key = "EMPTY"
openai_api_base = "http://<your-server-host>:8000/v1"

client = OpenAI(
    api_key=openai_api_key,
    base_url=openai_api_base,
)

model = "RedHatAI/Qwen3-Next-80B-A3B-Instruct-quantized.w8a8"

messages = [
    {"role": "user", "content": "Explain quantum mechanics clearly and concisely."},
]


outputs = client.chat.completions.create(
    model=model,
    messages=messages,
)

generated_text = outputs.choices[0].message.content
print(generated_text)

Creation

This model was quantized using the llm-compressor library as shown below.

Creation details

from datasets import load_dataset
from transformers import AutoModelForCausalLM, AutoTokenizer

from llmcompressor import oneshot
from llmcompressor.modifiers.quantization import QuantizationModifier
from llmcompressor.utils import dispatch_for_generation

# NOTE: Requires a minimum of transformers 4.57.0

MODEL_ID = "Qwen/Qwen3-Next-80B-A3B-Instruct"


# Select calibration dataset.
DATASET_ID = "garage-bAInd/Open-Platypus"
DATASET_SPLIT = "train"

# Select number of samples. 512 samples is a good place to start.
# Increasing the number of samples can improve accuracy.
NUM_CALIBRATION_SAMPLES = 1024
MAX_SEQUENCE_LENGTH = 8192

# Load model.
model = AutoModelForCausalLM.from_pretrained(MODEL_ID, torch_dtype="auto")
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)

# Configure the quantization algorithm and scheme.
# In this case, we:
#   * quantize the weights to int8 with per channel via ptq
#   * quantize the activations to int8 with dynamic per token
recipe = QuantizationModifier(
    targets="Linear", scheme="W8A8", ignore=[
        "lm_head",
        "re:.*mlp.gate$",
        "re:.*mlp.shared_expert_gate$",
        "re:.*linear_attn.*",
    ],
)

# Load calibration dataset.
ds = load_dataset(DATASET_ID, split=f"{DATASET_SPLIT}[:{NUM_CALIBRATION_SAMPLES}]")
ds = ds.shuffle(seed=42)

def preprocess(example):
    messages = [
        {"role": "user", "content": example["instruction"]},
        {"role": "assistant", "content": example["output"]},
    ]
    return {
        "text": tokenizer.apply_chat_template(
            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)

# Apply quantization.
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("========== SAMPLE GENERATION ==============")
dispatch_for_generation(model)
input_ids = tokenizer("Hello my name is", return_tensors="pt").input_ids.to(
    model.device
)
output = model.generate(input_ids, max_new_tokens=20)
print(tokenizer.decode(output[0]))
print("==========================================")

# Save to disk in compressed-tensors format.
SAVE_DIR = MODEL_ID.rstrip("/").split("/")[-1] + "-quantized.w8a8"
model.save_pretrained(SAVE_DIR)
tokenizer.save_pretrained(SAVE_DIR)

Evaluation

The model was evaluated on the AIME25, GPQA Diamond and Mathh 500 benchmarks using lighteval, and on MMLU-Pro, IFEval and GSM8k using lm-evaluation-harness. In all cases vLLM is used as the backend. All results were averaged over 6 repetitions with different random seeds.

Evaluation commands

Start vLLM server

vllm serve RedHatAI/Qwen3-Next-80B-A3B-Instruct-quantized.w8a8 --tensor_parallel_size 2

lm-evaluation-harness

lm_eval --model local-chat-completions \
  --tasks mmlu_pro_chat \
  --model_args "model=RedHatAI/Qwen3-Next-80B-A3B-Instruct-quantized.w8a8,max_length=262144,base_url=http://0.0.0.0:8000/v1/chat/completions,num_concurrent=128,max_retries=3,tokenized_requests=False,timeout=600,tokenizer_backend=None" \
  --apply_chat_template \
  --num_fewshot 5 \
  --fewshot_as_multiturn \
  --output_path mmlu_pro_qwen3_next_w8a8 \
  --gen_kwargs "do_sample=True,temperature=0.7,top_p=0.8,top_k=20,max_gen_toks=16000"

lm_eval --model local-chat-completions \
  --tasks ifeval \
  --model_args "model=RedHatAI/Qwen3-Next-80B-A3B-Instruct-quantized.w8a8,max_length=262144,base_url=http://0.0.0.0:8000/v1/chat/completions,num_concurrent=128,max_retries=3,tokenized_requests=False,timeout=600,tokenizer_backend=None" \
  --apply_chat_template \
  --output_path ifeval_qwen3_next_w8a8 \
  --gen_kwargs "do_sample=True,temperature=0.7,top_p=0.8,top_k=20,max_gen_toks=16000"

lm_eval --model local-chat-completions \
  --tasks gsm8k \
  --model_args "model=RedHatAI/Qwen3-Next-80B-A3B-Instruct-quantized.w8a8,max_length=262144,base_url=http://0.0.0.0:8000/v1/chat/completions,num_concurrent=128,max_retries=3,tokenized_requests=False,timeout=600,tokenizer_backend=None" \
  --apply_chat_template \
  --num_fewshot 5 \
  --fewshot_as_multiturn \
  --output_path gsm8k_qwen3_next_w8a8 \
  --gen_kwargs "do_sample=True,temperature=0.7,top_p=0.8,top_k=20,max_gen_toks=16000"

lighteval

litellm_config.yaml

model_parameters:
  provider: "hosted_vllm"
  model_name: "hosted_vllm/RedHatAI/Phi-4-reasoning-FP8-dynamic"
  base_url: "http://0.0.0.0:8000/v1"
  api_key: ""
  timeout: 600
  concurrent_requests: 128
  generation_parameters:
    temperature: 0.7
    top_k: 20
    top_p: 0.8
    max_new_tokens: 16000

lighteval endpoint litellm litellm_config.yaml \
    gpqa:diamond|0,math_500|0,aime25|0 \
    --output-dir qwen3_next_w8a8 \
    --save-details

Accuracy

Benchmark	Qwen3-Next-80B-A3B-Instruct	Qwen3-Next-80B-A3B-Instruct-quantized.w8a8 (this model)	Recovery
AIME25	62.78	65.00	103.5%
GPQA Diamond	74.58	75.17	100.8%
Math 500	89.73	90.57	100.9%
MMLU-Pro	78.62	78.85	100.3%
IFEval	91.45	91.51	100.1%
GSM8k	69.71	69.74	100.0%