Hugging Face's logo

RedHatAI
/
Qwen3-30B-A3B-quantized.w4a16

Text Generation
Transformers
Safetensors
qwen3_moe
neuralmagic
redhat
llmcompressor
quantized
INT4
conversational
compressed-tensors
Model card Files Community
Qwen3-30B-A3B-quantized.w4a16 / README.md
alexmarques's picture
alexmarques
Create README.md
1c842a5 verified
preview code
|
raw
history blame contribute delete
9.28 kB
metadata
library_name: transformers
license: apache-2.0
pipeline_tag: text-generation
base_model:
  - Qwen/Qwen3-30B-A3B
tags:
  - neuralmagic
  - redhat
  - llmcompressor
  - quantized
  - INT4

Qwen3-30B-A3B-quantized.w4a16

Model Overview

  • Model Architecture: Qwen3ForCausalLM
    • Input: Text
    • Output: Text
  • Model Optimizations:
    • Weight quantization: INT4
  • Intended Use Cases:
    • Reasoning.
    • Function calling.
    • Subject matter experts via fine-tuning.
    • Multilingual instruction following.
    • Translation.
  • Out-of-scope: Use in any manner that violates applicable laws or regulations (including trade compliance laws).
  • Release Date: 05/05/2025
  • Version: 1.0
  • Model Developers: RedHat (Neural Magic)

Model Optimizations

This model was obtained by quantizing the weights of Qwen3-30B-A3B to INT4 data type. This optimization reduces the number of bits per parameter from 16 to 4, reducing the disk size and GPU memory requirements by approximately 75%.

Only the weights of the linear operators within transformers blocks are quantized. Weights are quantized using a symmetric per-group scheme, with group size 128. The GPTQ algorithm is applied for quantization, as implemented in the llm-compressor library.

Deployment

This model can be deployed efficiently using the vLLM backend, as shown in the example below.

from vllm import LLM, SamplingParams
from transformers import AutoTokenizer

model_id = "RedHatAI/Qwen3-30B-A3B-quantized.w4a16"
number_gpus = 1
sampling_params = SamplingParams(temperature=0.6, top_p=0.95, top_k=20, min_p=0, max_tokens=256)

messages = [
    {"role": "user", "content": prompt}
]

tokenizer = AutoTokenizer.from_pretrained(model_id)

messages = [{"role": "user", "content": "Give me a short introduction to large language model."}]

prompts = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)

llm = LLM(model=model_id, tensor_parallel_size=number_gpus)

outputs = llm.generate(prompts, sampling_params)

generated_text = outputs[0].outputs[0].text
print(generated_text)

vLLM aslo supports OpenAI-compatible serving. See the documentation for more details.

Creation

Creation details This model was created with [llm-compressor](https://github.com/vllm-project/llm-compressor) by running the code snippet below. 
from llmcompressor.modifiers.quantization import GPTQModifier
from llmcompressor.transformers import oneshot
from transformers import AutoModelForCausalLM, AutoTokenizer

# Load model
model_stub = "Qwen/Qwen3-30B-A3B"
model_name = model_stub.split("/")[-1]

num_samples = 1024
max_seq_len = 8192

model = AutoModelForCausalLM.from_pretrained(model_stub)

tokenizer = AutoTokenizer.from_pretrained(model_stub)

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

ds = load_dataset("neuralmagic/LLM_compression_calibration", split="train")
ds = ds.map(preprocess_fn)

# Configure the quantization algorithm and scheme
recipe = GPTQModifier(
    ignore: ["lm_head", "re:.*gate$"]
    sequential_targets=["Qwen3DecoderLayer"],
    targets="Linear",
    scheme="W4A16",
    dampening_frac=0.01,
)

# Apply quantization
oneshot(
    model=model,
    dataset=ds, 
    recipe=recipe,
    max_seq_length=max_seq_len,
    num_calibration_samples=num_samples,
)

# Save to disk in compressed-tensors format
save_path = model_name + "-quantized.w4a16"
model.save_pretrained(save_path)
tokenizer.save_pretrained(save_path)
print(f"Model and tokenizer saved to: {save_path}")

Evaluation

The model was evaluated on the OpenLLM leaderboard tasks (versions 1 and 2), using lm-evaluation-harness, and on reasoning tasks using lighteval. vLLM was used for all evaluations.

Evaluation details

lm-evaluation-harness

lm_eval \
  --model vllm \
  --model_args pretrained="RedHatAI/Qwen3-30B-A3B-quantized.w4a16",dtype=auto,gpu_memory_utilization=0.5,max_model_len=8192,enable_chunk_prefill=True,tensor_parallel_size=1 \
  --tasks openllm \
  --apply_chat_template\
  --fewshot_as_multiturn \
  --batch_size auto

lm_eval \
  --model vllm \
  --model_args pretrained="RedHatAI/Qwen3-30B-A3B-quantized.w4a16",dtype=auto,gpu_memory_utilization=0.5,max_model_len=8192,enable_chunk_prefill=True,tensor_parallel_size=1 \
  --tasks mgsm \
  --apply_chat_template\
  --batch_size auto

lm_eval \
  --model vllm \
  --model_args pretrained="RedHatAI/Qwen3-30B-A3B-quantized.w4a16",dtype=auto,gpu_memory_utilization=0.5,max_model_len=16384,enable_chunk_prefill=True,tensor_parallel_size=1 \
  --tasks leaderboard \
  --apply_chat_template\
  --fewshot_as_multiturn \
  --batch_size auto

lighteval

lighteval_model_arguments.yaml

model_parameters:
  model_name: RedHatAI/Qwen3-30B-A3B-quantized.w4a16
  dtype: auto
  gpu_memory_utilization: 0.9
  max_model_length: 40960
  generation_parameters:
    temperature: 0.6
    top_k: 20
    min_p: 0.0
    top_p: 0.95
    max_new_tokens: 32768

lighteval vllm \
  --model_args lighteval_model_arguments.yaml \
  --tasks lighteval|aime24|0|0 \
  --use_chat_template = true

lighteval vllm \
  --model_args lighteval_model_arguments.yaml \
  --tasks lighteval|aime25|0|0 \
  --use_chat_template = true

lighteval vllm \
  --model_args lighteval_model_arguments.yaml \
  --tasks lighteval|math_500|0|0 \
  --use_chat_template = true

lighteval vllm \
  --model_args lighteval_model_arguments.yaml \
  --tasks lighteval|gpqa:diamond|0|0 \
  --use_chat_template = true

lighteval vllm \
  --model_args lighteval_model_arguments.yaml \
  --tasks extended|lcb:codegeneration \
  --use_chat_template = true

Accuracy

Category Benchmark Qwen3-30B-A3B Qwen3-30B-A3B-quantized.w4a16
(this model) 		Recovery
OpenLLM v1 MMLU (5-shot) 77.67 76.11 98.00%
ARC Challenge (25-shot) 63.40 62.97 99.3%
GSM-8K (5-shot, strict-match) 87.26 86.66 99.3%
Hellaswag (10-shot) 54.33 54.76 100.8%
Winogrande (5-shot) 66.77 64.33 96.3%
TruthfulQA (0-shot, mc2) 56.27 54.76 97.3%
Average 67.62 66.60 98.5%
OpenLLM v2 MMLU-Pro (5-shot) 47.45 45.38 95.6%
IFEval (0-shot) 86.26 84.86 98.4%
BBH (3-shot) 34.81 28.12 80.8%
Math-lvl-5 (4-shot) 52.14 56.99 109.3%
GPQA (0-shot) 0.31 0.60 ---
MuSR (0-shot) 8.09 9.05 ---
Average 38.18 37.50 98.2%
Multilingual MGSM (0-shot) 32.27 33,890 104.8%
Reasoning
(generation) 		AIME 2024 78.33 78.54 100.3%
AIME 2025 71.46 70.31 98.4%
GPQA diamond 62.63 62.12 99.2%
Math-lvl-5 97.60 97.20 99.6%
LiveCodeBench 60.66 58.75 96.9%