Model Architecture: Qwen/Qwen3-235B-A22B-Instruct-2507
- Input: Text
- Output: Text
Model Optimizations:
- Weight quantization: FP4
- Activation quantization: FP4
Out-of-scope: Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in languages other than English.
Release Date: 10/29/2025
Version: 1.0
Model Developers: RedHatAI

This model is a quantized version of Qwen/Qwen3-235B-A22B-Instruct-2507. It was evaluated on a several tasks to assess the its quality in comparison to the unquatized model.

Model Optimizations

This model was obtained by quantizing the weights and activations of Qwen/Qwen3-235B-A22B-Instruct-2507 to FP4 data type, ready for inference with vLLM>=0.9.1 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 and activations of the linear operators within transformers blocks are quantized using LLM Compressor.

Deployment

Use with vLLM

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

python

from vllm import LLM, SamplingParams
from transformers import AutoTokenizer

model_id = "RedHatAI/Qwen3-235B-A22B-Instruct-2507-NVFP4"
number_gpus = 1

sampling_params = SamplingParams(temperature=0.6, top_p=0.9, max_tokens=256)

tokenizer = AutoTokenizer.from_pretrained(model_id)

messages = [
    {"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"},
    {"role": "user", "content": "Who are you?"},
]

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

This model was created by applying LLM Compressor with calibration samples from UltraChat, as presented in the code snipet below.

python

from datasets import load_dataset

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

from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer
from llmcompressor.modeling.prepare import replace_modules_for_calibration

MODEL_ID = "Qwen/Qwen3-235B-A22B-Instruct-2507"

 #Load model.
model = AutoModelForCausalLM.from_pretrained(
    MODEL_ID, device_map=None, torch_dtype="auto"
)
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
print(model)


tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)

DATASET_ID = "HuggingFaceH4/ultrachat_200k"
DATASET_SPLIT = "train_sft"

NUM_CALIBRATION_SAMPLES = 256
MAX_SEQUENCE_LENGTH = 1024

# --- Replace MoE modules for calibration ---
model = replace_modules_for_calibration(model, calibrate_all_experts=False)

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

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


ds = ds.map(preprocess)

# Tokenize inputs.
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)

recipe = QuantizationModifier(
    targets="Linear",
    scheme="NVFP4",
    ignore=["re:.*lm_head.*", "re:.*mlp.gate$", "re:.*self_attn",
],
)

# Save to disk in compressed-tensors format.
SAVE_DIR = MODEL_ID.rstrip("/").split("/")[-1] + "-NVFP4"

# Apply quantization.
oneshot(
    model=model,
    dataset=ds,
    recipe=recipe,
    max_seq_length=MAX_SEQUENCE_LENGTH,
    num_calibration_samples=NUM_CALIBRATION_SAMPLES,
    output_dir=SAVE_DIR,
    pipeline="sequential",
    sequential_targets=["Qwen3MoeDecoderLayer"],
    calibrate_moe_context=True,

)
# Save to disk in compressed-tensors format.
model.save_pretrained(SAVE_DIR, save_compressed=True)
tokenizer.save_pretrained(SAVE_DIR)

Evaluation

This model was evaluated on the well-known OpenLLM v1, OpenLLM v2 and HumanEval_64 benchmarks using lm-evaluation-harness. The Reasoning evals were done using ligheval.

Accuracy

Category	Metric	Qwen/Qwen3-235B-A22B-Instruct-2507	RedHatAI/Qwen3-235B-A22B-Instruct-2507-NVFP4 (this model)	Recovery
OpenLLM V1	arc_challenge	72.78	72.27	99.29
gsm8k	89.92	90.37	100.50
hellaswag	87.77	87.54	99.74
mmlu	87.48	87.08	99.54
truthfulqa_mc2	62.10	60.89	98.06
winogrande	82.95	81.69	98.47
Average	80.50	79.97	101.43
OpenLLM V2	BBH	68.3	69.05	101.01%
MMLU-Pro	63.51	63.33	99.72
MuSR	45.11	45.77	101.46
IFEval	90.17	91.01	100.93
GPQA	34.56	35.49	102.7
Average	60.33	60.93	101.00
Reasoning	GPQA (Diamond, 0-shot)	69.19	64.65	93.44
Math-500 (0-shot)	94.20	89.20	94.69
Average	81.70	76.93	94.16
Coding	HumanEval_64 pass@2	96.67	96.46	99.78

Reproduction

The results were obtained using the following commands:

lm_eval \
  --model vllm \
  --model_args pretrained="RedHatAI/Qwen3-235B-A22B-Instruct-2507-NVFP4",dtype=auto,max_model_len=4096,tensor_parallel_size=2,enable_chunked_prefill=True,enforce_eager=True\
  --apply_chat_template \
  --fewshot_as_multiturn \
  --tasks openllm \
  --batch_size auto

OpenLLM v2

lm_eval \
  --model vllm \
  --model_args pretrained="RedHatAI/Qwen3-235B-A22B-Instruct-2507-NVFP4",dtype=auto,max_model_len=4096,tensor_parallel_size=2,enable_chunked_prefill=True,enforce_eager=True\
  --apply_chat_template \
  --fewshot_as_multiturn \
  --tasks leaderboard \
  --batch_size auto

HumanEval_64

lm_eval \
  --model vllm \
  --model_args pretrained="RedHatAI/Qwen3-235B-A22B-Instruct-2507-NVFP4",dtype=auto,max_model_len=4096,tensor_parallel_size=2,enable_chunked_prefill=True,enforce_eager=True\
  --apply_chat_template \
  --fewshot_as_multiturn \
  --tasks humaneval_64_instruct \
  --batch_size auto