RedHatAI/diffusiongemma-26B-A4B-it-NVFP4

This model is an NVFP4 quantized version of google/diffusiongemma-26B-A4B-it. The model has both weights and activations quantized to NVFP4 using vllm/llm-compressor and in the compressed-tensors format. It was evaluated on several tasks to assess its quality in comparison to the unquantized model using vLLM.

Deployment

VLLM_USE_V2_MODEL_RUNNER=1
vllm serve  RedHatAI/diffusiongemma-26B-A4B-it-NVFP4 \
    --trust-remote-code \
    --attention-backend TRITON_ATTN \
    --max-num-seqs 4 \
    --hf-overrides '{"diffusion_sampler": "entropy_bound", "diffusion_entropy_bound": 0.1}' \
    --default-chat-template-kwargs '{"enable_thinking": true}'

Creation

"""
Quantize DiffusionGemma to NVFP4 using LLM Compressor v0.11.0

Model: google/diffusiongemma-26B-A4B-it
- Total parameters: ~25.8B
- Expert parameters: 22.8B (88.4%)
- Non-expert parameters: 3.0B (11.6%)

Note: This will require a local update to transformers to support the model definition.
"""

import torch
from compressed_tensors.offload import dispatch_model
from datasets import load_dataset
from transformers import AutoProcessor
from transformers.models.diffusion_gemma import DiffusionGemmaForBlockDiffusion
from compressed_tensors.offload import dispatch_model

from llmcompressor import oneshot
from llmcompressor.modeling.diffusion_gemma4 import (  # noqa: F401
    CalibrationDiffusionGemmaTextExperts,
)
from llmcompressor.modifiers.quantization import QuantizationModifier

# Load model
MODEL_ID = "google/diffusiongemma-26B-A4B-it"

model = DiffusionGemmaForBlockDiffusion.from_pretrained(
    MODEL_ID, dtype="auto", trust_remote_code=True
)
processor = AutoProcessor.from_pretrained(MODEL_ID, trust_remote_code=True)

# CalibrationDiffusionGemmaTextExperts replaces the original
# DiffusionGemmaTextExperts class during calibration to:
# 1. Linearize the 3D expert tensors into individual nn.Linear modules
# 2. Ensure all experts are properly calibrated, even those not activated
#    for certain tokens during calibration

# Configure the quantization scheme
# NVFP4 (4-bit weights, 4-bit activations) for Linear layers
recipe = QuantizationModifier(
    targets="Linear",
    scheme="NVFP4",
    ignore=[
        "lm_head",
        "re:.*embed.*",
        "re:.*self_attn",
        "re:.*router",
        "re:.*vision_tower.*",
        "re:.*self_conditioning.*",
    ],
)

DATASET_ID = "neuralmagic/calibration"
NUM_CALIBRATION_SAMPLES = 256
MAX_SEQUENCE_LENGTH = 4096


ds = load_dataset(DATASET_ID, name="LLM", split=f"train[:{NUM_CALIBRATION_SAMPLES}]")


def preprocess_function(example):
    messgages = []
    for message in example["messages"]:
        messgages.append(
            {
                "role": message["role"],
                "content": [{"type": "text", "text": message["content"]}],
            }
        )

    return processor.apply_chat_template(
        messgages,
        return_tensors="pt",
        padding=False,
        truncation=True,
        max_length=MAX_SEQUENCE_LENGTH,
        tokenize=True,
        add_special_tokens=False,
        return_dict=True,
        add_generation_prompt=False,
    )


ds = ds.map(preprocess_function, batched=False, remove_columns=ds.column_names)


def data_collator(batch):
    assert len(batch) == 1
    return {
        key: (
            torch.tensor(value)
            if key != "pixel_values"
            else torch.tensor(value, dtype=torch.bfloat16).squeeze(0)
        )
        for key, value in batch[0].items()
    }


# Apply quantization with calibration data
oneshot(
    model=model,
    recipe=recipe,
    dataset=ds,
    max_seq_length=MAX_SEQUENCE_LENGTH,
    num_calibration_samples=NUM_CALIBRATION_SAMPLES,
    data_collator=data_collator,
    sequential_targets=[
        "DiffusionGemmaDecoderTextLayer", 
        "DiffusionGemmaEncoderTextLayer",  
    ],
)   

# Test sample generation
print("========== SAMPLE GENERATION ==============")
dispatch_model(model)

# "The reason the sky is blue is because" + chat template
input_ids = torch.tensor(
    [[
        2, 105, 2364, 107, 818, 3282, 506, 7217, 563, 3730, 563,
        1547, 106, 107, 105, 4368, 107
    ]]
).to(model.device)

output = model.generate(
    input_ids,
    max_new_tokens=100,
    max_denoising_steps=48,
)
print(processor.tokenizer.decode(output[0]))
print("==========================================\n\n")

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

Accuracy

The following metrics were generated when serving the quantized model with vLLM on a single B200 GPU.

Benchmark	google/diffusiongemma-26B-A4B-it	RedHatAI/diffusiongemma-26B-A4B-it-NVFP4	Recovery (%)
AIME 2025	0.437	0.427	97.7%
GPQA Diamond	0.641	0.644	100.5%
IFEval	0.879	0.866	98.5%
GSM8K	0.943	0.943	100.0%
MMLU 0-Shot	0.539	0.616	114.3%
Thinking
AIME 2025	0.650	0.637	98.0%
GPQA Diamond	0.698	0.677	97.0%
GSM8K	0.951	0.952	100.1%