This model was obtained by quantizing the weights and activations of swiss-ai/Apertus-70B-2509 to FP8 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 are quantized.

Deployment

Use with vLLM

Initialize vLLM server:

vllm serve RedHatAI/Apertus-70B-Instruct-2509-quantized.w4a16

Send requests to the server:

python

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/Apertus-70B-Instruct-2509-quantized.w4a16"

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

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

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

Creation

This model was created with llm-compressor by running the code snippet below.

Model Creation Code

python

from llmcompressor.modifiers.quantization import QuantizationModifier
from llmcompressor.transformers import oneshot
from transformers import AutoModelForCausalLM, AutoTokenizer

# Load model
model_stub = "swiss-ai/Apertus-70B-Instruct-2509"
model_name = model_stub.split("/")[-1]

model = AutoModelForCausalLM.from_pretrained(model_stub, dtype="auto")

tokenizer = AutoTokenizer.from_pretrained(model_stub)

# Configure the quantization algorithm and scheme
recipe = QuantizationModifier(
    ignore=["lm_head"],
    targets="Linear",
    scheme="FP8_dynamic",
)

# Apply quantization
oneshot(
    model=model,
    recipe=recipe,
)

# 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 OpenLLM Leaderboard V1, using the following command:

Evaluation Commands

OpenLLM Leaderboard V1:

lm_eval \
  --model vllm \
  --model_args pretrained="RedHatAI/Apertus-70B-Instruct-2509-quantized.w4a16",dtype=auto,add_bos_token=True,max_model_len=4096,tensor_parallel_size=2,gpu_memory_utilization=0.2,enable_chunked_prefill=True \
  --tasks openllm \
  --write_out \
  --batch_size auto \
  --output_path output_dir \
  --show_config

Accuracy

Category	Metric	swiss-ai/Apertus-70B-Instruct-2509	RedHatAI/Apertus-70B-Instruct-2509-quantized.w4a16	Recovery (%)
OpenLLM V1	ARC-Challenge (Acc-Norm, 25-shot)	70.82	70.65	99.8
GSM8K (Strict-Match, 5-shot)	73.69	73.45	99.7
HellaSwag (Acc-Norm, 10-shot)	86.23	85.67	99.4
MMLU (Acc, 5-shot)	69.21	68.25	98.6
TruthfulQA (MC2, 0-shot)	60.31	60.55	100.4
Winogrande (Acc, 5-shot)	80.74	80.03	99.1
Average Score	73.50	73.10	99.5