Mistral-Small-24B-Instruct-2501-quantized.w8a8

Model Overview

• Model Architecture: Mistral-Small-24B-Instruct-2501
  • Input: Text
  • Output: Text
• Model Optimizations:
  • Weight quantization: INT8
  • Activation quantization: INT8
• Release Date: 3/1/2025
• Version: 1.0
• Model Developers: Neural Magic

Quantized version of Mistral-Small-24B-Instruct-2501.

Model Optimizations

This model was obtained by quantizing the weights and activations to INT8 data type, ready for inference with vLLM. 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

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

from transformers import AutoTokenizer
from vllm import LLM, SamplingParams

max_model_len, tp_size = 4096, 1
model_name = "neuralmagic/Mistral-Small-24B-Instruct-2501-quantized.w8a8"
tokenizer = AutoTokenizer.from_pretrained(model_name)
llm = LLM(model=model_name, tensor_parallel_size=tp_size, max_model_len=max_model_len, trust_remote_code=True)
sampling_params = SamplingParams(temperature=0.3, max_tokens=256, stop_token_ids=[tokenizer.eos_token_id])

messages_list = [
    [{"role": "user", "content": "Who are you? Please respond in pirate speak!"}],
]

prompt_token_ids = [tokenizer.apply_chat_template(messages, add_generation_prompt=True) for messages in messages_list]

outputs = llm.generate(prompt_token_ids=prompt_token_ids, sampling_params=sampling_params)

generated_text = [output.outputs[0].text for output in outputs]
print(generated_text)

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

Creation

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

python quantize.py --model_path mistralai/Mistral-Small-24B-Instruct-2501 --quant_path "output_dir" --calib_size 1024 --dampening_frac 0.1 --observer minmax

from datasets import load_dataset
from transformers import AutoTokenizer
from llmcompressor.modifiers.quantization import GPTQModifier
from llmcompressor.transformers import SparseAutoModelForCausalLM, oneshot, apply
import argparse
from compressed_tensors.quantization import QuantizationScheme, QuantizationArgs, QuantizationType, QuantizationStrategy
from llmcompressor.modifiers.smoothquant import SmoothQuantModifier


parser = argparse.ArgumentParser()
parser.add_argument('--model_path', type=str)
parser.add_argument('--quant_path', type=str)
parser.add_argument('--calib_size', type=int, default=256)
parser.add_argument('--dampening_frac', type=float, default=0.01)
parser.add_argument('--observer', type=str, default="minmax")
args = parser.parse_args()

model = SparseAutoModelForCausalLM.from_pretrained(
    args.model_path,
    device_map="auto",
    torch_dtype="auto",
    use_cache=False,
    trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained(args.model_path)


NUM_CALIBRATION_SAMPLES = args.calib_size
DATASET_ID = "garage-bAInd/Open-Platypus"
DATASET_SPLIT = "train"
ds = load_dataset(DATASET_ID, split=DATASET_SPLIT)
ds = ds.shuffle(seed=42).select(range(NUM_CALIBRATION_SAMPLES))

def preprocess(example):
    concat_txt = example["instruction"] + "\n" + example["output"]
    return {"text": concat_txt}

ds = ds.map(preprocess)

def tokenize(sample):
    return tokenizer(
        sample["text"],
        padding=False,
        truncation=False,
        add_special_tokens=True,
    )


ds = ds.map(tokenize, remove_columns=ds.column_names)

recipe = [
    SmoothQuantModifier(smoothing_strength=0.8),
    GPTQModifier(
        targets=["Linear"],
        ignore=["lm_head"],
        scheme="W8A8",
        dampening_frac=args.dampening_frac,
        observer=args.observer,
    )
]
oneshot(
    model=model,
    dataset=ds,
    recipe=recipe,
    num_calibration_samples=args.calib_size,
    max_seq_length=8192,
)

# Save to disk compressed.
SAVE_DIR = args.quant_path
model.save_pretrained(SAVE_DIR, save_compressed=True)
tokenizer.save_pretP0+r\P0+r\rained(SAVE_DIR)

Evaluation

The model was evaluated on OpenLLM Leaderboard V1 and V2, using the following commands:

OpenLLM Leaderboard V1:

lm_eval \
  --model vllm \
  --model_args pretrained="neuralmagic/Mistral-Small-24B-Instruct-2501-quantized.w8a8",dtype=auto,add_bos_token=True,max_model_len=4096,tensor_parallel_size=1,gpu_memory_utilization=0.8,enable_chunked_prefill=True,trust_remote_code=True \
  --tasks openllm \
  --write_out \
  --batch_size auto \
  --output_path output_dir \
  --show_config

OpenLLM Leaderboard V2:

lm_eval \
  --model vllm \
  --model_args pretrained="neuralmagic/Mistral-Small-24B-Instruct-2501-quantized.w8a8",dtype=auto,add_bos_token=False,max_model_len=4096,tensor_parallel_size=1,gpu_memory_utilization=0.8,enable_chunked_prefill=True,trust_remote_code=True \
  --apply_chat_template \
  --fewshot_as_multiturn \
  --tasks leaderboard \
  --write_out \
  --batch_size auto \
  --output_path output_dir \
  --show_config

Accuracy

OpenLLM Leaderboard V1 evaluation scores

Metric	mistralai/Mistral-Small-24B-Instruct-2501	nm-testing/Mistral-Small-24B-Instruct-2501-quantized.w8a8
ARC-Challenge (Acc-Norm, 25-shot)	72.18	72.70
GSM8K (Strict-Match, 5-shot)	90.14	85.67
HellaSwag (Acc-Norm, 10-shot)	85.05	85.51
MMLU (Acc, 5-shot)	80.69	78.85
TruthfulQA (MC2, 0-shot)	65.55	65.90
Winogrande (Acc, 5-shot)	83.11	79.40
Average Score	79.45	78.01
Recovery (%)	100.00	98.18

OpenLLM Leaderboard V2 evaluation scores

Metric	mistralai/Mistral-Small-24B-Instruct-2501	nm-testing/Mistral-Small-24B-Instruct-2501-quantized.w8a8
IFEval (Inst-and-Prompt Level Strict Acc, 0-shot)	73.27	69.56
BBH (Acc-Norm, 3-shot)	45.18	47.47
MMLU-Pro (Acc, 5-shot)	38.83	41.59
Average Score	52.42	52.87
Recovery (%)	100.00	100.85
Math-Hard (Exact-Match, 4-shot)	6.35	13.95
GPQA (Acc-Norm, 0-shot)	8.29	11.78
MUSR (Acc-Norm, 0-shot)	7.84	12.90

Results on Math-Hard, GPQA, and MUSR are not considred for accuracy recovery calculation because the unquantized model has close to random prediction accuracy (6.35, 8.29, 7.84) which doesn't provide a reliable baseline for recovery calculation.