DeepSeek-V2.5-1210-quantized.w4a16

Model Overview

• Model Architecture: DeepSeek-V2.5-1210
  • Input: Text
  • Output: Text
• Model Optimizations:
  • Weight quantization: INT4
  • Activation quantization: None
• Release Date: 3/1/2025
• Version: 1.0
• Model Developers: Neural Magic

Quantized version of DeepSeek-V2.5-1210. It achieves an average score of 77.2 on the OpenLLM benchmark (version 1), whereas the unquantized model achieves 77.82.

Model Optimizations

This model was obtained by quantizing only the weights to INT4 data type, ready for inference with vLLM >= 0.5.2. This optimization reduces the number of bits per parameter from 16 to 4, reducing the disk size and GPU memory requirements by approximately 75%. The weights of the linear operators within transformers blocks are quantized, except the MLP routers.

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, 2
model_name = "neuralmagic-ent/DeepSeek-V2.5-1210-quantized.w4a16"
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 with the following command:

python quantize.py --model_path deepseek-ai/DeepSeek-V2.5-1210 --quant_path "output_dir" --calib_size 256 --dampening_frac 0.1 --observer minmax --actorder weight

`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.transformers.compression.helpers import calculate_offload_device_map
import torch


def parse_actorder(value):
    # Interpret the input value for --actorder
    if value.lower() == "false":
        return False
    elif value.lower() == "weight":
        return "weight"
    elif value.lower() == "group":
        raise ValueError("group not supported for TP>1 and MoEs")
    else:
        raise argparse.ArgumentTypeError("Invalid value for --actorder. Use 'group' or 'False'.")


parser = argparse.ArgumentParser()
parser.add_argument('--model_path', type=str)
parser.add_argument('--quant_path', type=str)
parser.add_argument('--num_bits', type=int, default=4)
parser.add_argument('--sequential_update', type=bool, default=True)
parser.add_argument('--calib_size', type=int, default=256)
parser.add_argument('--dampening_frac', type=float, default=0.05)
parser.add_argument('--observer', type=str, default="minmax")
parser.add_argument(
    '--actorder',
    type=parse_actorder,
    default=False,  # Default value is False
    help="Specify actorder as 'group' (string) or False (boolean)."
)

args = parser.parse_args()

device_map = calculate_offload_device_map(
    args.model_path,
    reserve_for_hessians=True,
    num_gpus=torch.cuda.device_count(),
    torch_dtype=torch.bfloat16,
    trust_remote_code=True,
)

model = SparseAutoModelForCausalLM.from_pretrained(
    args.model_path,
    device_map=device_map,
    torch_dtype=torch.bfloat16,
    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)

quant_scheme = QuantizationScheme(
    targets=["Linear"],
    weights=QuantizationArgs(
        num_bits=args.num_bits,
        type=QuantizationType.INT,
        symmetric=True,
        group_size=128,
        strategy=QuantizationStrategy.GROUP,
        observer=args.observer,
        actorder=args.actorder
    ),
    input_activations=None,
    output_activations=None,
)

recipe = [
    GPTQModifier(
        targets=["Linear"],
        ignore=["lm_head", "re:.*\.mlp\.gate$"],
        sequential_update=args.sequential_update,
        dampening_frac=args.dampening_frac,
        config_groups={"group_0": quant_scheme},
    )
]
oneshot(
    model=model,
    dataset=ds,
    recipe=recipe,
    num_calibration_samples=args.calib_size,
)

# Save to disk compressed.
SAVE_DIR = args.quant_path
model.save_pretrained(SAVE_DIR, save_compressed=True, skip_compression_stats=True)
tokenizer.save_pretrained(SAVE_DIR)

Evaluation

The model was evaluated on OpenLLM Leaderboard V1 using the following command:

OpenLLM Leaderboard V1:

lm_eval \
  --model vllm \
  --model_args pretrained="neuralmagic-ent/DeepSeek-V2.5-1210-quantized.w4a16",dtype=float16,add_bos_token=True,max_model_len=4096,tensor_parallel_size=4,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

Accuracy

OpenLLM Leaderboard V1 evaluation scores

Metric	deepseek-ai/DeepSeek-V2.5-1210	neuralmagic-ent/DeepSeek-V2.5-1210-quantized.w4a16
ARC-Challenge (Acc-Norm, 25-shot)	72.61	72.18
GSM8K (Strict-Match, 5-shot)	88.25	88.10
HellaSwag (Acc-Norm, 10-shot)	85.01	83.48
MMLU (Acc, 5-shot)	79.60	78.36
TruthfulQA (MC2, 0-shot)	57.18	57.01
Winogrande (Acc, 5-shot)	84.29	84.06
Average Score	77.82	77.20
Recovery	100.00	99.20