Deepseek R1 Distill Qwen 32B Quantized.w8a8

🚀 DeepSeek-R1-Distill-Qwen-32B-quantized.w8a8

A quantized version of DeepSeek-R1-Distill-Qwen-32B, optimized for reduced GPU memory usage and increased compute throughput.

🚀 Quick Start

This quantized model offers significant advantages in terms of memory usage and computational efficiency. You can quickly deploy it using the vLLM backend, as demonstrated in the usage examples below.

✨ Features

• Quantized Model: The weights and activations of this model are quantized to the INT8 data type, reducing GPU memory requirements by approximately 50% and increasing matrix - multiply compute throughput by approximately 2x.
• Efficient Deployment: Can be efficiently deployed using the vLLM backend, which also supports OpenAI - compatible serving.
• Good Performance: Achieves high accuracy on various benchmarks and up to 1.8x speedup in single - stream deployment and up to 2.2x speedup in multi - stream asynchronous deployment.

📦 Installation

To use this model, you need to have the necessary libraries installed. You can install them using pip:

pip install transformers vllm llmcompressor

💻 Usage Examples

Basic Usage

from transformers import AutoTokenizer
from vllm import LLM, SamplingParams

number_gpus = 1
model_name = "neuralmagic/DeepSeek-R1-Distill-Qwen-32B-quantized.w8a8"

tokenizer = AutoTokenizer.from_pretrained(model_name)
sampling_params = SamplingParams(temperature=0.6, max_tokens=256, stop_token_ids=[tokenizer.eos_token_id])
llm = LLM(model=model_name, tensor_parallel_size=number_gpus, trust_remote_code=True)

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)

Advanced Usage

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

# Load model
model_stub = "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B"
model_name = model_stub.split("/")[-1]

num_samples = 1024
max_seq_len = 8192

tokenizer = AutoTokenizer.from_pretrained(model_stub)

model = AutoModelForCausalLM.from_pretrained(
    model_stub,
    device_map="auto",
    torch_dtype="auto",
)

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

ds = load_dataset("neuralmagic/LLM_compression_calibration", split="train")
ds = ds.map(preprocess_fn)

# Configure the quantization algorithm and scheme
recipe = [
    SmoothQuantModifier(smoothing_strength=0.8),
    QuantizationModifier(
        targets="Linear",
        scheme="W8A8",
        ignore=["lm_head"],
        dampening_frac=0.01,
    ),
]

# Apply quantization
oneshot(
    model=model,
    dataset=ds, 
    recipe=recipe,
    max_seq_length=max_seq_len,
    num_calibration_samples=num_samples,
)

# Save to disk in compressed-tensors format
save_path = model_name + "-quantized.w8a8"
model.save_pretrained(save_path)
tokenizer.save_pretrained(save_path)
print(f"Model and tokenizer saved to: {save_path}")

📚 Documentation

Model Overview

This model was obtained by quantizing the weights and activations of DeepSeek-R1-Distill-Qwen-32B to the INT8 data type. Only the weights and activations of the linear operators within transformers blocks are quantized. Weights are quantized using a symmetric per - channel scheme, whereas activations are quantized using a symmetric per - token scheme. The GPTQ algorithm is applied for quantization, as implemented in the llm - compressor library.

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/DeepSeek-R1-Distill-Qwen-32B-quantized.w8a8",dtype=auto,max_model_len=4096,tensor_parallel_size=1,enable_chunked_prefill=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/DeepSeek-R1-Distill-Qwen-32B-quantized.w8a8",dtype=auto,max_model_len=4096,tensor_parallel_size=1,enable_chunked_prefill=True \
  --apply_chat_template \
  --fewshot_as_multiturn \
  --tasks leaderboard \
  --write_out \
  --batch_size auto \
  --output_path output_dir \
  --show_config

Accuracy

Inference Performance

This model achieves up to 1.8x speedup in single - stream deployment and up to 2.2x speedup in multi - stream asynchronous deployment, depending on hardware and use - case scenario. The following performance benchmarks were conducted with vLLM version 0.7.2, and GuideLLM.

guidellm --model neuralmagic/DeepSeek-R1-Distill-Qwen-32B-quantized.w8a8 --target "http://localhost:8000/v1" --data-type emulated --data "prompt_tokens=<prompt_tokens>,generated_tokens=<generated_tokens>" --max seconds 360 --backend aiohttp_server

Single - stream performance (measured with vLLM version 0.7.2)

Use case profiles: prompt tokens / generation tokens

QPD: Queries per dollar, based on on - demand cost at [Lambda Labs](https://lambdalabs.com/service/gpu - cloud) (observed on 2/18/2025).

Multi - stream asynchronous performance (measured with vLLM version 0.7.2)

Use case profiles: prompt tokens / generation tokens

🔧 Technical Details

The quantization process is based on the GPTQ algorithm, as implemented in the llm - compressor library. Only the weights and activations of the linear operators within transformers blocks are quantized. Weights are quantized using a symmetric per - channel scheme, and activations are quantized using a symmetric per - token scheme.

📄 License

This project is licensed under the MIT License.