amd/DeepSeek-R1-0528-mtp-ptpc

Note that the MTP layers of this model are also PTPC-quantized.

Model Overview

• Model Architecture: DeepSeek-R1-0528
  • Input: Text
  • Output: Text
• Supported Hardware Microarchitecture: AMD MI350/MI355
• ROCm: 7.0
• Operating System(s): Linux
• Inference Engine: SGLang/vLLM
• Model Optimizer: AMD-Quark (V0.10)
  • Weight quantization: Perchannel, FP8E4M3, Static
  • Activation quantization: Pertoken, FP8E4M3, Dynamic
• Calibration Dataset: Pile

This model was built with deepseek-ai DeepSeek-R1-0528 model by applying AMD-Quark for FP8E4M3 PTPC quantization.

Model Quantization

The model was quantized from deepseek-ai/DeepSeek-R1-0528 using AMD-Quark. The weights are quantized to FP8 and activations are quantized to FP8.

Preprocessing requirement:

Before executing the quantization script below, the original FP8 model must first be dequantized to BFloat16. You can either perform the dequantization manually using this conversion script, or use the pre-converted BFloat16 model available at unsloth/DeepSeek-R1-0528-BF16. You need to manually modify the transformers so that it can load the MTP layer. You can also directly use our modified model amd/DeepSeek-R1-0528-BF16 to perform quantitative analysis.

Quantization scripts:

# pip install amd-quark

from transformers import AutoTokenizer, AutoModelForCausalLM
from quark.torch import ModelQuantizer, export_safetensors
from quark.torch.quantization import FP8E4M3PerChannelSpec
from quark.torch.quantization.config.config import Config, QuantizationConfig

ckpt_path = "amd/DeepSeek-R1-0528-BF16"
exclude_layers = ["lm_head","*mlp.gate", "model.layers.61.eh_proj", "model.layers.61.shared_head.head"]
output_dir = ckpt_path.rstrip("/").split("/")[-1] + "-ptpc"

# Load the original floating-point model
model = AutoModelForCausalLM.from_pretrained(ckpt_path, device_map="auto", torch_dtype="auto", trust_remote_code=True)
model.eval()
tokenizer = AutoTokenizer.from_pretrained(ckpt_path)

# Set the quantization configuration
FP8_PER_CHANNEL_SPEC = FP8E4M3PerChannelSpec(is_dynamic=False, ch_axis=0).to_quantization_spec()
FP8_PER_TOKEN_DYNAMIC_SPEC = FP8E4M3PerChannelSpec(is_dynamic=True, ch_axis=1).to_quantization_spec()
W_FP8_PER_CHANNEL_STATIC_A_FP8_PER_TOKEN_DYNAMIC_CONFIG = QuantizationConfig(input_tensors=FP8_PER_TOKEN_DYNAMIC_SPEC, weight=FP8_PER_CHANNEL_SPEC)
quant_config = Config(global_quant_config=W_FP8_PER_CHANNEL_STATIC_A_FP8_PER_TOKEN_DYNAMIC_CONFIG, exclude=exclude_layers)

# Apply quantization
quantizer = ModelQuantizer(quant_config)
model = quantizer.quantize_model(model)

# Export quantized model
model = quantizer.freeze(model)
export_safetensors(model, output_dir)
tokenizer.save_pretrained(output_dir)

Deployment

This model can be deployed efficiently using the vLLM backends.

License

Modifications Copyright(c) 2025 Advanced Micro Devices, Inc. All rights reserved.