DeepSeek-V4-Flash-0731#

  • Context Length: 1048576

  • Model Name: DeepSeek-V4-Flash-0731

  • Languages: en, zh

  • Abilities: chat, reasoning, hybrid, tools

  • Description: Official DeepSeek-V4-Flash release with enhanced agentic capabilities and an attached DSpark speculative decoding module.

Specifications#

Model Spec 1 (fp8, 304 Billion)#

  • Model Format: fp8

  • Model Size (in billions): 304

  • Quantizations: fp8

  • Engines: vLLM

  • Model ID: deepseek-ai/DeepSeek-V4-Flash-0731

  • Model Hubs: Hugging Face, ModelScope

Execute the following command to launch the model, remember to replace ${quantization} with your chosen quantization method from the options listed above:

xinference launch --model-engine ${engine} --model-name DeepSeek-V4-Flash-0731 --size-in-billions 304 --model-format fp8 --quantization ${quantization}