iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
DeepSeek-V3’s Multi-Token Prediction (MTP) extends its training objective beyond predicting only the next token: at each position, it also predicts later tokens in sequence. At inference, those extra predictions can serve as proposals in speculative decoding. DeepSeek-AI reported an 85%–90% acceptance rate for the second-token prediction and 1.8× as many tokens per second in its evaluation; those figures describe the authors’ setup, not a guaranteed speedup for every deployment.
What is Multi-Token Prediction (MTP)?
In ordinary autoregressive language modeling, a model learns to predict the next token from the tokens that came before it. DeepSeek-V3’s MTP objective broadens that prediction scope: at each position, the model predicts multiple future tokens rather than limiting supervision to the immediate next token.
The additional predictions are made sequentially. At each prediction depth, the model preserves a causal chain, so a later prediction is conditioned on the preceding context and prediction steps. This is different from the earlier MTP approach discussed in DeepSeek-AI’s DeepSeek-V3 Technical Report, which predicts additional tokens in parallel through independent output heads.
Why does DeepSeek-V3 use MTP during training?
DeepSeek-AI presents two motivations for the objective. First, asking the model to predict multiple future tokens can densify the training signal and may improve data efficiency. Second, predicting farther ahead may encourage representations that better anticipate what comes next. The authors report improved performance in their evaluations, but these are findings from their report rather than independent confirmation that MTP will improve every model or task.
#1 Best Overall
- High-Performance AI Processing: The MX3 is designed to handle the most demanding AI computer vision workloads, delivering exceptional performance and efficiency.
- Flexible Integration: The MX3 can be easily integrated into your existing systems via its M.2 M-key form factor and support for Linux operating systems.
- Energy Efficient: The MX3 is designed to provide high performance while minimizing power consumption.
- Comprehensive Software Development Kit (SDK): The MX3 is supported by a comprehensive SDK that simplifies development and deployment.
- Hardware compatability: The MX3 is compatible with the PCI-SIG M.2 M-key 2280 Specification. It can be used with the Raspberry Pi 5 with a M-key 2280 HAT.
The report describes MTP as an auxiliary training objective. Its training rationale should be kept separate from the inference-time use of extra-token predictions: a denser learning signal is not itself a runtime speedup.
How does DeepSeek-V3 use MTP for speculative decoding?
Speculative decoding uses proposed future tokens to accelerate generation: a system can check proposed tokens against the model’s normal autoregressive process and accept proposals that are valid. DeepSeek-V3’s sequential extra-token predictions can be used as those proposals, allowing the decoding system to verify more than one token at a time.
Rank #2
- ✅Powered by 26 Tera-Operations Per Second (TOPS) Hailo-8 AI Processor. 2.5W typical power consumption
- ✅Scalable, enabling simultaneous processing of multi-streams & multi-models
- ✅Enabling real-time, low latency and high-efficiency AI inferencing on the edge devices
- ✅Supports TensorFlow, TensorFlow Lite, ONNX, Keras, Pytorch frameworks
- ✅Supports Linux and Windows. Supports the temperature range of -40°C to 85°C
In the evaluation reported by DeepSeek-AI, the second-token prediction was accepted 85%–90% of the time across the generation topics they evaluated. The report also gives a throughput result of 1.8× tokens per second with speculative decoding. Both numbers belong to the authors’ evaluated system and workloads; they are not a general guarantee for other models, inference engines, hardware, or generation tasks. The technical report is available at arXiv:2412.19437.
What do the model’s parameter and training figures mean?
DeepSeek-AI’s official repository lists DeepSeek-V3 with 671 billion total parameters and 37 billion activated per token. The repository says the Hugging Face model files total 685 billion parameters because they include an additional 14 billion parameters for the MTP modules. These counts describe different scopes: the 685B figure includes MTP weights, while the 671B figure is the main model total. The repository also lists a 128K context length for the DeepSeek-V3 and DeepSeek-V3-Base checkpoints. See the official DeepSeek-V3 repository for the checkpoint details.
DeepSeek-AI reports that full training used 14.8 trillion tokens and 2.788 million H800 GPU hours. Its $5.576 million training-cost estimate assumes a rental price of $2 per H800 GPU hour and excludes prior research and ablation costs; it is an estimate, not a complete accounting of all project costs. The official DeepSeek-V3 release announcement also describes the model’s training run.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What should developers check before enabling MTP?
DeepSeek-AI’s repository notes that community support for MTP is under active development. That does not establish that MTP is available in every inference engine or works across all hardware configurations. Check the current documentation for the specific runtime and deployment you plan to use before assuming its MTP support or performance.
Quick Recap
Best Value
- Professional AI & Creator Workstation: AMD Radeon AI PRO R9700 GPU with 32GB GDDR6 is engineered for AI development, professional content creation, and compute-intensive workloads.
- Massive 32GB Memory Capacity: 32GB of GDDR6 memory on a 256-bit bus provides ample bandwidth for large AI models, 8K video editing, and complex 3D rendering.
- Advanced RDNA 4 with AI Accelerators: 64 Compute Units with 3rd Gen Ray Tracing and dedicated 2nd Gen AI Accelerators for groundbreaking AI performance and visual computing.
- Professional Blower Cooling: Efficient single blower design exhausts heat directly out of the chassis, ideal for multi-GPU workstation and server configurations.
- Enterprise-Grade Thermal Solution: Vapor chamber heatsink with industrial Honeywell PTM7950 thermal interface material ensures reliable cooling under sustained professional loads.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →
