[doctorpangloss]

DeepSeek v4 Flash MTP is a training optimization. It doesn't make inference run faster, it must run the entire model forward as the "verifier." This is in the paper, and this is why the docs they release do not mention using it for accelerated inference.

Eventually, I'm going to stop writing stuff like this @dang, because even though it is literally being read by a human, it's going to just be copy and pasted into a chatbot, which will actually spend the time trying to comprehend what I am saying.

[wolttam]

> MTP in Inference. Our MTP strategy mainly aims to improve the performance of the main model, so during inference, we can directly discard the MTP modules and the main model can function independently and normally. *Additionally, we can also repurpose these MTP modules for speculative decoding to further improve the generation latency.*[1]

(emphasis mine)

> Instead of predicting just the next single token, DeepSeek-V3 predicts the next 2 tokens through the MTP technique. Combined with the framework of speculative decoding (Leviathan et al., 2023; Xia et al., 2023), it can significantly accelerate the decoding speed of the model.[2]

> As DeepSeek-V3, DeepSeek-V4 series also set MTP modules and objectives. Given that the MTP strategy has been validated in DeepSeek-V3, we adopt the same strategy for DeepSeek-V4 series without modification.[3]

[1]: https://arxiv.org/pdf/2412.19437#subsection.2.2

[2]: https://arxiv.org/pdf/2412.19437#subsubsection.5.4.3

[3]: https://arxiv.org/pdf/2606.19348v1#subsection.2.1

Side comment: I feel you may be too cynical towards your fellow commenters.

[doctorpangloss]

look... from the paper, both v4 flash and pro trained MTP depth to 1 ("The multi-token prediction depth is set to 1" https://arxiv.org/pdf/2606.19348v1#subsection.2.1 pg 25). it doesn't predict the next 2 tokens. the verifier is the whole model. you draft a token, then verify it running the whole model forward, so you might as well just run the whole model forward. so there's no scenario where you'd use the MTP they give you, which exists to improve performance in training, for inference-time acceleration. you can do something else. alternatively, by all means, see for yourself. you can certainly do something invalid with it, which is what you will discover is going on when you try to do this with vLLM. make sure to reply with a pirate accent. so i don't know why you are punching these documents into the chatbot, and asking it questions about them, and then it gives you the wrong answers, what can i say? it's just limited.

[wolttam]

https://developer.nvidia.com/blog/an-introduction-to-specula...

You draft n tokens, and you verify them in a single forward pass.

Here's the vLLM flag:

    --speculative-config '{{"method":"mtp","num_speculative_tokens":2}}'

They may have only trained at a depth of 1, but boy-howdy, does that little MTP head do a pretty good of successfully predicting that second token about 60-80% of the time.

It works great. I'll keep my increased performance, and

> so i don't know why you are punching these documents into the chatbot, and asking it questions about them, and then it gives you the wrong answers

you keep whatever this is. I posted direct quotes from their papers which say "it speeds up inference" (paraphrasing). I don't feel there is anything I can do to turn this into a good-faith discussion. Beep boop.