Article: https://proton.me/blog/deepseek
Calls it “Deepsneak”, failing to make it clear that the reason people love Deepseek is that you can download and it run it securely on any of your own private devices or servers - unlike most of the competing SOTA AIs.
I can’t speak for Proton, but the last couple weeks are showing some very clear biases coming out.
Those are not deepseek R1. They are unrelated models like llama3 from Meta or Qwen from Alibaba “distilled” by deepseek.
This is a common method to smarten a smaller model from a larger one.
Ollama should have never labelled them deepseek:8B/32B. Way too many people misunderstood that.
The 1.5B/7B/8B/13B/32B/70B models are all officially DeepSeek R1 models, that is what DeepSeek themselves refer to those models as. It is DeepSeek themselves who produced those models and released them to the public and gave them their names. And their names are correct, it is just factually false to say they are not DeepSeek R1 models. They are.
The “R1” in the name means “reasoning version one” because it does not just spit out an answer but reasons through it with an internal monologue. For example, here is a simple query I asked DeepSeek R1 13B:
However, on top of its answer, I can expand an option to see its internal monologue it went through before generating the answer, which you can find the internal monologue here because it’s too long to paste.
What makes these consumer-oriented models different is that that rather than being trained on raw data, they are trained on synthetic data from pre-existing models. That’s what the “Qwen” or “Llama” parts mean in the name. The 7B model is trained on synthetic data produced by Qwen, so it is effectively a compressed version of Qen. However, neither Qwen nor Llama can “reason,” they do not have an internal monologue.
This is why it is just incorrect to claim that something like DeepSeek R1 7B Qwen Distill has no relevance to DeepSeek R1 but is just a Qwen model. If it’s supposedly a Qwen model, why is it that it can do something that Qwen cannot do but only DeepSeek R1 can? It’s because, again, it is a DeepSeek R1 model, they add the R1 reasoning to it during the distillation process as part of its training. (I think they use the original R1 to produce the data related to the internal monologue which it is learns to copy.)
You got that backwards. They’re other models - qwen or llama - fine-tuned on synthetic data generated by Deepseek-R1. Specifically, reasoning data, so that they can learn some of its reasoning ability.
But the base model - and so the base capability there - is that of the corresponding qwen or llama model. Calling them “Deepseek-R1-something” doesn’t change what they fundamentally are, it’s just marketing.
I’m running deepseek-r1:14b-qwen-distill-fp16 locally and it produces really good results I find. Like yeah it’s a reduced version of the online one, but it’s still far better than anything else I’ve tried running locally.
Have you compared it with the regular qwen? It was sissy very good
The main difference is speed and memory usage. Qwen is a full-sized, high-parameter model while qwen-distill is a smaller model created using knowledge distillation to mimic qwen’s outputs. If you have the resources to run qwen fast then I’d just go with that.
Its so cute when chinese is sprinkled in randomly hehe my little bilingual robot in my pc