Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Genuinely curious - if one did not have any tooling at all (other than curl or wget or fetch, etc.) is there a one-liner to download only the deepseek model ?

  git clone --mirror ... ?


Yes, you need git with git lfs. Then just: `git clone https://huggingface.co/deepseek-ai/DeepSeek-R1`


You probably want the GGUF versions, though, which are the ones you can actually run on reasonable hardware.

Full R1 version here: https://huggingface.co/unsloth/DeepSeek-R1-GGUF/tree/main/De...

Various smaller versions: https://huggingface.co/unsloth/DeepSeek-R1-GGUF/tree/main

There is also a 'BF16' version on that page with 16-bit weights, but I don't understand what it is or where it came from, given that the original model is FP8-native.


  ssh user@rsync.net git lfs install
  Password:
  Git LFS initialized.
... so far so good ...

  ssh user@rsync.net git clone https://huggingface.co/deepseek-ai/DeepSeek-R1
  Password:
  Cloning into 'DeepSeek-R1'...
... success.

Done in the spirit of the Streisand Effect but I suppose the protest is watered down by the fact that this account is in Zurich ...


Ollama has underlying plain URLs but they are parts so you'd need something to combine them. But you can use curl.

Otherwise a docker one liner to run ollama should do it.

Docker is a "core util" right.




Consider applying for YC's Fall 2025 batch! Applications are open till Aug 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: