undefined | Better HN

0 pointsaddandsubtract1y ago0 comments

If you have ollama installed and 404GB free space: ollama pull deepseek-r1:671b

0 comments

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 ... ?

cyberax1y ago

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

CamperBob21y ago

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.

rsync1y ago

  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 ...

2030ai1y ago

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.

j / k navigate · click thread line to collapse

0 comments

rsync1y ago

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 ... ?

cyberax1y ago

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

CamperBob21y ago

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.

rsync1y ago

  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 ...

2030ai1y ago

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.

j / k navigate · click thread line to collapse