grok-1
grok-1 copied to clipboard
xai-org
Grok open release
It seems as if the model is being loaded in FP16. I've also noticed how `QuantizedWeight8bit` is imported in run.py, but not actually used. Is that for runtime quantization with...
raise ValueError(f'Number of devices {len(devices)} must equal the product ' ValueError: Number of devices 1 must equal the product of mesh_shape (1, 8)
# Someone made a full huggingface implementation which is way better than mine, so use this instead! https://huggingface.co/keyfan/grok-1-hf # Previous comment ## Very rough implementation, may be broken Part of...
Adds CPU execution to grok-1 model demo VERY SLOW! No one should process real world workloads this way. This is only meant for early dev work by those who don't...
Previous issues gives: we can compile models for portable use with local gpu, but with heavy quantization that sacrifices the performance. One example is comparing GGUF of Mixtral 8x7b 4bits...
On MacOS, without quotes the command to download checkpoint files fails. ``` levneiman@Levs-MBP ~/code/grok-1 % huggingface-cli download xai-org/grok-1 --repo-type model --include ckpt-0/* --local-dir checkpoints --local-dir-use-symlinks False zsh: no matches found:...
Conda env is very easy to setup environment of python and install package quickly. More info:https://docs.conda.io/en/latest/
I installed dm_haiku==0.0.12 according to the requirements. But I met some errors when I run 'python run.py'. I found that the haiku import has some problem. Why is that? I...
Hello everyone. - [x] I placed the downloaded weights into the correct directory `\checkpoints\ckpt-0` - [x] Changed the run(main_parameters) to circumvent the mesh issue: `ValueError: Number of devices 1 must...
i want to inference by 8*A6000(50G),how can i use 8-bit?
