Maxim Smirnov
Maxim Smirnov
How about now? While not as easy as writing `model.to(torch.device('tpu'))`, it's definitely not in alpha anymore. There are computer vision models one could train and run easily on colab/kaggle TPUs,...
> Components using old apis > > 1. rest-api > 2. concept-blending > 3. socketio (not been used for anything afaik) AFAIK, socketio is used in a publisher (https://github.com/opencog/opencog/blob/715d1831e3c8f0d0dc4cfdc7bb2aad0e34d558bd/opencog/python/web/socketio/atomspace_publisher.py). The...
If you use a list of `torch.LongTensor.zero_()`s instead, it seems to work.
No amount of encryption will save you from such people.
I'm looking into it. It seem to be an issue with flash attention and not vicuna/fastchat as such
If I replace `bf16 True` with `fp16 True` in the script args and also add `"fp16": {"enabled": true}` to my deepspeed config, the error changes to `RuntimeError: element 0 of...
> not sure yet but just stumbled on this [kyegomez/LongNet](https://github.com/kyegomez/LongNet) Just starting to look at it now. Great attempt, but it claims to be a simplified version. Then again, a...
I figured as much after a re-reading of the respective parts of the paper, but the whole "they encode them as if they were at the beginning of the local...
Oh. It seems like you mean the single file you'd get from running `llama.donwload` from pyllama. Let me try it out...
The first time the model is loaded from `./models`, there are indeed no checkpoints there, but they can be downloaded with the python or bash script from pyllama. As for...