Teknium

When I prompted Alpaca with ~800 tokens I got this error: Token indices sequence length is longer than the specified maximum sequence length for this model (521 > 512). Running...

Someone told me there is a deepspeed training option in the code, can I ask why its not the default? Do we know if it's far faster, and if so,...

What is the "odd train/loss" that is referenced to be caused by this flag? I have it on (and havent tested with it off) and my train/loss looks like this:...

Dataset for evol 250k?

4

I very much hope the dataset will be released, any info on this?

I do not understand how to run human eval

1

What do I do..

When I use scaleAI's deployment endpoint, if the generated text doesnt get sent back within ~30s, I get an error about a 504 error. If the response comes back quickly,...

https://github.com/OpenAccess-AI-Collective/axolotl/pull/441 ```File "/home/teknium/evaltest/axolotl/venv/lib/python3.10/site-packages/transformers/trainer_callback.py", line 388, in on_evaluate return self.call_event("on_evaluate", args, state, control, metrics=metrics) File "/home/teknium/evaltest/axolotl/venv/lib/python3.10/site-packages/transformers/trainer_callback.py", line 406, in call_event result = getattr(callback, event)( File "/home/teknium/evaltest/axolotl/src/axolotl/utils/callbacks.py", line 166, in on_evaluate (loss,...

I thought nvcc wasnt required @ 2.0.7 but is again?

22

```(venv) teknium@DESKTOP-UTIDMJ3:~/lm-evaluation-harness$ pip install flash-attn Collecting flash-attn Using cached flash_attn-2.1.1.tar.gz (2.3 MB) Preparing metadata (setup.py) ... error error: subprocess-exited-with-error × python setup.py egg_info did not run successfully. │ exit code:...