Abed khooli

Ktrain uses a forked version with tf.keras and TF2.x, so these won't work. I tried with one less epoch and got it highlighted correctly! 

I am keeping a workflow open to test. Using Chrome 77 (latest). Refresh had no effect. Signed in with Google account (still logged in). Error (repeated several times, some ws...

Re-ran `postgres=# sudo ckan datastore set-permissions | sudo -u postgres psql --set ON_ERROR_STOP=1` and it executed silently. Does that mean it failed? Also, if I `curl -X GET "http://127.0.0.1:5000/api/3/action/datastore_search?resource_id=_table_metadata"` I...

Does the above fix work with Ckan 2.10? The extension installed OK but server won't start if views are added to plugins (ex. resource_proxy shp_view). ckantoolkit doesn't seem to be...

For this task, Colab and Kaggle can only help in data preparation (may be not Kaggle if output is limited to 5 GB). For actual training, more compute resources are...

This is what I know of so far (copied as is from a notebook): ``` ### datasets: https://dl.acm.org/doi/abs/10.1145/2911451.2914677 (https://sites.google.com/view/arabicweb16) # http://opus.nlpl.eu/ # https://www.kaggle.com/linuxscout/tashkeela # https://traces1.inria.fr/oscar/ shuffled by line # https://github.com/zaidalyafeai/ARBML#datasets...

I also see same on a model I quantized using bnb (4bits). I suspect it is related to the model itself although it is based on Llama-2. Base model is...

@danielhanchen yes. I feel Unsloth quantization is different than BitsAndBytes and this could cause that. I get much better output with Qwen2 although the above models should be better.

This is a copy of my code using Qwen and it worked with Unsloth: ``` import torch from transformers import AutoModelForCausalLM, BitsAndBytesConfig from transformers import AutoTokenizer quantization_config = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_quant_type="nf4",...

@iterniter Did you test it on Colab? Do you mean w/out ludwig in first line? Still error with `torch._custom_ops` missing.