premmotgi

icon indicating a website link www.premmotgi.com

icon company https://tools.id.vodafone.com/git/customerid/cid-openid-gw.git icon location Santa Clara icon location Software Development Engineer (Java EE8, Spring Boot, Multithreading, JavaScript, HTML, CSS, Blockchain) Currently pursuing Masters of Science in Computer Scien

Results 3 issues of premmotgi

Finetuning Llama3.1 70B Model OOM error with LoRa technique

### System Info ```shell When I use the k8s sample example for lora for llama3 8B model it works fine. But for 70b model it fails with OOM. Total number...

bug

Tokenizer error: eos_token_id not found error - Incorrect assignment of variables

Incorrect assignment: https://github.com/huggingface/optimum-habana/blob/feb6545202ef6812445bbca680eb3ad50ffe5aa8/examples/language-modeling/run_lora_clm.py#L704 When running a model customization for 8B and 70B with customized dataset such as databricks-dolly-15k. I get error related to Tokenizer class does not have add_eos_token_id variable...

How to disable Checkpointing for Full tuning or PEFT runs?

4 comment

I am trying to run single GPU to multinode distributed fine tuning for Llama3-70B and Llama3 8B Models. Below is my training configuration: SFT (Llama3 8B & 70B) Epochs: 3...

enhancement