Sanyam Bhutani comments

Results 18 comments of


                                            Sanyam Bhutani

Llama2 on EC2 tutorial using Runhouse.

@rohinb2 Since the repo has been updated for 3.1 Model(s)-would you be interested in updating your PR for the same please?

freezing layers have differenct behaves for different models

@hjc3613 please let us know if you would be interested in contributing with a PR! For now, I will close the issue since its not on the current roadmap

Is it possible to restart finetuning from an existing checkpoint?

Closing as solved!

Clarification regarding best practice for pad_token

Closing because of no response. However, @ShengYun-Peng please feel free to send a PR still if you have the time and interest! Thanks!

Function or Tool calling

@avianion Thanks for the feedback above! The 3.1 model(s) support function calling out of the box, you can find more details on [MODEL CARD](https://llama.meta.com/docs/model-cards-and-prompt-formats/llama3_1) page, please re-open issue incase you...

NCCL communicator error: Socket Timeout when finetuning 70B model on 2 * (8* A100(80G))

@tginart please let us know if you would still be interested in sharing some more details for us to repro. Thanks!

When fine-tuning the model, the loss is greater than 1

Could you kindly try our updated Fine-tunning scripts and re-open if you still face the same issue?

Update llama_guard_version in inference.py

Closing since the repo has been updated but please feel free to update the PR if you are still interested in contributing!

BUG2

Closing but please re-open if you are interested in contributing to any feature, would be happy to help you @Phani1609. Thanks!

Training running out of memory on 1st backward pass of 2nd epoch

Please feel free to re-open if you still have an issue!