-
Notifications
You must be signed in to change notification settings - Fork 49
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[Model] Add the Infinity-Instruct SFT code #278
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Remove the train_
prefix from the file names since they are in the train
directory.
examples/qwen/dist_start.sh
Outdated
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We have updated the launcher to use the unified run.py
. Please remove dist_start.sh
, dist_stop.sh
, env.sh
, and run.sh
.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
These files (dist_start.sh, dist_stop.sh, env.sh, run.sh and the directories of tokenizers) have been removed.
flagscale/train/train_aquila.py
Outdated
total_tokens = loss_mask.sum() | ||
loss = torch.cat([torch.sum(losses.view(-1) * loss_mask).view(1), total_tokens.view(1)]) | ||
|
||
loss = torch.cat([torch.sum(torch.masked_select(losses.view(-1) , loss_mask==1)).view(1), total_tokens.view(1)]) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This loss will also be used in pre-training. If the SFT requires a different one, we may need a better way to distinguish them.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I add a new file called train_aquila_sft.py to distinguish the loss.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Please rename conf_qwen
to conf
since this folder is already in the parent qwen
folder.
In /conf/train/qwen_2.5_1.5b.yaml, adding ckpt_format, ckpt_convert_format and ckpt_convert_save to convert checkpoints .
Have renamed conf_qwen. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
No description provided.