Skip to content

pullrequests Search Results · repo:OptimalScale/LMFlow language:Python

Filter by

518 results
 (105 ms)

518 results

inOptimalScale/LMFlow (press backspace or delete to remove)

OptimalScale/LMFlow
fix lr to 1e-5

As per discussion, we should have the learning rate as 1e-5.
  • lxaw
  • 1
  • Opened 
    on Jul 10
  • #952

OptimalScale/LMFlow
Explicitly mention # of training tokens

Add Q/A for number of tokens that participants can use, as this was not directly mentioned in the original README.
  • lxaw
  • 1
  • Opened 
    on May 14
  • #951

OptimalScale/LMFlow
[05/14/2025 02:00 PM EDT] Update train.sh

Just to make more clear that dataset path must be entered by user. Also added a comment that project dir will show users things like loss over the finetuning process.
  • lxaw
  • 1
  • Opened 
    on May 14
  • #950

OptimalScale/LMFlow
Update README, add small example script for upload model to HF, reupload example json dataset

Changes: 1. Changed reference from LoRA to DoRA in ### Training section. 2. Added the small dataset example in ### Prepare Dataset (it was not there). 3. Added section on what hyperparameters are expected ...
  • lxaw
  • 4
  • Opened 
    on May 14
  • #949

OptimalScale/LMFlow
[05/14/2025 09:32 AM EDT]: Add commonly asked questions in our FAQ

Added commonly asked questions to the FAQ section. These are: 1. Where to ask general question -- Go to Discord 2. How to resume from checkpoint -- Use --resume_from_checkpoint 3. How to test validation ...
  • lxaw
  • 1
  • Opened 
    on May 14
  • #948

OptimalScale/LMFlow
[fix] fix eval using part of train

Fix #823 , we will merge to main later after some tests
  • wheresmyhair
  • Opened 
    on May 13
  • #947

OptimalScale/LMFlow
support load from lora checkpoint

Description Support load from lora checkpoint. Tested in #945 . How to use --resume_from_checkpoint path/to/your/lora/checkpoint-50
  • wheresmyhair
  • 1
  • Opened 
    on May 13
  • #946

OptimalScale/LMFlow
[data4elm] support resume from lora checkpoint

Description Support resume from a lora checkpoint. How to use --resume_from_checkpoint path/to/your/lora/checkpoint-50
  • wheresmyhair
  • 1
  • Opened 
    on May 7
  • #945

OptimalScale/LMFlow
[lxaw: 04/26/2025 06:59 EDT] Add text on switching branches in README

  • lxaw
  • 1
  • Opened 
    on Apr 26
  • #944

OptimalScale/LMFlow
Add comprehensive tokenization tests, update diagram, and adjust code to handle edge cases

call stack diagram for dataset
  • Yuncong-Cao
  • 3
  • Opened 
    on Apr 6
  • #943