# Accuracy Logging

**URL:** <https://community.wandb.ai/t/accuracy-logging/8846>\
**Category:** W&B Help\
**Tags:** wandb\
**Created:** [January 22, 2025, 12:57pm UTC](https://community.wandb.ai/t/accuracy-logging/8846 "2025-01-22T12:57:53Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![tesswatt](https://avatars.discourse-cdn.com/v4/letter/t/ac91a4/32.png) [@tesswatt](https://community.wandb.ai/u/tesswatt)\
**Post date:** [January 22, 2025, 12:57pm UTC](https://community.wandb.ai/t/accuracy-logging/8846/1 "2025-01-22T12:57:53Z")

</div>

Hi there,

I want to log my model’s accuracy after each epoch and its final accuracy at the end but I cannot find a simple way of doing this.

I am following this tutorial: [Brev.dev Console](https://console.brev.dev/launchable/deploy?userID=p2mzt91a8&orgID=jnj0c501d&launchableID=env-2hpxJ6HArVk5jzOYgJmFDJfvNmH&instance=A10G%40g5.12xlarge&diskStorage=300&cloudID=devplane-brev-1&python=3.10&cuda=12.2.2&file=https%3A%2F%2Fgithub.com%2Fbrevdev%2Fnotebooks%2Fblob%2Fmain%2Fllava-finetune.ipynb&name=Fine-tune+and+deploy+multimodal+LLaVA-1.5)

My deepspeed/wandb code is as follows:

> !deepspeed LLaVA/llava/train/train\_mem.py   
> –lora\_enable True --lora\_r 128 --lora\_alpha 256 --mm\_projector\_lr 2e-5   
> –deepspeed LLaVA/scripts/zero3.json   
> –model\_name\_or\_path liuhaotian/llava-v1.5-13b   
> –version v1   
> –data\_path ./dataset/train/dataset.json   
> –image\_folder ./dataset/images   
> –vision\_tower openai/clip-vit-large-patch14-336   
> –mm\_projector\_type mlp2x\_gelu   
> –mm\_vision\_select\_layer -2   
> –mm\_use\_im\_start\_end False   
> –mm\_use\_im\_patch\_token False   
> –image\_aspect\_ratio pad   
> –group\_by\_modality\_length True   
> –bf16 True   
> –output\_dir ./checkpoints/llava-v1.5-13b-task-lora   
> –num\_train\_epochs 10   
> –per\_device\_train\_batch\_size 16   
> –per\_device\_eval\_batch\_size 4   
> –gradient\_accumulation\_steps 1   
> –evaluation\_strategy “no”   
> –save\_strategy “steps”   
> –save\_steps 50000   
> –save\_total\_limit 1   
> –learning\_rate 2e-4   
> –weight\_decay 0.   
> –warmup\_ratio 0.03   
> –lr\_scheduler\_type “cosine”   
> –logging\_steps 1   
> –tf32 True   
> –model\_max\_length 2048   
> –gradient\_checkpointing True   
> –dataloader\_num\_workers 4   
> –lazy\_preprocess True   
> –report\_to wandb

I have tried using:

> metric\_for\_best\_model “accuracy”  
> evaluation\_strategy “epoch”  
> evaluation\_strategy “steps”  
> do\_eval True

But so far nothing is working.

Any help or advice would be appreciated.

---

<div class="post-metadata">

**Author:** ![tesswatt](https://avatars.discourse-cdn.com/v4/letter/t/ac91a4/32.png) [@tesswatt](https://community.wandb.ai/u/tesswatt)\
**Post date:** [February 13, 2025, 4:40pm UTC](https://community.wandb.ai/t/accuracy-logging/8846/2 "2025-02-13T16:40:28Z")

</div>

Hi there, any help or advice on this?