h2o-llmstudio icon indicating copy to clipboard operation
h2o-llmstudio copied to clipboard

H2O LLM Studio - a framework and no-code GUI for fine-tuning LLMs. Documentation: https://docs.h2o.ai/h2o-llmstudio/

Results 157 h2o-llmstudio issues
Sort by recently updated
recently updated
newest added
trafficstars

We currently only directly support the push to HF for the GUI via: https://github.com/h2oai/h2o-llmstudio/blob/main/app_utils/sections/experiment.py#L1596 When using the CLI, there is no dedicated util function. We should prepare a notebook and/or...

area/core

### 🐛 Bug I am getting the warning below and the nightly Docker image doesn't see my GPU. I have RTX 3090 with Driver Version: 470.182.03 CUDA Version: 11.4 on...

type/bug

### 🐛 Bug Setting both LORA and FSDP options to true while fine tuning results in > ValueError: FlatParameter requires uniform dtype but got torch.float16 and torch.float32 ### To Reproduce...

type/bug

### 🐛 Bug My experiment fails pretty early, with the following stacktrace upon BLEU evaluation: ``` 2023-05-18 17:58:52,287 - INFO: Validation BLEU: 0.32177 2023-05-18 17:58:52,333 - ERROR: Exception occurred during...

type/bug

### 🚀 Feature I'd like to experiment with fine tuning small T5 based models, but it looks like there are some assumptions made in the code about the type of...

type/feature

### 🐛 Bug Starting a new experiment from `cfg.yaml` causes an error if number of gpu specified in `cfg.yaml` exceeds the number of gpus on the target machine. Within the...

type/bug

### 🔧 Proposed code refactoring Redirect stdout logging from huggingface download to the logging module Something on these line could solve it (untested): ```python import logging from transformers import logging...

area/core

### 🚀 Feature Replace Summary tab with a model card can be the same model card as being published ### Motivation More useful than the rarely used experiment summary

type/feature

The current plotting functionality could need some rework, some suggestions: - For training plotting, I would just plot the whole sample, and (color)mark the labels. - For validation insights we...

area/core

This PR: - Removes the casting to float32 in case of LORA + float16 - Adds a new mode for pushing to HF which first loads the model on CPU...