h2o-llmstudio

h2o-llmstudio copied to clipboard

### 🚀 Feature I trained an experiment on a specific GPU of a multi-gpu machine. When selecting the `chat` tab within the experiment the model attempts to load on the...

RobMulla

[CODE IMPROVEMENT] Check default RLHF parameters

3

### 🔧 Proposed code refactoring Check if our default hyperparameters (e.g. [kl_target](https://github.com/h2oai/h2o-llmstudio/blob/main/llm_studio/python_configs/text_causal_language_modeling_config.py#L154)) are correct, see: https://github.com/lvwerra/trl/commit/b56e8b327733baa81c3ef0d6508f08e1b3e33939 and https://github.com/lvwerra/trl/issues/462 Also, RLHF training is quite unstable w.r.t. parameter choices, see e.g. issues...

maxjeblick

[FEATURE] DeepSpeed in H2O-LLMStudio

1

### 🚀 Feature Add DeepSpeed to h2o-llmstudio ### Motivation Many reports say that DeepSpeed allows us to finetune LMMs on cheaper hardware - https://www.databricks.com/blog/2023/03/20/fine-tuning-large-language-models-hugging-face-and-deepspeed.html. Any reason why h2o-llmstudio doesn't have...

binga

[BUG] Int8 finetuning throwing a type error

3

### 🐛 Bug Int8 finetuning throwing a type error I'm trying to finetune EleutherAI/pythia-2.8b-deduped model with oasst dataset on a machine with 8 V100 GPUs. Only the following parameters are...

binga

This PR add the option to change the gpu id used for the chatbot in the Settings tab. Gpu ids are starting at 1, to be in sync with the...

maxjeblick

closes #469 

pascal-pfeiffer

[FEATURE] Add ability to specify dataset problem type during data import

1

### 🚀 Feature LLM studio supports multiple problem types (Causal Modeling/ Classification), however, during data import, it expects dataset to be named exactly as the problem type, otherwise proper problem...

MartinBarus

### 🔧 Proposed code refactoring Replace `backbone.print_trainable_parameters()` by an explicit log call to propagate the information to the UI. In addition, we could think of adding it to the model...

maxjeblick

### 🚀 Feature Provide a button to pick a random validation sample from the experiment and run it in the chat window.

psinger

ValueError: invalid literal for int() with base 10: ‘Failed to initialize NVML: Unknown Error’

8

### 🐛 Bug q.app q.user q.client report_error: True q.events q.args report_error: True stacktrace Traceback (most recent call last): File “/workspace/./llm_studio/app_utils/handlers.py”, line 78, in handle await home(q) File “/workspace/./llm_studio/app_utils/sections/home.py”, line 66,...

jldroid19