H2OTest / documentation /docs /guide /experiments /experiment-settings.md
elineve's picture
Upload 301 files
07423df
|
raw
history blame
12 kB
metadata
description: All the settings needed for creating an experiment are explored in this page.

import GeneralSettingsDataset from '../../tooltips/experiments/_dataset.mdx'; import GeneralSettingsProblemType from '../../tooltips/experiments/_problem-type.mdx'; import GSImportConfigFromYaml from '../../tooltips/experiments/_import-config-from-yaml.mdx'; import GSExperimentName from '../../tooltips/experiments/_experiment-name.mdx'; import GSLLMBackbone from '../../tooltips/experiments/_llm-backbone.mdx'; import DSTrainDataframe from '../../tooltips/experiments/_train-dataframe.mdx'; import DSvalidationStrategy from '../../tooltips/experiments/_validation-strategy.mdx'; import DSvalidationSize from '../../tooltips/experiments/_validation-size.mdx'; import DSdataSample from '../../tooltips/experiments/_data-sample.mdx'; import DSpromptColumn from '../../tooltips/experiments/_prompt-column.mdx'; import DSsystemColumn from '../../tooltips/experiments/_system-column.mdx'; import DSanswerColumn from '../../tooltips/experiments/_answer-column.mdx'; import DSparentIdColumn from '../../tooltips/experiments/_parent-id-column.mdx'; import DStextPromptStart from '../../tooltips/experiments/_text-prompt-start.mdx'; import DStextAnswerSeparator from '../../tooltips/experiments/_text-answer-separator.mdx'; import DSadaptiveKlControl from '../../tooltips/experiments/_adaptive-kl-control.mdx'; import DSaddEosTokentoprompt from '../../tooltips/experiments/_add-eos-token-to-prompt.mdx'; import DSaddEosTokentoanswer from '../../tooltips/experiments/_add-eos-token-to-answer.mdx'; import DSmaskPromptlabels from '../../tooltips/experiments/_mask-prompt-labels.mdx'; import TSmaxLengthPrompt from '../../tooltips/experiments/_max-length-prompt.mdx'; import TSmaxLengthAnswer from '../../tooltips/experiments/_max-length-answer.mdx'; import TSmaxLength from '../../tooltips/experiments/_max-length.mdx'; import TSaddpromptanswertokens from '../../tooltips/experiments/_add-prompt-answer-tokens.mdx'; import TSpaddingQuantile from '../../tooltips/experiments/_padding-quantile.mdx'; import TSuseFast from '../../tooltips/experiments/_use-fast.mdx'; import ASBackboneDtype from '../../tooltips/experiments/_backbone-dtype.mdx'; import ASGradientcheckpointing from '../../tooltips/experiments/_gradient-checkpointing.mdx'; import ASforceEmbeddingGradients from '../../tooltips/experiments/_force-embedding-gradients.mdx'; import ASintermediateDropout from '../../tooltips/experiments/_intermediate-dropout.mdx'; import ASpretrainedWeights from '../../tooltips/experiments/_pretrained-weights.mdx'; import TSoptimizer from '../../tooltips/experiments/_optimizer.mdx'; import TSlossfunction from '../../tooltips/experiments/_loss-function.mdx'; import TSlearningRate from '../../tooltips/experiments/_learning-rate.mdx'; import TSuseflashattention2 from '../../tooltips/experiments/_use-flash-attention-2.mdx'; import TSbatchSize from '../../tooltips/experiments/_batch-size.mdx'; import TSepochs from '../../tooltips/experiments/_epochs.mdx'; import TSschedule from '../../tooltips/experiments/_schedule.mdx'; import TSwarmupEpochs from '../../tooltips/experiments/_warmup-epochs.mdx'; import TSweightDecay from '../../tooltips/experiments/_weight-decay.mdx'; import TSGradientclip from '../../tooltips/experiments/_gradient-clip.mdx'; import TSgradAccumulation from '../../tooltips/experiments/_grad-accumulation.mdx'; import TSlora from '../../tooltips/experiments/_lora.mdx'; import TSloraR from '../../tooltips/experiments/_lora-r.mdx'; import TSloraAlpha from '../../tooltips/experiments/_lora-alpha.mdx'; import TSloraDropout from '../../tooltips/experiments/_lora-dropout.mdx'; import TSloraTargetModules from '../../tooltips/experiments/_lora-target-modules.mdx'; import TSsavebestcheckpoint from '../../tooltips/experiments/_save-best-checkpoint.mdx'; import TSevaluationepochs from '../../tooltips/experiments/_evaluation-epochs.mdx'; import TSevaluationbeforetraining from '../../tooltips/experiments/_evaluate-before-training.mdx'; import TStrainvalidationdata from '../../tooltips/experiments/_train-validation-data.mdx'; import TSuseRHLF from '../../tooltips/experiments/_use-rlhf.mdx'; import TSrewardModel from '../../tooltips/experiments/_reward-model.mdx'; import TSinitialKlCoefficient from '../../tooltips/experiments/_initial-kl-coefficient.mdx'; import TSklTarget from '../../tooltips/experiments/_kl-target.mdx'; import TSklHorizon from '../../tooltips/experiments/_kl-horizon.mdx'; import TSadvantagesGamma from '../../tooltips/experiments/_advantages-gamma.mdx'; import TSadvantagesLambda from '../../tooltips/experiments/_advantages-lambda.mdx'; import TSppoClipPolicy from '../../tooltips/experiments/_ppo-clip-policy.mdx'; import TSppoClipValue from '../../tooltips/experiments/_ppo-clip-value.mdx'; import TSscalingFactorValueLoss from '../../tooltips/experiments/_scaling-factor-value-loss.mdx'; import TSppoEpochs from '../../tooltips/experiments/_ppo-epochs.mdx'; import TSppoBatchSize from '../../tooltips/experiments/_ppo-batch-size.mdx'; import TSppoGenerateTemp from '../../tooltips/experiments/_ppo-generate-temperature.mdx'; import TSoffloadRewardModel from '../../tooltips/experiments/_offload-reward-model.mdx'; import AStokenmaskprobability from '../../tooltips/experiments/_token-mask-probability.mdx'; import ASskipParentprobability from '../../tooltips/experiments/_skip-parent-probability.mdx'; import ASrandomparentprobability from '../../tooltips/experiments/_random-parent-probability.mdx'; import ASneftunenoisealpha from '../../tooltips/experiments/_neftune_noise_alpha.mdx'; import PSmetric from '../../tooltips/experiments/_metric.mdx'; import PSmetricgptmodel from '../../tooltips/experiments/_metric-gpt-model.mdx'; import PSmetricgpttemplate from '../../tooltips/experiments/_metric-gpt-template.mdx'; import PSminlengthinference from '../../tooltips/experiments/_min-length-inference.mdx'; import PSmaxlengthinference from '../../tooltips/experiments/_max-length-inference.mdx'; import PSbatchsizeinference from '../../tooltips/experiments/_batch-size-inference.mdx'; import PSdosample from '../../tooltips/experiments/_do-sample.mdx'; import PSnumbeams from '../../tooltips/experiments/_num-beams.mdx'; import PStemperature from '../../tooltips/experiments/_temperature.mdx'; import PSrepetitionpenalty from '../../tooltips/experiments/_repetition-penalty.mdx'; import PSstoptokens from '../../tooltips/experiments/_stop-tokens.mdx'; import PStopk from '../../tooltips/experiments/_top-k.mdx'; import PStopp from '../../tooltips/experiments/_top-p.mdx'; import ESgpus from '../../tooltips/experiments/_gpus.mdx'; import ESmixedprecision from '../../tooltips/experiments/_mixed-precision.mdx'; import EScompilemodel from '../../tooltips/experiments/_compile-model.mdx'; import ESfindunusedparameters from '../../tooltips/experiments/_find-unused-parameters.mdx'; import EStrustremotecode from '../../tooltips/experiments/_trust-remote-code.mdx'; import EShuggingfacebranch from '../../tooltips/experiments/_huggingface-branch.mdx'; import ESnumofworkers from '../../tooltips/experiments/_number-of-workers.mdx'; import ESseed from '../../tooltips/experiments/_seed.mdx'; import LSlogger from '../../tooltips/experiments/_logger.mdx'; import LSneptuneproject from '../../tooltips/experiments/_neptune-project.mdx';

Experiment settings

The settings for creating an experiment are grouped into the following sections:

The settings under each category are listed and described below.

General settings

Dataset

Problem type

Import config from YAML

Experiment name

LLM backbone

Dataset settings

Train dataframe

Validation strategy

Validation size

Data sample

System column

Prompt column

Answer column

Parent ID column

Text prompt start

Text answer separator

Adaptive Kl control

Add EOS token to prompt

Add EOS token to answer

Mask prompt labels

Tokenizer settings

Max length prompt

Max length answer

Max length

Add prompt answer tokens

Padding quantile

Use fast

Architecture settings

Backbone Dtype

Gradient Checkpointing

Force Embedding Gradients

Intermediate dropout

Pretrained weights

Training settings

Loss function

Optimizer

Learning rate

Use Flash Attention 2

Batch size

Epochs

Schedule

Warmup epochs

Weight decay

Gradient clip

Grad accumulation

Lora

Lora R

Lora Alpha

Lora dropout

Lora target modules

Save best checkpoint

Evaluation epochs

Evaluate before training

Train validation data

Use RLHF

Reward model

Adaptive KL control

Initial KL coefficient

KL target

KL Horizon

Advantages gamma

Advantages Lambda

PPO clip policy

PPO clip value

Scaling factor value loss

PPO epochs

PPO Batch Size

PPO generate temperature

Offload reward model

Augmentation settings

Token mask probability

Skip parent probability

Random parent probability

Neftune noise alpha

Prediction settings

Metric

Metric GPT model

Metric GPT template

Min length inference

Max length inference

Batch size inference

Do sample

Num beams

Temperature

Repetition penalty

Stop tokens

Top K

Top P

Environment settings

GPUs

Mixed precision

Compile model

Find unused parameters

Trust remote code

Huggingface branch

Number of workers

Seed

Logging settings

Logger

Neptune project