Skip to main content

H2O LLM Studio performance

Setting up and running H2O LLM Studio requires the following minimal prerequisites. This page lists out the speed and performance metrics of H2O LLM Studio based on different hardware setups.

The following metrics were measured.

  • Hardware setup: The type and number of computing devices used to train the model.
  • LLM backbone: The underlying architecture of the language model. For more information, see LLM backbone.
  • Quantization: A technique used to reduce the size and memory requirements of the model. For more information, see Quantization.
  • Train: The amount of time it took to train the model in hours and minutes.
  • Validation: The amount of time it took to validate the mode in hours and minutes.
Hardware setupLLM backboneQuantizationTrain (hh:mm:ss)Validation (hh:mm:ss)
8xA10Gh2oai/h2ogpt-4096-llama2-7bbfloat1611:353:32
4xA10Gh2oai/h2ogpt-4096-llama2-7bbfloat1621:1306:35
2xA10Gh2oai/h2ogpt-4096-llama2-7bbfloat1637:0412:21
1xA10Gh2oai/h2ogpt-4096-llama2-7bbfloat161:25:2915:50
8xA10Gh2oai/h2ogpt-4096-llama2-7bnf414:2606:13
4xA10Gh2oai/h2ogpt-4096-llama2-7bnf426:5511:59
2xA10Gh2oai/h2ogpt-4096-llama2-7bnf448:2423:37
1xA10Gh2oai/h2ogpt-4096-llama2-7bnf41:26:5942:17
8xA10Gh2oai/h2ogpt-4096-llama2-13bbfloat16OOMOOM
4xA10Gh2oai/h2ogpt-4096-llama2-13bbfloat16OOMOOM
2xA10Gh2oai/h2ogpt-4096-llama2-13bbfloat16OOMOOM
1xA10Gh2oai/h2ogpt-4096-llama2-13bbfloat16OOMOOM
8xA10Gh2oai/h2ogpt-4096-llama2-13bnf425:0710:58
4xA10Gh2oai/h2ogpt-4096-llama2-13bnf448:4321:25
2xA10Gh2oai/h2ogpt-4096-llama2-13bnf41:30:4542:06
1xA10Gh2oai/h2ogpt-4096-llama2-13bnf42:44:361:14:20
8xA10Gh2oai/h2ogpt-4096-llama2-70bnf4OOMOOM
4xA10Gh2oai/h2ogpt-4096-llama2-70bnf4OOMOOM
2xA10Gh2oai/h2ogpt-4096-llama2-70bnf4OOMOOM
1xA10Gh2oai/h2ogpt-4096-llama2-70bnf4OOMOOM
---------------
4xA100 80GBh2oai/h2ogpt-4096-llama2-7bbfloat167:043:55
2xA100 80GBh2oai/h2ogpt-4096-llama2-7bbfloat1613:147:23
1xA100 80GBh2oai/h2ogpt-4096-llama2-7bbfloat1623:3613:25
4xA100 80GBh2oai/h2ogpt-4096-llama2-7bnf49:446:30
2xA100 80GBh2oai/h2ogpt-4096-llama2-7bnf418:3412:16
1xA100 80GBh2oai/h2ogpt-4096-llama2-7bnf434:0621:51
4xA100 80GBh2oai/h2ogpt-4096-llama2-13bbfloat1611:465:56
2xA100 80GBh2oai/h2ogpt-4096-llama2-13bbfloat1621:5411:17
1xA100 80GBh2oai/h2ogpt-4096-llama2-13bbfloat1639:1018:55
4xA100 80GBh2oai/h2ogpt-4096-llama2-13bnf416:5110:35
2xA100 80GBh2oai/h2ogpt-4096-llama2-13bnf432:0521:00
1xA100 80GBh2oai/h2ogpt-4096-llama2-13bnf459:1136:53
4xA100 80GBh2oai/h2ogpt-4096-llama2-70bnf41:13:3346:02
2xA100 80GBh2oai/h2ogpt-4096-llama2-70bnf42:20:441:33:42
1xA100 80GBh2oai/h2ogpt-4096-llama2-70bnf44:23:572:44:51
info

The runtimes were gathered using the default parameters.

Expand to see the default parameters 
architecture:
    backbone_dtype: int4
    gradient_checkpointing: true
    intermediate_dropout: 0.0
    pretrained: true
    pretrained_weights: ''
augmentation:
    random_parent_probability: 0.0
    skip_parent_probability: 0.0
    token_mask_probability: 0.0
dataset:
    add_eos_token_to_answer: true
    add_eos_token_to_prompt: true
    add_eos_token_to_system: true
    answer_column: output
    chatbot_author: H2O.ai
    chatbot_name: h2oGPT
    data_sample: 1.0
    data_sample_choice:
    - Train
    - Validation
    limit_chained_samples: false
    mask_prompt_labels: true
    parent_id_column: None
    personalize: false
    prompt_column:
    - instruction
    system_column: None
    text_answer_separator: <|answer|>
    text_prompt_start: <|prompt|>
    text_system_start: <|system|>
    train_dataframe: /data/user/oasst/train_full.pq
    validation_dataframe: None
    validation_size: 0.01
    validation_strategy: automatic
environment:
    compile_model: false
    find_unused_parameters: false
    gpus:
    - '0'
    - '1'
    - '2'
    - '3'
    - '4'
    - '5'
    - '6'
    - '7'
    huggingface_branch: main
    mixed_precision: true
    number_of_workers: 8
    seed: -1
    trust_remote_code: true
experiment_name: default-8-a10g
llm_backbone: h2oai/h2ogpt-4096-llama2-7b
logging:
    logger: None
    neptune_project: ''
output_directory: /output/...
prediction:
    batch_size_inference: 0
    do_sample: false
    max_length_inference: 256
    metric: BLEU
    metric_gpt_model: gpt-3.5-turbo-0301
    metric_gpt_template: general
    min_length_inference: 2
    num_beams: 1
    num_history: 4
    repetition_penalty: 1.2
    stop_tokens: ''
    temperature: 0.0
    top_k: 0
    top_p: 1.0
problem_type: text_causal_language_modeling
tokenizer:
    add_prompt_answer_tokens: false
    max_length: 512
    padding_quantile: 1.0
training:
    batch_size: 2
    differential_learning_rate: 1.0e-05
    differential_learning_rate_layers: []
    drop_last_batch: true
    epochs: 1
    evaluate_before_training: false
    evaluation_epochs: 1.0
    grad_accumulation: 1
    gradient_clip: 0.0
    learning_rate: 0.0001
    lora: true
    use_dora: false
    lora_alpha: 16
    lora_dropout: 0.05
    lora_r: 4
    lora_target_modules: ''
    loss_function: TokenAveragedCrossEntropy
    optimizer: AdamW
    save_checkpoint: "last"
    schedule: Cosine
    train_validation_data: false
    warmup_epochs: 0.0
    weight_decay: 0.0
Feedback