Merge branch 'main' into on/doc-problem-types

h2oai · Sep 27, 2024 · f952547 · f952547
2 parents bd6ec02 + 8790178
commit f952547
Show file tree

Hide file tree

Showing 13 changed files with 691 additions and 679 deletions.
diff --git a/Pipfile b/Pipfile
@@ -14,7 +14,7 @@ python_version = "3.10"
 [packages]
 torch = {index = "pytorch", version = "==2.4.0+cu121"}
 tqdm = ">=4.65.0, <5.0.0"
-transformers = "==4.44.2"
+transformers = "==4.45.0"
 numpy = ">=1.26.0, <2.0.0"
 pandas = ">=2.2.0, <3.0.0"
 scikit-learn = ">=1.5.1, <2.0.0"

diff --git a/Pipfile.lock b/Pipfile.lock
diff --git a/README.md b/README.md
@@ -161,7 +161,7 @@ If you are running H2O LLM Studio with a custom environment other than Pipenv, y
 H2O_WAVE_MAX_REQUEST_SIZE=25MB \
 H2O_WAVE_NO_LOG=true \
 H2O_WAVE_PRIVATE_DIR="/download/@output/download" \
-wave run app
+wave run llm_studio.app
 ```
 
 If you are using the [nightly conda environment](#nightly-conda-virtual-environment), you can run ```make llmstudio-conda```.

diff --git a/app.toml.template b/app.toml.template
@@ -15,14 +15,12 @@ CPUReservation = "10"
 MemoryReservation = "118Gi"
 MemoryLimit = "118Gi"
 GPUCount = 1
-GPUType = "training-gpu"
 VolumeMount = "/home/llmstudio/mount"
 VolumeSize = "1Ti"
 ResourceVolumeSize = "1Ti"
 EnableSHM = true
 EnableOIDC = true
 RoutingMode = "BASE_URL"
-CustomImage = "docker.io/h2oairelease/h2oai-llmstudio-app:v{{VERSION}}"
 
 [[Env]]
 Name = "H2O_LLM_STUDIO_WORKDIR"
@@ -39,3 +37,7 @@ Value = "True"
 [[Env]]
 Name = "H2O_WAVE_PRIVATE_DIR"
 Value = "/download/@/home/llmstudio/mount/output/download"
+
+[[Env]]
+Name = "HF_HUB_ENABLE_HF_TRANSFER"
+Value = "0"
diff --git a/documentation/docs/get-started/set-up-llm-studio.md b/documentation/docs/get-started/set-up-llm-studio.md
@@ -199,7 +199,7 @@ H2O LLM Studio requires the following minimum requirements:
   H2O_WAVE_MAX_REQUEST_SIZE=25MB \ <br></br>
   H2O_WAVE_NO_LOG=True \ <br></br>
   H2O_WAVE_PRIVATE_DIR="/download/@output/download" \ <br></br>
-  wave run app
+  wave run llm_studio.app
   </code>
   </pre>
  </p>
@@ -255,7 +255,7 @@ If you are running H2O LLM Studio with a custom environment other than Pipenv, s
 H2O_WAVE_MAX_REQUEST_SIZE=25MB \
 H2O_WAVE_NO_LOG=True \
 H2O_WAVE_PRIVATE_DIR="/download/@output/download" \
-wave run app
+wave run llm_studio.app
 ```
 
 ### Run using Docker from a nightly build

diff --git a/llm_studio/app_utils/config.py b/llm_studio/app_utils/config.py
@@ -3,6 +3,7 @@
 from types import SimpleNamespace
 
 import toml
+from huggingface_hub.constants import _is_true
 
 toml_root_dir = os.path.abspath(
  os.path.join(os.path.dirname(os.path.abspath(__file__)), "../..")
@@ -43,6 +44,47 @@ def get_size(x):
 url = f"http://{host}:{port}/"
 
 
+if os.getenv("H2O_LLM_STUDIO_DEFAULT_LM_MODELS"):
+ default_causal_language_models = [
+ mdl.strip() for mdl in os.getenv("H2O_LLM_STUDIO_DEFAULT_LM_MODELS").split(",")
+ ]
+else:
+ default_causal_language_models = [
+ "h2oai/h2o-danube3-500m-base",
+ "h2oai/h2o-danube3-500m-chat",
+ "h2oai/h2o-danube3-4b-base",
+ "h2oai/h2o-danube3-4b-chat",
+ "h2oai/h2o-danube2-1.8b-base",
+ "h2oai/h2o-danube2-1.8b-chat",
+ "meta-llama/Llama-3.2-1B-Instruct",
+ "meta-llama/Llama-3.2-3B-Instruct",
+ "meta-llama/Meta-Llama-3.1-8B-Instruct",
+ "meta-llama/Meta-Llama-3.1-70B-Instruct",
+ "mistralai/Mistral-7B-v0.3",
+ "mistralai/Mistral-7B-Instruct-v0.2",
+ "google/gemma-2-2b-it",
+ "google/gemma-2-9b-it",
+ "microsoft/Phi-3-mini-4k-instruct",
+ "microsoft/Phi-3-medium-4k-instruct",
+ "Qwen/Qwen2-7B-Instruct",
+ "Qwen/Qwen2-72B-Instruct",
+ ]
+
+if os.getenv("H2O_LLM_STUDIO_DEFAULT_S2S_MODELS"):
+ default_sequence_to_sequence_models = [
+ mdl.strip() for mdl in os.getenv("H2O_LLM_STUDIO_DEFAULT_S2S_MODELS").split(",")
+ ]
+else:
+ default_sequence_to_sequence_models = [
+ "t5-small",
+ "t5-base",
+ "t5-large",
+ "google/flan-t5-small",
+ "google/flan-t5-base",
+ "google/flan-t5-large",
+ "google/flan-ul2",
+ ]
+
 default_cfg = {
  "url": url,
  "name": "H2O LLM Studio",
@@ -67,6 +109,8 @@ def get_size(x):
  "text_sequence_to_sequence_modeling_config",
  "text_dpo_modeling_config",
  ],
+ "default_causal_language_models": default_causal_language_models,
+ "default_sequence_to_sequence_models": default_sequence_to_sequence_models,
  "problem_categories": ["text"],
  "dataset_keys": [
  "train_dataframe",
@@ -117,8 +161,8 @@ def get_size(x):
  "default_wandb_project": os.getenv("WANDB_PROJECT", ""),
  "default_wandb_entity": os.getenv("WANDB_ENTITY", ""),
  "default_huggingface_api_token": os.getenv("HF_TOKEN", ""),
- "default_hf_hub_enable_hf_transfer": os.getenv(
- "HF_HUB_ENABLE_HF_TRANSFER", True
+ "default_hf_hub_enable_hf_transfer": _is_true(
+ os.getenv("HF_HUB_ENABLE_HF_TRANSFER", "1")
  ),
  "default_openai_azure": os.getenv("OPENAI_API_TYPE", "open_ai") == "azure",
  "default_openai_api_token": os.getenv("OPENAI_API_KEY", ""),

diff --git a/llm_studio/python_configs/text_causal_classification_modeling_config.py b/llm_studio/python_configs/text_causal_classification_modeling_config.py
@@ -4,6 +4,7 @@
 
 import llm_studio.src.datasets.text_causal_classification_ds
 import llm_studio.src.plots.text_causal_classification_modeling_plots
+from llm_studio.app_utils.config import default_cfg
 from llm_studio.python_configs.base import DefaultConfig, DefaultConfigProblemBase
 from llm_studio.python_configs.text_causal_language_modeling_config import (
  ConfigNLPAugmentation,
@@ -155,7 +156,11 @@ class ConfigNLPCausalClassificationLogging(ConfigNLPCausalLMLogging):
 class ConfigProblemBase(DefaultConfigProblemBase):
  output_directory: str = f"output/{os.path.basename(__file__).split('.')[0]}"
  experiment_name: str = field(default_factory=generate_experiment_name)
- llm_backbone: str = "h2oai/h2o-danube3-500m-chat"
+ llm_backbone: str = (
+ "h2oai/h2o-danube3-500m-chat"
+ if "h2oai/h2o-danube3-500m-chat" in default_cfg.default_causal_language_models
+ else default_cfg.default_causal_language_models[0]
+ )
 
  dataset: ConfigNLPCausalClassificationDataset = field(
  default_factory=ConfigNLPCausalClassificationDataset
@@ -188,24 +193,7 @@ def __post_init__(self):
  self._visibility["output_directory"] = -1
 
  self._possible_values["llm_backbone"] = possible_values.String(
- values=(
- "h2oai/h2o-danube3-500m-base",
- "h2oai/h2o-danube3-500m-chat",
- "h2oai/h2o-danube3-4b-base",
- "h2oai/h2o-danube3-4b-chat",
- "h2oai/h2o-danube2-1.8b-base",
- "h2oai/h2o-danube2-1.8b-chat",
- "meta-llama/Meta-Llama-3.1-8B-Instruct",
- "meta-llama/Meta-Llama-3.1-70B-Instruct",
- "mistralai/Mistral-7B-v0.3",
- "mistralai/Mistral-7B-Instruct-v0.2",
- "google/gemma-2-2b-it",
- "google/gemma-2-9b-it",
- "microsoft/Phi-3-mini-4k-instruct",
- "microsoft/Phi-3-medium-4k-instruct",
- "Qwen/Qwen2-7B-Instruct",
- "Qwen/Qwen2-72B-Instruct",
- ),
+ values=default_cfg.default_causal_language_models,
  allow_custom=True,
  )
 

diff --git a/llm_studio/python_configs/text_causal_language_modeling_config.py b/llm_studio/python_configs/text_causal_language_modeling_config.py
@@ -6,6 +6,7 @@
 import torch
 
 import llm_studio.src.datasets.text_causal_language_modeling_ds
+from llm_studio.app_utils.config import default_cfg
 from llm_studio.python_configs.base import DefaultConfig, DefaultConfigProblemBase
 from llm_studio.src import possible_values
 from llm_studio.src.augmentations.nlp_aug import BaseNLPAug
@@ -613,7 +614,11 @@ def __post_init__(self):
 class ConfigProblemBase(DefaultConfigProblemBase):
  output_directory: str = f"output/{os.path.basename(__file__).split('.')[0]}"
  experiment_name: str = field(default_factory=generate_experiment_name)
- llm_backbone: str = "h2oai/h2o-danube3-500m-base"
+ llm_backbone: str = (
+ "h2oai/h2o-danube3-500m-base"
+ if "h2oai/h2o-danube3-500m-base" in default_cfg.default_causal_language_models
+ else default_cfg.default_causal_language_models[0]
+ )
 
  dataset: ConfigNLPCausalLMDataset = field(default_factory=ConfigNLPCausalLMDataset)
  tokenizer: ConfigNLPCausalLMTokenizer = field(
@@ -640,24 +645,7 @@ def __post_init__(self):
  self._visibility["output_directory"] = -1
 
  self._possible_values["llm_backbone"] = possible_values.String(
- values=(
- "h2oai/h2o-danube3-500m-base",
- "h2oai/h2o-danube3-500m-chat",
- "h2oai/h2o-danube3-4b-base",
- "h2oai/h2o-danube3-4b-chat",
- "h2oai/h2o-danube2-1.8b-base",
- "h2oai/h2o-danube2-1.8b-chat",
- "meta-llama/Meta-Llama-3.1-8B-Instruct",
- "meta-llama/Meta-Llama-3.1-70B-Instruct",
- "mistralai/Mistral-7B-v0.3",
- "mistralai/Mistral-7B-Instruct-v0.2",
- "google/gemma-2-2b-it",
- "google/gemma-2-9b-it",
- "microsoft/Phi-3-mini-4k-instruct",
- "microsoft/Phi-3-medium-4k-instruct",
- "Qwen/Qwen2-7B-Instruct",
- "Qwen/Qwen2-72B-Instruct",
- ),
+ values=default_cfg.default_causal_language_models,
  allow_custom=True,
  )
 

diff --git a/llm_studio/python_configs/text_causal_regression_modeling_config.py b/llm_studio/python_configs/text_causal_regression_modeling_config.py
@@ -4,6 +4,7 @@
 
 import llm_studio.src.datasets.text_causal_regression_ds
 import llm_studio.src.plots.text_causal_classification_modeling_plots
+from llm_studio.app_utils.config import default_cfg
 from llm_studio.python_configs.base import DefaultConfig, DefaultConfigProblemBase
 from llm_studio.python_configs.text_causal_classification_modeling_config import (
  ConfigNLPCausalClassificationAugmentation as ConfigNLPCausalRegressionAugmentation,
@@ -109,7 +110,11 @@ def __post_init__(self):
 class ConfigProblemBase(DefaultConfigProblemBase):
  output_directory: str = f"output/{os.path.basename(__file__).split('.')[0]}"
  experiment_name: str = field(default_factory=generate_experiment_name)
- llm_backbone: str = "h2oai/h2o-danube3-500m-chat"
+ llm_backbone: str = (
+ "h2oai/h2o-danube3-500m-chat"
+ if "h2oai/h2o-danube3-500m-chat" in default_cfg.default_causal_language_models
+ else default_cfg.default_causal_language_models[0]
+ )
 
  dataset: ConfigNLPCausalRegressionDataset = field(
  default_factory=ConfigNLPCausalRegressionDataset
@@ -142,24 +147,7 @@ def __post_init__(self):
  self._visibility["output_directory"] = -1
 
  self._possible_values["llm_backbone"] = possible_values.String(
- values=(
- "h2oai/h2o-danube3-500m-base",
- "h2oai/h2o-danube3-500m-chat",
- "h2oai/h2o-danube3-4b-base",
- "h2oai/h2o-danube3-4b-chat",
- "h2oai/h2o-danube2-1.8b-base",
- "h2oai/h2o-danube2-1.8b-chat",
- "meta-llama/Meta-Llama-3.1-8B-Instruct",
- "meta-llama/Meta-Llama-3.1-70B-Instruct",
- "mistralai/Mistral-7B-v0.3",
- "mistralai/Mistral-7B-Instruct-v0.2",
- "google/gemma-2-2b-it",
- "google/gemma-2-9b-it",
- "microsoft/Phi-3-mini-4k-instruct",
- "microsoft/Phi-3-medium-4k-instruct",
- "Qwen/Qwen2-7B-Instruct",
- "Qwen/Qwen2-72B-Instruct",
- ),
+ values=default_cfg.default_causal_language_models,
  allow_custom=True,
  )
 

diff --git a/llm_studio/python_configs/text_dpo_modeling_config.py b/llm_studio/python_configs/text_dpo_modeling_config.py
@@ -3,6 +3,7 @@
 from typing import Any
 
 import llm_studio.src.datasets.text_dpo_modeling_ds
+from llm_studio.app_utils.config import default_cfg
 from llm_studio.python_configs.base import DefaultConfigProblemBase
 from llm_studio.python_configs.text_causal_language_modeling_config import (
  ConfigNLPAugmentation,
@@ -108,7 +109,11 @@ class ConfigProblemBase(DefaultConfigProblemBase):
  output_directory: str = f"output/{os.path.basename(__file__).split('.')[0]}"
  experiment_name: str = field(default_factory=generate_experiment_name)
 
- llm_backbone: str = "h2oai/h2o-danube3-500m-chat"
+ llm_backbone: str = (
+ "h2oai/h2o-danube3-500m-chat"
+ if "h2oai/h2o-danube3-500m-chat" in default_cfg.default_causal_language_models
+ else default_cfg.default_causal_language_models[0]
+ )
 
  dataset: ConfigDPODataset = field(default_factory=ConfigDPODataset)
  tokenizer: ConfigNLPCausalLMTokenizer = field(
@@ -129,23 +134,6 @@ def __post_init__(self):
  super().__post_init__()
  self._visibility["output_directory"] = -1
  self._possible_values["llm_backbone"] = possible_values.String(
- values=(
- "h2oai/h2o-danube3-500m-base",
- "h2oai/h2o-danube3-500m-chat",
- "h2oai/h2o-danube3-4b-base",
- "h2oai/h2o-danube3-4b-chat",
- "h2oai/h2o-danube2-1.8b-base",
- "h2oai/h2o-danube2-1.8b-chat",
- "meta-llama/Meta-Llama-3.1-8B-Instruct",
- "meta-llama/Meta-Llama-3.1-70B-Instruct",
- "mistralai/Mistral-7B-v0.3",
- "mistralai/Mistral-7B-Instruct-v0.2",
- "google/gemma-2-2b-it",
- "google/gemma-2-9b-it",
- "microsoft/Phi-3-mini-4k-instruct",
- "microsoft/Phi-3-medium-4k-instruct",
- "Qwen/Qwen2-7B-Instruct",
- "Qwen/Qwen2-72B-Instruct",
- ),
+ values=default_cfg.default_causal_language_models,
  allow_custom=True,
  )
diff --git a/llm_studio/python_configs/text_sequence_to_sequence_modeling_config.py b/llm_studio/python_configs/text_sequence_to_sequence_modeling_config.py
@@ -2,6 +2,7 @@
 from dataclasses import dataclass, field
 from typing import Any, Dict, List
 
+from llm_studio.app_utils.config import default_cfg
 from llm_studio.python_configs.base import DefaultConfigProblemBase
 from llm_studio.python_configs.text_causal_language_modeling_config import (
  ConfigNLPAugmentation,
@@ -72,7 +73,11 @@ def __post_init__(self):
 class ConfigProblemBase(DefaultConfigProblemBase):
  output_directory: str = f"output/{os.path.basename(__file__).split('.')[0]}"
  experiment_name: str = field(default_factory=generate_experiment_name)
- llm_backbone: str = "t5-small"
+ llm_backbone: str = (
+ "t5-small"
+ if "t5-small" in default_cfg.default_sequence_to_sequence_models
+ else default_cfg.default_sequence_to_sequence_models[0]
+ )
 
  dataset: ConfigNLPSeq2SeqDataset = field(default_factory=ConfigNLPSeq2SeqDataset)
  tokenizer: ConfigNLPCausalLMTokenizer = field(
@@ -99,15 +104,7 @@ def __post_init__(self):
  self._visibility["output_directory"] = -1
 
  self._possible_values["llm_backbone"] = possible_values.String(
- values=(
- "t5-small",
- "t5-base",
- "t5-large",
- "google/flan-t5-small",
- "google/flan-t5-base",
- "google/flan-t5-large",
- "google/flan-ul2",
- ),
+ values=default_cfg.default_sequence_to_sequence_models,
  allow_custom=True,
  )
 

diff --git a/pyproject.toml b/pyproject.toml
@@ -1,6 +1,6 @@
 [project]
 name = "h2o-llmstudio"
-version = "1.13.0-dev"
+version = "1.14.0-dev"
 readme = "README.md"
 license = {file = "LICENSE"}