Merge branch 'main' into conda_pack_build_related_fixes

Author: qiuosier

.pre-commit-config.yaml

@@ -45,7 +45,7 @@ repos:
       rev: v8.18.4
       hooks:
           - id: gitleaks
-            exclude: .github/workflows/reusable-actions/set-dummy-conf.yml
+            exclude: .github/workflows/reusable-actions/set-dummy-conf.yml|./tests/operators/common/test_load_data.py
     # Oracle copyright checker
     - repo: https://github.com/oracle-samples/oci-data-science-ai-samples/
       rev: 1bc5270a443b791c62f634233c0f4966dfcc0dd6

CODEOWNERS

@@ -1 +1 @@
-* @darenr @mayoor @mrDzurb @VipulMascarenhas @qiuosier
+* @darenr @mayoor @mrDzurb @VipulMascarenhas @qiuosier @ahosler

ads/aqua/app.py

@@ -1,5 +1,4 @@
 #!/usr/bin/env python
-# -*- coding: utf-8 -*-
 # Copyright (c) 2024 Oracle and/or its affiliates.
 # Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/
 
@@ -175,7 +174,7 @@ def create_model_version_set(
                         f"Invalid model version set name. Please provide a model version set with `{tag}` in tags."
                     )
 
-            except:
+            except Exception:
                 logger.debug(
                     f"Model version set {model_version_set_name} doesn't exist. "
                     "Creating new model version set."
@@ -254,7 +253,7 @@ def if_artifact_exist(self, model_id: str, **kwargs) -> bool:
 
         try:
             response = self.ds_client.head_model_artifact(model_id=model_id, **kwargs)
-            return True if response.status == 200 else False
+            return response.status == 200
         except oci.exceptions.ServiceError as ex:
             if ex.status == 404:
                 logger.info(f"Artifact not found in model {model_id}.")
@@ -302,15 +301,15 @@ def get_config(self, model_id: str, config_file_name: str) -> Dict:
                 config_path,
                 config_file_name=config_file_name,
             )
-        except:
+        except Exception:
             # todo: temp fix for issue related to config load for byom models, update logic to choose the right path
             try:
                 config_path = f"{artifact_path.rstrip('/')}/config/"
                 config = load_config(
                     config_path,
                     config_file_name=config_file_name,
                 )
-            except:
+            except Exception:
                 pass
 
         if not config:
@@ -343,7 +342,7 @@ def build_cli(self) -> str:
         params = [
             f"--{field.name} {getattr(self,field.name)}"
             for field in fields(self.__class__)
-            if getattr(self, field.name)
+            if getattr(self, field.name) is not None
         ]
         cmd = f"{cmd} {' '.join(params)}"
         return cmd

ads/aqua/common/entities.py

@@ -0,0 +1,17 @@
+#!/usr/bin/env python
+# Copyright (c) 2024 Oracle and/or its affiliates.
+# Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/
+
+
+class ContainerSpec:
+    """
+    Class to hold to hold keys within the container spec.
+    """
+
+    CONTAINER_SPEC = "containerSpec"
+    CLI_PARM = "cliParam"
+    SERVER_PORT = "serverPort"
+    HEALTH_CHECK_PORT = "healthCheckPort"
+    ENV_VARS = "envVars"
+    RESTRICTED_PARAMS = "restrictedParams"
+    EVALUATION_CONFIGURATION = "evaluationConfiguration"

ads/aqua/common/enums.py

@@ -1,5 +1,4 @@
 #!/usr/bin/env python
-# -*- coding: utf-8 -*-
 # Copyright (c) 2024 Oracle and/or its affiliates.
 # Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/
 
@@ -8,6 +7,7 @@
 ~~~~~~~~~~~~~~
 This module contains the set of enums used in AQUA.
 """
+
 from ads.common.extended_enum import ExtendedEnumMeta
 
 
@@ -38,21 +38,34 @@ class Tags(str, metaclass=ExtendedEnumMeta):
     READY_TO_IMPORT = "ready_to_import"
     BASE_MODEL_CUSTOM = "aqua_custom_base_model"
     AQUA_EVALUATION_MODEL_ID = "evaluation_model_id"
+    MODEL_FORMAT = "model_format"
+    MODEL_ARTIFACT_FILE = "model_file"
 
 
 class InferenceContainerType(str, metaclass=ExtendedEnumMeta):
     CONTAINER_TYPE_VLLM = "vllm"
     CONTAINER_TYPE_TGI = "tgi"
+    CONTAINER_TYPE_LLAMA_CPP = "llama-cpp"
 
 
 class InferenceContainerTypeFamily(str, metaclass=ExtendedEnumMeta):
     AQUA_VLLM_CONTAINER_FAMILY = "odsc-vllm-serving"
     AQUA_TGI_CONTAINER_FAMILY = "odsc-tgi-serving"
+    AQUA_LLAMA_CPP_CONTAINER_FAMILY = "odsc-llama-cpp-serving"
 
 
 class InferenceContainerParamType(str, metaclass=ExtendedEnumMeta):
     PARAM_TYPE_VLLM = "VLLM_PARAMS"
     PARAM_TYPE_TGI = "TGI_PARAMS"
+    PARAM_TYPE_LLAMA_CPP = "LLAMA_CPP_PARAMS"
+
+
+class EvaluationContainerTypeFamily(str, metaclass=ExtendedEnumMeta):
+    AQUA_EVALUATION_CONTAINER_FAMILY = "odsc-llm-evaluate"
+
+
+class FineTuningContainerTypeFamily(str, metaclass=ExtendedEnumMeta):
+    AQUA_FINETUNING_CONTAINER_FAMILY = "odsc-llm-fine-tuning"
 
 
 class HuggingFaceTags(str, metaclass=ExtendedEnumMeta):

ads/aqua/common/utils.py

@@ -10,14 +10,26 @@
 import os
 import random
 import re
+import shlex
+import subprocess
+from datetime import datetime, timedelta
 from functools import wraps
 from pathlib import Path
 from string import Template
 from typing import List, Union
 
 import fsspec
 import oci
+from cachetools import TTLCache, cached
+from huggingface_hub.hf_api import HfApi, ModelInfo
+from huggingface_hub.utils import (
+    GatedRepoError,
+    HfHubHTTPError,
+    RepositoryNotFoundError,
+    RevisionNotFoundError,
+)
 from oci.data_science.models import JobRun, Model
+from oci.object_storage.models import ObjectSummary
 
 from ads.aqua.common.enums import (
     InferenceContainerParamType,
@@ -34,6 +46,7 @@
     COMPARTMENT_MAPPING_KEY,
     CONSOLE_LINK_RESOURCE_TYPE_MAPPING,
     CONTAINER_INDEX,
+    HF_LOGIN_DEFAULT_TIMEOUT,
     MAXIMUM_ALLOWED_DATASET_IN_BYTE,
     MODEL_BY_REFERENCE_OSS_PATH_KEY,
     SERVICE_MANAGED_CONTAINER_URI_SCHEME,
@@ -44,8 +57,7 @@
     VLLM_INFERENCE_RESTRICTED_PARAMS,
 )
 from ads.aqua.data import AquaResourceIdentifier
-from ads.common.auth import default_signer
-from ads.common.decorator.threaded import threaded
+from ads.common.auth import AuthState, default_signer
 from ads.common.extended_enum import ExtendedEnumMeta
 from ads.common.object_storage_details import ObjectStorageDetails
 from ads.common.oci_resource import SEARCH_TYPE, OCIResource
@@ -213,7 +225,6 @@ def read_file(file_path: str, **kwargs) -> str:
         return UNKNOWN
 
 
-@threaded()
 def load_config(file_path: str, config_file_name: str, **kwargs) -> dict:
     artifact_path = f"{file_path.rstrip('/')}/{config_file_name}"
     signer = default_signer() if artifact_path.startswith("oci://") else {}
@@ -228,6 +239,32 @@ def load_config(file_path: str, config_file_name: str, **kwargs) -> dict:
     return config
 
 
+def list_os_files_with_extension(oss_path: str, extension: str) -> [str]:
+    """
+    List files in the specified directory with the given extension.
+
+    Parameters:
+    - oss_path: The path to the directory where files are located.
+    - extension: The file extension to filter by (e.g., 'txt' for text files).
+
+    Returns:
+    - A list of file paths matching the specified extension.
+    """
+
+    oss_client = ObjectStorageDetails.from_path(oss_path)
+
+    # Ensure the extension is prefixed with a dot if not already
+    if not extension.startswith("."):
+        extension = "." + extension
+    files: List[ObjectSummary] = oss_client.list_objects().objects
+
+    return [
+        file.name[len(oss_client.filepath) :].lstrip("/")
+        for file in files
+        if file.name.endswith(extension)
+    ]
+
+
 def is_valid_ocid(ocid: str) -> bool:
     """Checks if the given ocid is valid.
 
@@ -503,6 +540,7 @@ def container_config_path():
     return f"oci://{AQUA_SERVICE_MODELS_BUCKET}@{CONDA_BUCKET_NS}/service_models/config"
 
 
+@cached(cache=TTLCache(maxsize=1, ttl=timedelta(hours=5), timer=datetime.now))
 def get_container_config():
     config = load_config(
         file_path=container_config_path(),
@@ -743,6 +781,33 @@ def get_ocid_substring(ocid: str, key_len: int) -> str:
     return ocid[-key_len:] if ocid and len(ocid) > key_len else ""
 
 
+def upload_folder(os_path: str, local_dir: str, model_name: str) -> str:
+    """Upload the local folder to the object storage
+
+    Args:
+        os_path (str): object storage URI with prefix. This is the path to upload
+        local_dir (str): Local directory where the object is downloaded
+        model_name (str): Name of the huggingface model
+    Retuns:
+        str: Object name inside the bucket
+    """
+    os_details: ObjectStorageDetails = ObjectStorageDetails.from_path(os_path)
+    if not os_details.is_bucket_versioned():
+        raise ValueError(f"Version is not enabled at object storage location {os_path}")
+    auth_state = AuthState()
+    object_path = os_details.filepath.rstrip("/") + "/" + model_name + "/"
+    command = f"oci os object bulk-upload --src-dir {local_dir} --prefix {object_path} -bn {os_details.bucket} -ns {os_details.namespace} --auth {auth_state.oci_iam_type} --profile {auth_state.oci_key_profile} --no-overwrite"
+    try:
+        logger.info(f"Running: {command}")
+        subprocess.check_call(shlex.split(command))
+    except subprocess.CalledProcessError as e:
+        logger.error(
+            f"Error uploading the object. Exit code: {e.returncode} with error {e.stdout}"
+        )
+
+    return f"oci://{os_details.bucket}@{os_details.namespace}" + "/" + object_path
+
+
 def is_service_managed_container(container):
     return container and container.startswith(SERVICE_MANAGED_CONTAINER_URI_SCHEME)
 
@@ -881,6 +946,8 @@ def get_container_params_type(container_type_name: str) -> str:
         return InferenceContainerParamType.PARAM_TYPE_VLLM
     elif InferenceContainerType.CONTAINER_TYPE_TGI in container_type_name.lower():
         return InferenceContainerParamType.PARAM_TYPE_TGI
+    elif InferenceContainerType.CONTAINER_TYPE_LLAMA_CPP in container_type_name.lower():
+        return InferenceContainerParamType.PARAM_TYPE_LLAMA_CPP
     else:
         return UNKNOWN
 
@@ -905,3 +972,93 @@ def get_restricted_params_by_container(container_type_name: str) -> set:
         return TGI_INFERENCE_RESTRICTED_PARAMS
     else:
         return set()
+
+
+def get_huggingface_login_timeout() -> int:
+    """This helper function returns the huggingface login timeout, returns default if not set via
+    env var.
+    Returns
+    -------
+    timeout: int
+        huggingface login timeout.
+
+    """
+    timeout = HF_LOGIN_DEFAULT_TIMEOUT
+    try:
+        timeout = int(
+            os.environ.get("HF_LOGIN_DEFAULT_TIMEOUT", HF_LOGIN_DEFAULT_TIMEOUT)
+        )
+    except ValueError:
+        pass
+    return timeout
+
+
+def format_hf_custom_error_message(error: HfHubHTTPError):
+    """
+    Formats a custom error message based on the Hugging Face error response.
+
+    Parameters
+    ----------
+    error (HfHubHTTPError): The caught exception.
+
+    Raises
+    ------
+    AquaRuntimeError: A user-friendly error message.
+    """
+    # Extract the repository URL from the error message if present
+    match = re.search(r"(https://huggingface.co/[^\s]+)", str(error))
+    url = match.group(1) if match else "the requested Hugging Face URL."
+
+    if isinstance(error, RepositoryNotFoundError):
+        raise AquaRuntimeError(
+            reason=f"Failed to access `{url}`. Please check if the provided repository name is correct. "
+            "If the repo is private, make sure you are authenticated and have a valid HF token registered. "
+            "To register your token, run this command in your terminal: `huggingface-cli login`",
+            service_payload={"error": "RepositoryNotFoundError"},
+        )
+
+    if isinstance(error, GatedRepoError):
+        raise AquaRuntimeError(
+            reason=f"Access denied to `{url}` "
+            "This repository is gated. Access is restricted to authorized users. "
+            "Please request access or check with the repository administrator. "
+            "If you are trying to access a gated repository, ensure you have a valid HF token registered. "
+            "To register your token, run this command in your terminal: `huggingface-cli login`",
+            service_payload={"error": "GatedRepoError"},
+        )
+
+    if isinstance(error, RevisionNotFoundError):
+        raise AquaRuntimeError(
+            reason=f"The specified revision could not be found at `{url}` "
+            "Please check the revision identifier and try again.",
+            service_payload={"error": "RevisionNotFoundError"},
+        )
+
+    raise AquaRuntimeError(
+        reason=f"An error occurred while accessing `{url}` "
+        "Please check your network connection and try again. "
+        "If you are trying to access a gated repository, ensure you have a valid HF token registered. "
+        "To register your token, run this command in your terminal: `huggingface-cli login`",
+        service_payload={"error": "Error"},
+    )
+
+
+@cached(cache=TTLCache(maxsize=1, ttl=timedelta(hours=5), timer=datetime.now))
+def get_hf_model_info(repo_id: str) -> ModelInfo:
+    """Gets the model information object for the given model repository name. For models that requires a token,
+    this method assumes that the token validation is already done.
+
+    Parameters
+    ----------
+    repo_id: str
+        hugging face model repository name
+
+    Returns
+    -------
+        instance of ModelInfo object
+
+    """
+    try:
+        return HfApi().model_info(repo_id=repo_id)
+    except HfHubHTTPError as err:
+        raise format_hf_custom_error_message(err) from err

ads/aqua/config/config.py

@@ -1,5 +1,4 @@
 #!/usr/bin/env python
-# -*- coding: utf-8 -*-
 # Copyright (c) 2024 Oracle and/or its affiliates.
 # Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/
 
@@ -14,5 +13,6 @@ def get_finetuning_config_defaults():
             "BM.GPU.A10.4": {"batch_size": 1, "replica": 1},
             "BM.GPU4.8": {"batch_size": 4, "replica": 1},
             "BM.GPU.A100-v2.8": {"batch_size": 6, "replica": 1},
+            "BM.GPU.H100.8": {"batch_size": 6, "replica": 1},
         }
     }