Merge pull request #38 from iryna-kondr/feature_palm

Added Azure OpenAI + Google PaLM support

.gitignore

@@ -161,3 +161,4 @@ cython_debug/
 test.py
 tmp.ipynb
 tmp.py
+*.pickle

README.md

@@ -43,6 +43,21 @@ SKLLMConfig.set_openai_org("<YOUR_ORGANISATION>")
 - If you have a free trial OpenAI account, the [rate limits](https://platform.openai.com/docs/guides/rate-limits/overview) are not sufficient (specifically 3 requests per minute). Please switch to the "pay as you go" plan first.
 - When calling `SKLLMConfig.set_openai_org`, you have to provide your organization ID and **NOT** the name. You can find your ID [here](https://platform.openai.com/account/org-settings).
 
+### Using Azure OpenAI
+
+```python
+from skllm.config import SKLLMConfig
+
+SKLLMConfig.set_openai_key("<YOUR_KEY>") #use azure key instead
+SKLLMConfig.set_azure_api_base("<API_BASE>")
+
+# start with "azure::" prefix when setting the model name
+model_name = "azure::<model_name>"
+# e.g. ZeroShotGPTClassifier(openai_model="azure::gpt-3.5-turbo")
+```
+
+Note: Azure OpenAI is not supported by the preprocessors at the moment.
+
 ### Using GPT4ALL
 
 In addition to OpenAI, some of the models can use [gpt4all](https://gpt4all.io/index.html) as a backend.
@@ -66,18 +81,20 @@ ZeroShotGPTClassifier(openai_model="gpt4all::ggml-gpt4all-j-v1.3-groovy")
 
 When running for the first time, the model file will be downloaded automatially.
 
-At the moment only the following estimators support gpt4all as a backend:
-
-- `ZeroShotGPTClassifier`
-- `MultiLabelZeroShotGPTClassifier`
-- `FewShotGPTClassifier`
-
 When using gpt4all please keep the following in mind:
 
 1. Not all gpt4all models are commercially licensable, please consult gpt4all website for more details.
 2. The accuracy of the models may be much lower compared to ones provided by OpenAI (especially gpt-4).
 3. Not all of the available models were tested, some may not work with scikit-llm at all.
 
+## Supported models by a non-standard backend
+
+At the moment only the following estimators support non-standard backends (gpt4all, azure):
+
+- `ZeroShotGPTClassifier`
+- `MultiLabelZeroShotGPTClassifier`
+- `FewShotGPTClassifier`
+
 ### Zero-Shot Text Classification
 
 One of the powerful ChatGPT features is the ability to perform text classification without being re-trained. For that, the only requirement is that the labels must be descriptive.
@@ -205,6 +222,28 @@ clf.fit(X, y)
 labels = clf.predict(X)
 ```
 
+### Text Classification with Google PaLM 2
+
+At the moment 3 PaLM based models are available in test mode:
+- `ZeroShotPaLMClassifier` - zero-shot text classification with PaLM 2;
+- `PaLMClassifier` - fine-tuneable text classifier with PaLM 2;
+- `PaLM` - fine-tuneable estimator that can be trained on arbitrary text input-output pairs.
+
+Example: 
+
+```python
+from skllm.models.palm import PaLMClassifier
+from skllm.datasets import get_classification_dataset
+
+X, y = get_classification_dataset()
+
+clf = PaLMClassifier(n_update_steps=100)
+clf.fit(X, y)
+labels = clf.predict(X)
+```
+
+A more detailed documentation will follow soon. For now, please refer to our [official guide on Medium](https://medium.com/@iryna230520).
+
 ### Text Vectorization
 
 As an alternative to using GPT as a classifier, it can be used solely for data preprocessing. `GPTVectorizer` allows to embed a chunk of text of arbitrary length to a fixed-dimensional vector, that can be used with virtually any classification or regression model.
@@ -273,18 +312,3 @@ t = GPTTranslator(openai_model="gpt-3.5-turbo", output_language="English")
 translated_text = t.fit_transform(X)
 ```
 
-## Roadmap 🧭
-
-- [x] Zero-Shot Classification with OpenAI GPT 3/4
-  - [x] Multiclass classification
-  - [x] Multi-label classification
-- [ ] Few-Shot classifier
-  - [x] Multiclass classification
-  - [ ] Multi-label classification
-- [x] GPT Vectorizer
-- [x] ChatGPT models
-- [ ] InstructGPT models
-- [ ] InstructGPT Fine-tuning (optional)
-- [ ] Open source models
-
-*The order of the elements in the roadmap is arbitrary and does not reflect the planned order of implementation.*

skllm/__init__.py

@@ -1,7 +1,7 @@
 # ordering is important here to prevent circular imports
-from skllm.models.gpt_zero_shot_clf import (
+from skllm.models.gpt.gpt_zero_shot_clf import (
     MultiLabelZeroShotGPTClassifier,
     ZeroShotGPTClassifier,
 )
-from skllm.models.gpt_few_shot_clf import FewShotGPTClassifier
-from skllm.models.gpt_dyn_few_shot_clf import DynamicFewShotGPTClassifier
+from skllm.models.gpt.gpt_few_shot_clf import FewShotGPTClassifier
+from skllm.models.gpt.gpt_dyn_few_shot_clf import DynamicFewShotGPTClassifier

skllm/completions.py

@@ -3,14 +3,19 @@
 
 
 def get_chat_completion(
-    messages: dict, openai_key: str=None, openai_org: str=None, model: str="gpt-3.5-turbo", max_retries: int=3
+    messages: dict,
+    openai_key: str = None,
+    openai_org: str = None,
+    model: str = "gpt-3.5-turbo",
+    max_retries: int = 3,
 ):
-    """
-    Gets a chat completion from the OpenAI API.
-    """
+    """Gets a chat completion from the OpenAI API."""
     if model.startswith("gpt4all::"):
         return _g4a_get_chat_completion(messages, model[9:])
     else:
+        api = "azure" if model.startswith("azure::") else "openai"
+        if api == "azure":
+            model = model[7:]
         return _oai_get_chat_completion(
-            messages, openai_key, openai_org, model, max_retries
+            messages, openai_key, openai_org, model, max_retries, api=api
         )

skllm/config.py

@@ -3,21 +3,121 @@
 
 _OPENAI_KEY_VAR = "SKLLM_CONFIG_OPENAI_KEY"
 _OPENAI_ORG_VAR = "SKLLM_CONFIG_OPENAI_ORG"
+_AZURE_API_BASE_VAR = "SKLLM_CONFIG_AZURE_API_BASE"
+_AZURE_API_VERSION_VAR = "SKLLM_CONFIG_AZURE_API_VERSION"
+_GOOGLE_PROJECT = "GOOGLE_CLOUD_PROJECT"
 
-class SKLLMConfig():
-    
+
+class SKLLMConfig:
     @staticmethod
     def set_openai_key(key: str) -> None:
+        """Sets the OpenAI key.
+
+        Parameters
+        ----------
+        key : str
+            OpenAI key.
+        """
         os.environ[_OPENAI_KEY_VAR] = key
 
     @staticmethod
     def get_openai_key() -> Optional[str]:
+        """Gets the OpenAI key.
+
+        Returns
+        -------
+        Optional[str]
+            OpenAI key.
+        """
         return os.environ.get(_OPENAI_KEY_VAR, None)
-    
+
     @staticmethod
     def set_openai_org(key: str) -> None:
+        """Sets OpenAI organization ID.
+
+        Parameters
+        ----------
+        key : str
+            OpenAI organization ID.
+        """
         os.environ[_OPENAI_ORG_VAR] = key
 
     @staticmethod
-    def get_openai_org() -> Optional[str]:
-        return os.environ.get(_OPENAI_ORG_VAR, None)
+    def get_openai_org() -> str:
+        """Gets the OpenAI organization ID.
+
+        Returns
+        -------
+        str
+            OpenAI organization ID.
+        """
+        return os.environ.get(_OPENAI_ORG_VAR, "")
+
+    @staticmethod
+    def get_azure_api_base() -> str:
+        """Gets the API base for Azure.
+
+        Returns
+        -------
+        str
+            URL to be used as the base for the Azure API.
+        """
+        base = os.environ.get(_AZURE_API_BASE_VAR, None)
+        if base is None:
+            raise RuntimeError("Azure API base is not set")
+        return base
+
+    @staticmethod
+    def set_azure_api_base(base: str) -> None:
+        """Set the API base for Azure.
+
+        Parameters
+        ----------
+        base : str
+            URL to be used as the base for the Azure API.
+        """
+        os.environ[_AZURE_API_BASE_VAR] = base
+
+    @staticmethod
+    def set_azure_api_version(ver: str) -> None:
+        """Set the API version for Azure.
+
+        Parameters
+        ----------
+        ver : str
+            Azure API version.
+        """
+        os.environ[_AZURE_API_VERSION_VAR] = ver
+
+    @staticmethod
+    def get_azure_api_version() -> str:
+        """Gets the API version for Azure.
+
+        Returns
+        -------
+        str
+            Azure API version.
+        """
+        return os.environ.get(_AZURE_API_VERSION_VAR, "2023-05-15")
+
+    @staticmethod
+    def get_google_project() -> Optional[str]:
+        """Gets the Google Cloud project ID.
+
+        Returns
+        -------
+        Optional[str]
+            Google Cloud project ID.
+        """
+        return os.environ.get(_GOOGLE_PROJECT, None)
+
+    @staticmethod
+    def set_google_project(project: str) -> None:
+        """Sets the Google Cloud project ID.
+
+        Parameters
+        ----------
+        project : str
+            Google Cloud project ID.
+        """
+        os.environ[_GOOGLE_PROJECT] = project

skllm/google/completions.py

@@ -0,0 +1,41 @@
+from time import sleep
+
+from vertexai.preview.language_models import ChatModel, TextGenerationModel
+
+# TODO reduce code duplication for retrying logic
+
+
+def get_completion(model: str, text: str, max_retries: int = 3):
+    for _ in range(max_retries):
+        try:
+            if model.startswith("text-"):
+                model = TextGenerationModel.from_pretrained(model)
+            else:
+                model = TextGenerationModel.get_tuned_model(model)
+            response = model.predict(text, temperature=0.0)
+            return response.text
+        except Exception as e:
+            error_msg = str(e)
+            error_type = type(e).__name__
+            sleep(3)
+    print(
+        f"Could not obtain the completion after {max_retries} retries: `{error_type} ::"
+        f" {error_msg}`"
+    )
+
+
+def get_completion_chat_mode(model: str, context: str, text: str, max_retries: int = 3):
+    for _ in range(max_retries):
+        try:
+            model = ChatModel.from_pretrained(model)
+            chat = model.start_chat(context=context)
+            response = chat.send_message(text, temperature=0.0)
+            return response.text
+        except Exception as e:
+            error_msg = str(e)
+            error_type = type(e).__name__
+            sleep(3)
+    print(
+        f"Could not obtain the completion after {max_retries} retries: `{error_type} ::"
+        f" {error_msg}`"
+    )

skllm/google/tuning.py

@@ -0,0 +1,13 @@
+from pandas import DataFrame
+from vertexai.preview.language_models import TextGenerationModel
+
+
+def tune(model: str, data: DataFrame, train_steps: int = 100):
+    model = TextGenerationModel.from_pretrained(model)
+    model.tune_model(
+        training_data=data,
+        train_steps=train_steps,
+        tuning_job_location="europe-west4",  # the only supported training location atm
+        tuned_model_location="us-central1",  # the only supported deployment location atm
+    )
+    return model  # ._job