Merge pull request #60 from corenting/add_groq_cloud

feat: add groq cloud + llama3 model

README.md

@@ -1,10 +1,11 @@
 # roboquote
 
-Generate random "inspirational" quotes images by using an AI text generation model through the Hugging Face Inference API.
+Generate random "inspirational" quotes images by using an AI text generation model.
 
 The following models can be used:
-- [bigscience/bloom](https://huggingface.co/bigscience/bloom)
-- [mistralai/Mixtral-8x7B-Instruct-v0.1](https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1)
+- [bigscience/bloom](https://huggingface.co/bigscience/bloom) through Hugging Face Inference API
+- [mistralai/Mixtral-8x7B-Instruct-v0.1](https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1) through Hugging Face Inference API
+- [meta-llama/Meta-Llama-3-70B](https://huggingface.co/meta-llama/Meta-Llama-3-70B) through GroqCloud API
 
 ## Examples
 
@@ -23,10 +24,10 @@ The following models can be used:
 ### Installation
 1. Install the project with [poetry](https://python-poetry.org/) by doing `poetry install`.
 2. Set environment variables for configuration (the project uses [environs](https://github.com/sloria/environs) so you can also put the variable in a [env file](https://github.com/sloria/environs#reading-env-files)):
-    - `HUGGING_FACE_API_TOKEN` (**required**): your Hugging Face Inference API token.
+    - `HUGGING_FACE_ACCESS_TOKEN` (**required**): your Hugging Face access token for their Inference API
+    - `GROQ_CLOUD_API_KEY` (**required**): your GroqCloud API key for their API
     - `LOG_LEVEL` (optional, default is `WARNING`): log level of the application
     - `WEB_DEBUG` (optional, default is `False`): if you want to run the web app in debug mode (should not be required)
-    - `HIDE_HUGGING
 
 ### CLI usage
 
@@ -45,6 +46,7 @@ It can be launched quickly locally with [uvicorn](https://pypi.org/project/uvico
 - [atomicparade](https://github.com/atomicparade) for the [code used to do the text auto wrapping](https://github.com/atomicparade/pil_autowrap/blob/main/pil_autowrap/pil_autowrap.py)
 - [BigScience Workshop](https://huggingface.co/bigscience/) for the BLOOM model used
 - [Google Fonts](https://fonts.google.com/) for the fonts used in the pictures.
+- [Groq](https://groq.com/) for their inference API
 - [Hugging Face](https://huggingface.co/) for the inference API
 - [Mistral AI](https://mistral.ai/) for the Mistral-7B-Instruct-v0.1 model used
 - [Unsplash](unsplash.com) for the background images.

main.py

@@ -1,15 +1,20 @@
 """Main CLI entrypoint."""
 
 import asyncio
+from typing import Annotated
 
+import click
 import typer
 
 from roboquote.background_image import (
     get_random_background_from_unsplash_by_theme,
     get_random_background_search_query,
 )
 from roboquote.entities.generate_options import GenerateOptions
-from roboquote.entities.text_model import TextModel
+from roboquote.entities.large_language_model import (
+    AVAILABLE_LARGE_LANGUAGE_MODELS,
+    AVAILABLE_LARGE_LANGUAGE_MODELS_NAMES,
+)
 from roboquote.quote_text_generation import get_random_quote
 from roboquote.result_image import generate_image
 
@@ -19,22 +24,32 @@
 @app.command()
 def generate(
     filename: str,
-    blur: bool = typer.Option(True, help="Add a blur on the background."),
-    blur_intensity: int | None = typer.Option(
-        None, help="If blur is enabled,the blur intensity level."
-    ),
-    background: str | None = typer.Option(
-        default=None,
-        help="If specified, use this string as the search query "
-        + "for the background image instead of a random one. "
-        + "Works best with simple queries like 'mountain', 'sea' etc.",
-    ),
-    text_model: TextModel = typer.Option(
-        default=TextModel.MISTRAL_8X7B_INSTRUCT.value,
-        help="The text generation model to use.",
-    ),
+    blur: Annotated[bool, typer.Option(help="Add a blur on the background.")] = True,
+    blur_intensity: Annotated[
+        int | None, typer.Option(help="If blur is enabled,the blur intensity level.")
+    ] = None,
+    background: Annotated[
+        str | None,
+        typer.Option(
+            help="If specified, use this string as the search query "
+            + "for the background image instead of a random one. "
+            + "Works best with simple queries like 'mountain', 'sea' etc.",
+        ),
+    ] = None,
+    model_name: Annotated[
+        str,
+        typer.Option(
+            click_type=click.Choice(AVAILABLE_LARGE_LANGUAGE_MODELS_NAMES),
+            help="The name of the LLM to use for the text generation.",
+        ),
+    ] = AVAILABLE_LARGE_LANGUAGE_MODELS_NAMES[0],
 ) -> None:
     """Generate a new image with the given filename."""
+    # Get model
+    large_language_model = next(
+        model for model in AVAILABLE_LARGE_LANGUAGE_MODELS if model.name == model_name
+    )
+
     # Get a random background category if not specified
     if background is None:
         background = get_random_background_search_query()
@@ -55,7 +70,7 @@ def generate(
     )
 
     # Get text to use
-    text = asyncio.run(get_random_quote(background, text_model))
+    text = asyncio.run(get_random_quote(background, large_language_model))
 
     # Generate and save image
     generated_image = generate_image(

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "roboquote"
-version = "0.3.0"
+version = "0.4.0"
 description = "A generator of inspirational quotes"
 authors = ["corenting <corenting@gmail.com>"]
 license = "MIT"

roboquote/__init__.py

@@ -4,7 +4,7 @@
 
 from roboquote import config
 
-__version__ = "0.3.0"
+__version__ = "0.4.0"
 
 # Setup logger
 logger.remove()

roboquote/config.py

@@ -6,5 +6,7 @@
 env.read_env()
 
 WEB_DEBUG: bool = env.bool("WEB_DEBUG", False)
-HUGGING_FACE_API_TOKEN: str = env.str("HUGGING_FACE_API_TOKEN")
 LOG_LEVEL: str = env.str("LOG_LEVEL", "WARNING")
+
+HUGGING_FACE_ACCESS_TOKEN: str = env.str("HUGGING_FACE_ACCESS_TOKEN")
+GROQ_CLOUD_API_KEY: str = env.str("GROQ_CLOUD_API_KEY")

roboquote/constants.py

@@ -1,9 +1,3 @@
 """Constants used in the code."""
 
-from roboquote.entities.text_model import TextModel
-
 FONTS_PATH = "fonts"
-
-HUGGING_FACE_BASE_API_URL = "https://api-inference.huggingface.co/models/"
-
-DEFAULT_WEB_TEXT_MODEL = TextModel.MISTRAL_8X7B_INSTRUCT

roboquote/entities/large_language_model.py

@@ -0,0 +1,47 @@
+from dataclasses import dataclass
+from enum import Enum
+
+
+class LargeLanguageModelPromptType(Enum):
+    CONTINUE = "continue"  # where the model continues an existing sentence
+    INSTRUCT = "instruct"  # where we instruct the model
+
+
+class LargeLanguageModelAPI(Enum):
+    GROQ_CLOUD = "GroqCloud"
+    HUGGING_FACE = "Hugging Face"
+
+
+@dataclass
+class LargeLanguageModel:
+    name: str
+    prompt_type: LargeLanguageModelPromptType
+    api: LargeLanguageModelAPI
+    prompt_start: str | None = None
+    prompt_end: str | None = None
+
+
+# Ordered by preferred models
+AVAILABLE_LARGE_LANGUAGE_MODELS = [
+    LargeLanguageModel(
+        name="llama3-70b-8192",
+        prompt_type=LargeLanguageModelPromptType.INSTRUCT,
+        api=LargeLanguageModelAPI.GROQ_CLOUD,
+    ),
+    LargeLanguageModel(
+        name="mistralai/Mixtral-8x7B-Instruct-v0.1",
+        prompt_type=LargeLanguageModelPromptType.INSTRUCT,
+        api=LargeLanguageModelAPI.HUGGING_FACE,
+        prompt_start="<s>[[INST]",
+        prompt_end="[/INST]",
+    ),
+    LargeLanguageModel(
+        name="bigscience/bloom",
+        prompt_type=LargeLanguageModelPromptType.CONTINUE,
+        api=LargeLanguageModelAPI.HUGGING_FACE,
+    ),
+]
+
+AVAILABLE_LARGE_LANGUAGE_MODELS_NAMES = [
+    model.name for model in AVAILABLE_LARGE_LANGUAGE_MODELS
+]