This is an AI translated post.
Building an AI Full Stack with Open Source
- Writing language: Korean
- •
- Base country: All countries
- •
- Information Technology
Select Language
Summarized by durumis AI
- The open source ecosystem is experiencing an AI open source renaissance, with many models being released by the Open LLM camp.
- Various inference and serving tools and LLM monitoring and management tools are emerging for LLM utilization.
- Various frameworks for developing LLM-based applications are being introduced.
With the surge of numerous open-source projects related to AI, the open-source ecosystem is experiencing a true renaissance in AI open-source. Following the success of LangChain, many open-source projects have emerged, rapidly filling the gaps in AI industry systems.
Open LLM
LLM (Large Language Model), the core of generative AI, is divided into two axes: Closed LLM led by GPT and Open LLM in the Llama camp. Mistral team has released its model under an open-source license, drawing attention from many with its outstanding performance. Open LLM is primarily managed and provided through Hugging Face.
Mixtral-8x7B-Instruct-v0.1(Apache-2.0)
https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0\.1
- Expert Mixture of Experts (SMoE) model was adopted.
- It shows performance that surpasses the Llama 2 70B model and even GPT-3.5 175B.
- It ranks third in Chatbot Arena, a blind chatbot test, following GPT-4 and Claude-2.
Llama-2-7b-chat(Llama 2 Community)
https://huggingface.co/meta-llama/Llama-2-7b-chat
- It is a license that allows commercial use for services with less than 700 million monthly active users.
- Numerous derivative models based on Llama-2 have emerged.
phi-2(MIT)
https://huggingface.co/microsoft/phi-2
- It is a lightweight model with 2.7B parameters released by MS.
- Tests on common sense, language understanding, and logical reasoning have shown that its performance is better than the 13B model.
LLM Inference and Serving
To effectively utilize well-trained LLMs, tools are needed that can provide fast inference and efficient management of computing resources.
Ollama(MIT)
https://github.com/jmorganca/ollama
- It allows you to run LLMs up to 7B in size directly on local environments like Mac, Linux, and Windows.
- You can download and run models with simple commands.
- Models can be managed through the CLI, and simple chatting is possible.
- Various applications are possible through the provided API.
vLLM(Apache-2.0)
https://github.com/vllm-project/vllm
- It is a fast and easy-to-use library for LLM inference and serving.
- Supports models provided by Hugging Face.
- Provides distributed processing, parallel processing, streaming output, and OpenAI compatible API.
- Supports Nvidia and AMD GPUs.
KServe(Apache-2.0)
https://github.com/kserve/kserve- It is a platform for ML model inference that can be built in a Kubernetes environment. - Provides an abstraction interface for scaling, networking, and monitoring.
LLM Proxying
LiteLLM(MIT)
https://github.com/BerriAI/litellm
- Integrates various LLM APIs and provides proxying.
- Follows the API format of OpenAI.
- Provides user-specific API authentication management.
One API(MIT)
https://github.com/songquanpeng/one-api
- It enables instant access to all large models through the standard OpenAI API format.
- Supports various LLMs and also provides proxy services.
- Load balancing and multi-deployment are possible, and it provides user management and group features.
AI Gateway(MIT)
https://github.com/Portkey-AI/gateway
- Provides connections to over 100 LLMs with a single, fast, and familiar API.
- Ensures fast access with a small installation size.
LLM Monitoring Great Expectations(Apache-2.0)
https://github.com/great-expectations/great_expectations
- Helps data teams build a shared understanding of their data through quality testing, documentation, and profiling.
- It can be integrated with CI/CD pipelines to add data quality precisely where needed.
LangFuse(MIT)
https://github.com/langfuse/langfuse
- Provides open-source LLM visibility, analysis, rapid management, evaluation, testing, monitoring, logging, and tracing.
- You can explore and debug complex logs and traces from a visual UI.
- Enterprise features are planned for the future.
Giskard(Apache-2.0, Dual License)
https://github.com/Giskard-AI/giskard
- It can automatically detect vulnerabilities in AI models, from tabular models to LLMs, such as bias, data leakage, spurious correlations, hallucinations, toxicity, and security issues.
- It scans for vulnerabilities in AI models and automatically generates test suites to support quality assurance processes for ML models and LLMs.
- Provides a SaaS platform for detecting AI safety risks in deployed LLM applications. (Premium)
LLM Framework
LangChain (MIT)
https://github.com/langchain-ai/langchain
- It is a framework for developing applications powered by language models.
- It is available in Python and Javascript, providing an abstraction layer that integrates numerous libraries.
- You can also deploy the built LangChain as an API.
LlamaIndex(MIT)
https://github.com/run-llama/llama_index
- It is a data-centric framework for LLM applications.
- Provides data connectors for collecting existing data sources and data formats (API, PDF, documents, SQL, etc.).
- Provides methods for structuring data (indexes, graphs) to make it easily usable by LLMs.
Haystack(Apache-2.0)
https://github.com/deepset-ai/haystack
- It is an LLM framework for easy construction of search augmented generation (RAG), document search, question answering, and answer generation.
- It is built based on the pipeline concept.
Flowise(Apache-2.0)
https://github.com/FlowiseAI/Flowise
- You can build custom LLM flows by dragging and dropping the UI.
LangFlow(MIT)
https://github.com/logspace-ai/langflow
- It makes it easy to experiment and prototype LangChain pipelines.
- It is executed using the CLI, and it also supports deploying Langflow to Google Cloud Platform (GCP).
Spring AI(Apache-2.0)
https://github.com/spring-projects/spring-ai
- AI framework provided by Spring Framework (still in snapshot stage)
- Supports API integration based on OpenAI and MS Azure, providing an abstraction layer
- The goal is to make it easier and more scalable to implement AI features using AI Templates
Data Juicer(Apache-2.0)
https://github.com/alibaba/data-juicer
- It is an open-source project released by Alibaba, providing a one-stop data processing system for LLMs.
- It provides a systematic library comprised of over 20 reusable configuration recipes, over 50 core OPs, and a rich functional dedicated toolkit.
- With automated report generation capabilities, it allows for in-depth data analysis to gain a deeper understanding of data sets.