- Видео 15
- Просмотров 25 790
LLMOps Space
Добавлен 19 сен 2023
LLMOps.Space is a global community for LLM practitioners. 💡📚
The community focuses on content, discussions, and events around topics related to deploying LLMs into production. 🚀
Join discord: llmops.space/discord
The community focuses on content, discussions, and events around topics related to deploying LLMs into production. 🚀
Join discord: llmops.space/discord
Building the Fine-Tuning Pipeline for Alignment of LLMs 🏗️ | Nebius AI
In this session, Maksim Nekrashevich, ML & LLM Engineer from Nebius AI discussed the key aspects of aligning LLMs and explored how to set up the necessary infrastructure to maintain a versatile alignment pipeline.
Topics that will be covered:
✅ Incorporating LLMs into the data collection for supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to maximize efficiency.
✅ Techniques for instilling desired behaviors in LLMs through the strategic use of prompt tuning.
✅ An exploration of cutting-edge workflow management and how it facilitates rapid prototyping of highly-intensive distributed training procedures.
About LLMOps Space -
LLMOps.Space is a global community f...
Topics that will be covered:
✅ Incorporating LLMs into the data collection for supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to maximize efficiency.
✅ Techniques for instilling desired behaviors in LLMs through the strategic use of prompt tuning.
✅ An exploration of cutting-edge workflow management and how it facilitates rapid prototyping of highly-intensive distributed training procedures.
About LLMOps Space -
LLMOps.Space is a global community f...
Просмотров: 401
Видео
Traceability and Observability in Multi-Step LLM Systems | Langfuse | LLMOps
Просмотров 8362 месяца назад
In this session, Marc Klingen, CEO & Co-Founder at Langfuse (github.com/langfuse/langfuse), talked about advanced techniques and best practices for implementing traceability and observability in multi-step LLM systems. He also discussed how leveraging OpenTelemetry (OTel) can provide deep insights into RAG systems. 💡 Topics that were covered: ✅ Multi-Step LLM Applications: Understanding the com...
Vector and Keyword Search for Enhanced LLM Performance 🚀 | LLMOps
Просмотров 2854 месяца назад
In this session, Ohad Levi, Co-founder & CEO at Hyperspace, discussed the transformative power of Hybrid Search and its importance within the Gen AI LLM tech stack. Topics that were covered: ✅ The integration of classic keyword search functions and advanced vector search algorithms in the session. ✅ The use of metadata filtering, aggregations, and TF/IDF ranking. ✅ Advanced vector search algori...
Private RAG with Open Source and Custom LLMs 🚀 | BentoML | OpenLLM
Просмотров 1,8 тыс.4 месяца назад
In this session, Chaoyu Yang, Founder and CEO at BentoML, talked about the practical considerations of building private Retrieval-Augmented Generation (RAG) applications, utilizing a mix of open source and custom LLMs. He also talked about OpenLLM (github.com/bentoml/OpenLLM) and how it can help with LLM Deployments. ✅ Topics that were covered: ✅ The benefits of self-hosting open source LLMs or...
Building Multi-Modal Search and RAG with Vector Databases 🚀 | LLMOps
Просмотров 9375 месяцев назад
In this session, Zain from @Weaviate talked about using open-source multimodal embeddings in conjunction with large generative multimodal models to perform cross-modal search and multimodal retrieval augmented generation (MM-RAG). Topics that were covered: ✅ Multimodal Embedding Models: Learn how these models integrate various data forms, including images, text, audio, and sensory information, ...
Config-Based Development for LLMs: Versioning, Routing, & Evaluating LLMs 🚀 | LLMOps
Просмотров 6596 месяцев назад
In this talk, Sarmad, CEO of LastMile AI talked about how engineering teams can successfully transition their Generative AI apps from prototype to production with open-source and remote LLMs. He covered how to manage AI prompts and model generation settings as JSON-serializable configs - enabling version control, evaluation, and rapid prototyping. Topics that were covered: ✅ JSON-serializable c...
Fine-tuning LLMs with Hugging Face SFT 🚀 | QLoRA | LLMOps
Просмотров 1,8 тыс.7 месяцев назад
In this session, Harpreet from Deci AI talked about the nuances of supervised fine-tuning, instruction tuning, and the powerful techniques that bridge the gap between model objectives and user-specific requirements. He also demonstrated how to fine-tune LLMs using Hugging Face SFT. Topics that were covered: ✅ Specialized Fine-Tuning: Adapt LLMs for niche tasks using labeled data. ✅ Introduction...
The Science of LLM Benchmarks: Methods, Metrics, and Meanings | LLMOps
Просмотров 2,2 тыс.8 месяцев назад
In this talk, Jonathan discussed LLM benchmarks and their performance evaluation metrics. He addressed intriguing questions such as whether Gemini truly outperformed Open AI GPT-4V. He covered how to review benchmarks effectively and understand popular benchmarks like ARC, HellSwag, MMLU, and more. A step-by-step process to assess these benchmarks critically, helping you understand the strength...
Orchestrating RAG: Retrieval, Canopy, & Pinecone 🚀 | LLMOps
Просмотров 7438 месяцев назад
In this talk, Roy from Pinecone talked about the different aspects of orchestrating a RAG pipeline. He covered the challenges in building successful AI applications that scale to billions of documents, how Canopy is addressing those challenges, and the role of Vector Databases in the modern AI Stack. Topics that were covered: ✅ A deep understanding of the different aspects of orchestrating a RA...
Evaluating LLM-Based Apps: New Product Release | Deepchecks LLM Validation
Просмотров 6359 месяцев назад
In this session, Shir Chorev, CTO at Deepchecks, and Yaron, VP Product at Deepchecks discussed LLM hallucinations, evaluation methodologies, golden sets, and gave a live demonstration of the new Deepchecks LLM evaluation module. Deepchecks LLM Evaluation: deepchecks.com/solutions/llm-evaluation/ Topics that were covered: ✅ Hallucinations: Cases when the model generates outputs that aren’t groun...
Securing AI & LLM based Applications: Best Practices | LLMOps
Просмотров 7149 месяцев назад
In this talk, Itamar, founder of Prompt Security discussed security, privacy, and safety challenges associated with AI and LLM-based applications in production. Topics that were covered: ✅ Understanding the security landscape of LLM-based applications. ✅ Practical examples of potential attack scenarios. ✅ Methodological and technological measures to mitigate security risks. ✅ Best practices for...
Advanced RAG: Chunking, Embeddings, and Vector Databases 🚀 | LLMOps
Просмотров 10 тыс.10 месяцев назад
In this talk, Yujian from Zilliz talked about advanced RAG concepts including Chunking, Embeddings, and Vector Databases in RAG (Retrieval Augmented Generation) models Topics that were covered: ✅ Chunking: Understand the concept of chunking and its role in improving the efficiency of information retrieval. Learn how to implement chunking in RAG to optimize the retrieval of relevant information....
Building LLM Powered Apps: Best Practices | Qdrant | Vector Databases | LLMOps 🚀
Просмотров 55710 месяцев назад
In this talk, Kacper Łukawski, from Qdrant discussed best practices for building LLM-based apps and their integrations. As the adoption of these models becomes more widespread, it's essential to understand the potential technical hurdles that could impact the system's performance and scalability. Topics that were covered: ✅ Learn about the key considerations when designing applications powered ...
Leveraging Open Source Large Language Models (LLMs) for Production | Dstack 🚀
Просмотров 78711 месяцев назад
In this talk, Andrey, founder of dstack, discussed open-source Large Language Models (LLMs) and their practical applications. The session comprehensively compares open-source LLMs and proprietary options, such as OpenAI, and discusses the economic aspects of hosting these models. Topics that are covered: ✅ A comparative analysis of open-source LLMs and proprietary large language models. ✅ An ex...
Building an LLMOps Stack for Large Language Models | LLMs
Просмотров 3,8 тыс.11 месяцев назад
In this talk, our speakers Rafael and Puneet from @Databricks talked about the construction and optimization of LLMOps architecture. They discussed various components including MLFlow for Large Language Models (LLMs), Vector Databases, embeddings, and compute optimizations. Topics that are covered: ✅ MLFlow for LLMs: Discover the role of MLFlow in managing and streamlining LLMs. ✅ Vector Databa...
Talk begins at 8:00
Great tutorial, appreciate it guys!
Worst
What is the loss function used?
Good interview, I really enjoy it. Uniquely, I think it would be great less interruptions between each slide that Chaoyu was talking. It is only a constructive criticism, I think that distracts the audience. Keep it going, amazing content!
Thanks for putting the recording and sending an email notification about it!
😄
this is very interesting and can help in building quite useful applications
can you please share colab notebook also is there a way to put path of dataset and model locally (like path form my pc ) ?
colab.research.google.com/drive/1-xGUUad2O3Y3V0-qBxES7JPD4pKB4FI9?usp=sharing
You mention you’re the only distributed vector db. Is that true? There are multiple distributed vector dbs including Elasticsearch. What exactly makes you the only one?
it's just marketing or they are living under a rock lol
Can i use different embedding models for chunk embedding and query embedding
Thank you
Thank you for the session. 🎉
Fantastic
Thank you for the video.