Explorar

28 technologies shown

• Unsloth framework Fine-tuning implementation with hand-written kernels for speed and low memory.
• CrewAI framework Defines agent teams with roles and tasks, and runs them in order.
• LoRA technique Fine-tuning by training small low-rank adapter matrices instead of all weights.
• RLAIF technique Reinforcement learning from AI feedback instead of human feedback.
• TensorRT-LLM tool NVIDIA's compiler and runtime for optimised LLM inference on its GPUs.
• E5 model Text embedding model family trained on weakly supervised pairs.
• Haystack framework Framework for building retrieval and question-answering pipelines.
• Jailbreak technique Prompt that circumvents a model's refusal behaviour.
• Jina Embeddings model Embedding models designed around long inputs and multilingual coverage.
• MI300X hardware AMD's datacentre accelerator with a large unified memory pool.
• Reinforcement Learning concept Learning a policy from rewards obtained by acting in an environment.
• Text Generation Inference tool Hugging Face's serving stack for transformer models.
• Typesense database Typo-tolerant search engine with vector support, aimed at developer experience.
• Vespa database Search platform combining vector and structured ranking in one query.
• Ragas tool Evaluation library that scores retrieval and answer quality separately.
• Command R model Model family oriented toward retrieval and tool use.
• LanceDB database Embedded vector database built on a columnar format for multimodal data.
• LangSmith tool Tracing and evaluation for model applications, showing each step of a run.
• Nomic Embed model Open embedding model with an auditable training set and long context.
• Tokenization technique Splitting text into the subword units a language model actually reads.
• Axolotl framework Configuration-driven fine-tuning runner for many model architectures.
• DeepEval library Open-source framework of tests for model applications.
• GAN architecture Generator and discriminator trained against each other to synthesise realistic data.
• Instructor model Embedding model that takes a task instruction alongside the text.
• KV Cache technique Cached attention keys and values that make generating one token at a time affordable.
• TruLens library Instrumentation and feedback functions for evaluating model applications.
• RNN architecture Recurrent neural network that carries a hidden state along a sequence.
• LSTM architecture Recurrent network with gating that lets it retain information over longer sequences.