Foundation Models

Large models pre-trained on broad, unlabeled data that can be used as-is or fine-tuned for specific tasks.

Inside Foundation Models (13)

Hugging Face Transformers — A library providing pretrained models and tools for natural language processing tasks.
Jamba — Jamba is a foundation model by AI21 Labs, optimized for long document analysis and instruction following.
Codex — OpenAI's GPT-3 descendant for code generation and completion, trained on public code.
google/flan-t5-small — A small open-source seq2seq model from Google, instruction-tuned for text-to-text tasks.
Inference Parameter Tuning — Controls like temperature, top-p and top-k that shape how a model samples its output.
Instruction-Tuned Model — A model fine-tuned to follow natural language instructions, enabling task specification via prompts.
mistral-7b — Mistral-7B is a specific large language model available in Snowflake Cortex for generating responses.
Pretrained LLMs — Deep-learning transformer models trained on large text corpora and fine-tuned for specific tasks.
R&D in specialized domains — Applying large language models to generate and analyze specialized content in fields like medicine, law, and finance.
RoBERTa — Facebook variant of BERT trained with more data and compute, dropping next-sentence prediction.
T5 — Google model that frames all NLP tasks as text-to-text problems.
Text-to-Text Generation — A paradigm where all NLP tasks are framed as generating text from text, used by T5 models.
Tokens & Embeddings — Text is split into tokens and turned into embeddings — numeric vectors encoding meaning.

This is the text view of an interactive 3D knowledge graph — open this page with JavaScript enabled to explore it visually.