Language & NLP

Pure C/C++ LLM inference library — run quantized models on CPU, Metal, CUDA and more.

The Natural Language Toolkit — Python's foundational NLP library for tokenization, POS tagging, parsing, and corpora.

State-of-the-art sentence and text embeddings — compute semantic similarity, clustering, and dense retrieval.

Fast, memory-mapped dataset library for NLP and ML — 50,000+ datasets, streaming, and Arrow-backed processing.

Run powerful open-source LLMs locally — privacy-first desktop app and Python SDK, no internet required.

Ultra-fast tokenizer library (Rust core) — BPE, WordPiece, SentencePiece — tokenize GBs in seconds.

Industrial-strength NLP library for Python with pre-trained pipelines for tokenization, NER, parsing, and more.

Run PyTorch training scripts on any hardware — single GPU, multi-GPU, TPU — with minimal code changes.

Microsoft's deep learning optimization library — train trillion-parameter models with ZeRO memory parallelism.

8-bit and 4-bit quantization for PyTorch — run and fine-tune LLMs on consumer GPUs with minimal quality loss.

Fine-tune LLMs with RLHF, PPO, DPO, SFT, and GRPO — the standard library for aligning language models.

Unified fine-tuning framework for 100+ LLMs — SFT, RLHF, DPO, LoRA, QLoRA via Web UI or CLI.

Apache-licensed 7B language model from Mistral AI — beats Llama 2 13B on most benchmarks at half the size.

Hugging Face Transformers: state-of-the-art pre-trained models for NLP, vision, audio, and multimodal tasks.

Facebook AI's library for efficient text classification and word representation learning — trains in seconds on millions of documents.

The largest open prompt collection — thousands of curated system prompts for personas, coding, writing, education, and creative tasks. CC0 licensed.

MIT-licensed SOTA English embedding model from BAAI — top MTEB leaderboard performer, commercial-friendly.

Microsoft's MIT-licensed 3.8B SLM — instruction-tuned, runs on CPU/mobile, punches far above its weight class.

384-dimensional sentence embeddings with tens of millions of downloads — fast, compact, and remarkably accurate for semantic search and clustering.

MIT-licensed comprehensive guide and prompt library — techniques, examples, and templates for every major LLM prompting method.

Modular topic modeling framework using transformer embeddings and c-TF-IDF for interpretable, coherent topics.

Alibaba's Apache-2.0 7B instruction model — top multilingual performance, 128k context, strong at coding and math.

Simple NLP library with SOTA models for NER, POS tagging, chunking, text classification, and contextual string embeddings.

Apache-2.0 text embedding model with 8192-token context — the first open, auditable, long-context embedding.

Hugging Face library for parameter-efficient fine-tuning — LoRA, QLoRA, IA3, and more. Fine-tune billion-parameter models on a single consumer GPU.

MIT-licensed 4.2M instruction dataset — GPT-4/3.5 augmented CoT traces that power top open-source fine-tunes.

Production-ready NLP pipeline framework for building search, RAG, and question-answering systems with any LLM.

Apache-2.0 fine-tuning framework — train any HuggingFace model with LoRA/QLoRA/full-fine-tune via a single YAML config.

CC-BY-SA-3.0 instruction dataset of 15k human-written prompts — the first commercially licensed open instruction dataset.

Stanford NLP's Python library for tokenisation, sentence segmentation, NER, dependency parsing, and coreference across 70+ languages.

Apache-2.0 RAG evaluation framework — faithfulness, answer relevancy, context recall, and more in one pip install.

All-in-one semantic search, RAG, and LLM workflow engine — embeddings, vector DB, and pipelines in one library.

Open-source annotation platform for building high-quality fine-tuning and RLHF datasets; integrates with Hugging Face Hub.

Integrates LLMs as spaCy pipeline components for NER, classification, lemmatisation, and relation extraction with zero/few-shot prompting.