Alternatives to Transformers

Apache 2.0 open-weight model family from TII Abu Dhabi, spanning hybrid Transformer-Mamba, Arabic, reasoning, and multimodal vision models.

Open-source Dutch RoBERTa/NeoBERT language models you fine-tune yourself, including the EU AI Act-compliant RobBERT-2026.

Open-source Python library for constraining LLM output with regex, context-free grammars, and inline control flow — archived since October 2023.

Open-source add-on to Hugging Face Transformers for training, composing, and sharing parameter-efficient fine-tuning adapters.

Predibase is a managed platform for fine-tuning and serving open-source LLMs, now part of Rubrik.

Together Compute is an AI-native cloud for running open-source models — serverless inference, batch jobs, model shaping, and GPU clusters in one platform.

Open-source TypeScript toolkit for building AI apps and agents across 100+ models with streaming, tools, and fallbacks

Enterprise AI platform that cuts token cost for agent workloads by routing across models and tuning small open models to frontier quality.

Open-source Python library that constrains LLM decoding with finite-state machines so Pydantic, JSON Schema, regex, and grammar outputs come out valid every

Liquid AI's open-weight LFM2.5 model family runs native text, vision, and audio AI locally on CPU, GPU, or NPU.

Open-source Python library that turns type-annotated functions into testable, validated LLM calls.

An open-source Ruby gem that puts 17 AI providers — chat, vision, audio, agents — behind one consistent API.

Open-source omni-modal LLM demo on Hugging Face that fuses speech, text, and vision with zero-shot voice personalization

Open-source PyTorch toolkit for compressing transformer NLP models via knowledge distillation.

Open-source Java tokenizer library implementing OpenAI's BPE encodings for local token counting, encoding, and decoding

Databricks' open-source instruction-following LLM, built on Pythia-12B and fine-tunable on a single A100 in about 30 minutes.

Linum is a two-person research lab that trains text-to-video models from scratch and releases the weights under Apache 2.0.

Open-source Go SDK giving one unified API across 25+ LLM providers with streaming, typed structured output, and MCP support.

Open-source JavaScript library that runs small LLMs like Llama 3.2 1B Instruct inside the browser via WebAssembly and WebGPU.

Drop-in AI autocomplete for any text input — two lines of code, OpenAI models underneath.

Paper Instruments trains open-source, domain-specific frontier models and agent tooling for specialist knowledge work.

Reka builds omni models for real-time video reasoning that run on-device, not just in the cloud.

StepFun ships frontier and open-weight multimodal models plus a full voice stack — Realtime, ASR, TTS and music — behind one OpenAI-compatible endpoint.

Zhipu AI builds the open-source GLM model family and a full-stack MaaS platform for coding, multimodal, and long-horizon agent work.

Frontier code models built for ultra-long-context software engineering and AI research automation.

Mistral sells sovereign AI: frontier open-weight models you can own, self-host, or run on EU inference.

Open-source multimodal model with native function calling, built for agents that both see images and take action on them.

ModelScope is Alibaba Cloud's open-source Model-as-a-Service hub for finding, fine-tuning, and deploying AI models.

Baichuan-7B is a free 7B bilingual Chinese-English base LLM released in 2023 on Hugging Face for fine-tuning and benchmark research.

Open-source on-premise RAG framework from the team behind Zylon — 57k GitHub stars, 100% local document Q&A.