Model catalog

Choose a recipe

FastFlowLM curates the most requested families and publishes tuned manifests under flm pull <model>. We validate every build on Ryzen™ AI laptops and provide matching cards—see them here. The catalog already covers models spanning Qwen3.8 (27B, MTP and DFlash 2 speculative decoding), Qwen3.6 (MoE), Qwen3.5 (Vision), Gemma4 (omni), Nemotron Bolt (DFlash speculative decoding), Qwen3, GPT-OSS-20B (MoE), DeepSeek-R1, Whisper (Audio), Hy-MT2 (Translation), MiniCPM-V 4.7 (Vision, 1B and 35B-A3B MoE), MedGemma (Medical), Decision 3.0 (d3-edge, d3-lite, d3-nano, d3-mini, d3-flash, d3-turbo, d3-ultra, the System One decision model of vLLM Semantic Router), and other leading open families.

View model docs

Supported families

Each manifest describes quantization, context window, tokenizer, and recommended memory, so there are no surprises after download.

Qwen3.8 (Vision, MTP and DFlash 2 speculative decoding) Qwen3.6 (Vision and MoE) Gemma 4 (omni) Nemotron Bolt (NVIDIA, tool use; DFlash speculative decoding) Qwen3.5 (Vision) Qwen 3 (Vision) MiniCPM-V 4.7 (Vision, incl. 35B-A3B MoE) GPT-OSS (MoE) LFM2 / LFM3 (Liquid AI) EmbeddingGemma (RAG) Whisper (Audio) Hy-MT2 (Translation) SmolVLA (Robotics) Decision 3.0 (System One) MedGemma (Medical) DeepSeek R-1 LLaMA 3.x

Roadmap

Upcoming drops

Follow the GitHub project for rolling releases. We publish nightly builds for community testing and stable channels every few weeks.

Fill this in

Coming soon!

Stay in the loop

Release notes Discord