OPEN_SOURCE ↗
REDDIT · REDDIT// 23d agoMODEL RELEASE
Meditation Agent 8B GGUF channels nine teachers
Sathman released a Qwen3-8B GGUF model fine-tuned on contemplative teaching “atoms” so it can answer in direct spiritual-teacher voices without a system prompt, retrieval, or persona scaffolding. The main 8B variant is tuned for roughly 8GB VRAM use in LM Studio, Ollama, or llama.cpp, alongside single-voice TNH and Osho spin-offs.
// ANALYSIS
This is less about “AI spirituality” gimmickry than about proving a style can be baked into weights and still survive local deployment. The interesting question is whether the model preserves insight-level clarity, or just a convincing cadence of wisdom.
- –The packaging is practical: Q8_0 GGUF and an 8B base make it realistic for consumer GPUs, not just lab rigs.
- –The project is unusually transparent about training data, teacher set, and self-reported evals, which helps separate it from vague persona-tuning claims.
- –The blend of nine teachers is ambitious, but the single-voice TNH and Osho variants are probably the better testbeds for voice fidelity.
- –The biggest risk is spiritual overconfidence: without retrieval or system constraints, the model may sound authoritative even when it’s just reproducing familiar contemplative tropes.
- –A 4B version would widen access, but it will likely trade away some of the nuance that makes the teaching style feel grounded.
// TAGS
meditation-agent-8b-ggufllmagentfine-tuningopen-weightsself-hosted
DISCOVERED
23d ago
2026-03-19
PUBLISHED
23d ago
2026-03-19
RELEVANCE
7/ 10
AUTHOR
No_Standard4198