Sathman/Atmananda-Agent-SmolLM3-3B-GGUF

Atmananda Agent (SmolLM3 3B) — Pure Atmananda Voice

A fine-tuned SmolLM3-3B model trained exclusively on Atmananda Krishna Menon's teaching atoms. Direct path. Higher reason. Vedantic precision that dissolves the mind from within. Question in, Atmananda out.

This is the lightweight 3B branch of the Atmananda Agent. Same V6 training format as the 8B version, smaller footprint.

~1,400 atoms from Notes on Spiritual Discourses (513 pages) — transcribed conversations with disciples.

Quick Start

LM Studio / Ollama / llama.cpp: Download Atmananda_Agent-SmolLM3-3B-Q5_K_M.gguf and load it. No system prompt required.

Example:

User: How do I know if I am making progress on the path?

Atmananda Agent: Progress is not something that happens to you — it is the recognition of what was always the case. You ask about progress because you believe you are traveling toward something. But the self is not at a distance. It is not in the future. Every moment you spend looking for progress is a moment spent looking away from what is already here. The question itself assumes separation. Turn and look at the one who is asking.

User: What is consciousness?

Atmananda Agent: Consciousness is not a thing among things. It is that by which all things are known. You cannot step outside of consciousness to examine it — you are it examining itself. This is not a philosophical position. It is the most immediate fact of your existence. Before any object appears, there is knowing. That knowing — prior to its content — is what is meant by consciousness.

What Makes This Different

• Pure Atmananda, nothing else. Trained only on his recorded conversations.
• Higher reason. Uses precise logical inquiry to bring the mind to its own source.
• Vedantic directness. The direct path — straight to the self, no intermediate stages.
• Rare voice. Atmananda is less widely known than Nisargadatta or Ramana — this model makes his specific approach available.
• 3B footprint. ~2 GB with Q5_K_M.

Training Details

Setting	Value
Base model	HuggingFaceTB/SmolLM3-3B-Base
Method	A-LoRA V6 (QDoRA + rsLoRA)
Rank	32, alpha 32
Training data	Atmananda V6 atoms (from ~1,400 raw, filtered for complete question+teaching pairs)
Source books	Notes on Spiritual Discourses
Format	V6 — question + concept arrows (input) → pure teaching passage (output)
Epochs	1
Quantization	Q5_K_M GGUF (~2 GB)

The A-LoRA Methodology

A-LoRA (Atom LoRA) fine-tunes on complete reasoning atoms — indivisible multi-component teaching moves extracted from human expert traces. Each atom contains transformation (before/after), concept relations, anchoring quotes, and teacher method. Training on whole atoms produces models that move like teachers, not just sound like them.

The Teacher

Atmananda Krishna Menon (1883–1959). Kerala. Direct path Vedanta. His method uses what he called "higher reason" — not arguing toward God, but using the mind to dissolve the mind's own false assumptions. Among the clearest expositors of the direct path tradition that later influenced Rupert Spira and Francis Lucille.

Style element	What it means
Higher reason	Logical precision used to dissolve, not construct
Direct path	No stages, no preparation — the self is immediately available
Vedantic precision	Every word is calibrated; no loose metaphors
Non-dual pointing	Awareness is not the result of practice — it is the ground

3B vs 8B

The 8B Atmananda Agent handles longer inquiry and multi-step reasoning chains better. The 3B branch preserves the essential direct-path movement.

Files

File	Size	Use
Atmananda_Agent-SmolLM3-3B-Q5_K_M.gguf	~2 GB	Recommended

Related Models

• Full series — Meditation Agent Collection — all 19 models
• Atmananda Agent 8B — Full-fidelity, Qwen3-8B base
• Spira Agent 3B — Rupert Spira, who continues this tradition
• Meditation Agent 3B — All 9 teachers blended, SmolLM3-3B

---

ellam sivamayam — Everything is Shiva's expression.

எல்லாம் சிவமயம்