---
language: en
license: apache-2.0
tags:
  - gguf
  - contemplative-ai
  - fine-tuned
  - nondual
  - teaching
  - spirituality
  - awareness
  - e-atoms
  - meditation-agent
  - v2
  - smollm3
  - krishnamurti
base_model: HuggingFaceTB/SmolLM3-3B-Base
pipeline_tag: text-generation
---

# Meditation Agent: Krishnamurti v2 (SmolLM3 3B)

A lightweight single-teacher contemplative model from the **Meditation Agent v2** line for Krishnamurti.

This is the `v2` successor to [Krishnamurti-Agent-SmolLM3-3B-GGUF](https://huggingface.co/Sathman/Krishnamurti-Agent-SmolLM3-3B-GGUF). It is trained on the newer **E-atoms** corpus, where the target is `teaching_e`, not the older atom surface.

Collections:

- New v2 collection: [Meditation Agent v2 - EAtoms Teacher Series](https://huggingface.co/collections/Sathman/meditation-agent-v2-eatoms-teacher-series)
- Previous collection: [Meditation Agent - Non-Dual Self-Realization Teacher Series](https://huggingface.co/collections/Sathman/meditation-agent-non-dual-self-realization-teacher-series)

## Why Download This

- **Single-teacher branch.** This model is tuned specifically for Krishnamurti, not the mixed multi-teacher voice.
- **Small local footprint.** The current release file is `Q5_K_M`, which runs comfortably on local setups that cannot host larger models.
- **Newer v2 corpus.** The `E-atoms` line is sharper, less synthetic, and less likely to collapse into generic spiritual assistant prose.

## Teacher Signal

This branch is tuned for choiceless awareness, dismantling of the observer, and fierce refusal of method dependence.

## Sample Outputs

These are from the local `5q` smoke eval on the finished merged model.

### Example 1

**Prompt:** `How does fear relate to thought?`

> Look at your fear--not just as an idea about it, but actually feel the quality of that contraction in you. Is there any movement in that? Or is there only resistance against something? Thought has created this whole structure called 'fear,' hasn't it? You think: I should not be afraid; I must overcome my fears through willpower or analysis. But look closely now--can you see how thought itself creates what it calls fear? The thinker and his thoughts are identical when he observes fear with judgment rather than direct perception. There's no separation between them then. So we're not dealing with a problem that needs solving; we've been looking at it all wrong from the beginning.


### Example 2

**Prompt:** `Is there a self?`

> What is this'self' you're asking about? Let us look at it directly without assuming anything. When I say something like, 'I am confused,' who is that observer separate from the confusion itself--the thinker and thought? Is not the thinker made entirely of thoughts? Thought creating thought--isn't that what we call consciousness? So when you ask if there's a self, perhaps you should examine whether this consciousness has any substance or independence beyond its own creation.


### Example 3

**Prompt:** `What do I do with anger when it arises?`

> When you are angry, what exactly are you experiencing? There is the sensation in your body -- that tightening of muscles, that flush of heat -- and then there's this mental state: frustration, resentment, justification. These two things are not separate events but one unified movement. You can't stop being angry without understanding both parts. The physical feeling needs attention too, like a wound that must be cleaned before healing can begin. But more crucially, look at how thoughts have built up around that anger over time -- the stories about who did this wrong to me, why they deserve my wrath, how this should make them suffer. Those narratives create the fuel for future explosions. So don't suppress or run from either part; examine them together as one thing. When you see clearly how anger operates through the body and mind simultaneously, something fundamental shifts.


## Training Summary

- **Base model:** `HuggingFaceTB/SmolLM3-3B-Base`
- **Training format:** `V6E`
- **Teacher examples:** `2,204`
- **Adapter recipe:** `QDoRA + rsLoRA`
- **Rank:** `32`
- **Epochs:** `1`
- **Completion-only loss:** enabled

## Run Metrics

- **Run name:** `teacher-smollm3-v6e-krishnamurti-l2-r32`
- **Training runtime:** `19.9 minutes`
- **Final train loss:** `1.6916`
- **Final eval loss:** `1.7245`
- **Final eval token accuracy:** `0.5588`

## Files

- `Krishnamurti_Agent-SmolLM3-3B-v2-Q5_K_M.gguf` (`2.2 GB`)

## Honest Limits

- This is still a compact `3B` model, so some broader contemplative phrasing can leak into non-teacher-specific prompts.
- The `5q` smoke eval is encouraging, but it is not a full benchmark or final quality proof.

## Public Naming

- **Public repo:** `Meditation-Agent-Krishnamurti-SmolLM3-3B-v2-GGUF`
- **Internal model folder:** `Krishnamurti_Agent-SmolLM3-3B-v2`