Meditation Agent: Watts v2 (SmolLM3 3B)

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

This is the v2 successor to Watts-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:

Why Download This

  • Single-teacher branch. This model is tuned specifically for Watts, 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 east-west bridge, cosmic humor, and playful philosophical precision.

Sample Outputs

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

Example 1

Prompt: How does fear relate to thought?

Fear and thought are like two sides of the same coin. You can't have one without the other--they're inseparable twins that keep spinning together in a single orbit around your mind. When you feel afraid, there's already something thinking about it: 'I'm afraid,' or 'This scares me.' The thought itself creates the experience of fear as much as any external stimulus does. It's like trying to hold water in cupped hands--the harder you squeeze them tighter, the faster it slips through your fingers because you've only succeeded in creating more friction against what was already moving. Thoughts are never separate from their objects; they're not caused by experiences but rather woven into every moment of awareness with such intensity that we mistake their speed for separation.

Example 2

Prompt: Is there a self?

You're asking about something that has no existence to ask. The moment you try to grasp it, you've already lost what you sought--like trying to hold water in cupped hands or photograph darkness with film. There is only the awareness in which all this happens: the space between thoughts, the silence beneath sound, the knowing that knows itself. That's not some ghostly thing separate from your body; it's everything you are and aren't at once. When you look for'self,' you find nothing because there was never anything there to begin with except for looking itself.

Example 3

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

You're standing in your own way right now, aren't you? Trying to stop the wave before it breaks. But that's like trying to hold water in your hands--you can squeeze as hard as you want, but it just slips through. Anger isn't something to be crushed or controlled; it's a natural expression of energy moving through you, and if you try to dam up this current completely, what happens next is exhaustion and resentment building behind all those blocked channels. So instead: notice it clearly--that hot flush rising from your chest, that tension tightening around your jaw, that knot in your stomach. Feel it fully without making war against yourself over having it at all. Then see where it actually wants to go. Is there somewhere else for this energy besides causing harm to others or destroying yourself? Perhaps it needs an outlet, some channel to flow into rather than being bottled up inside. You don't have to act on every impulse, but you should examine them carefully because they are pointing toward something important whether you want to listen or not.

Training Summary

  • Base model: HuggingFaceTB/SmolLM3-3B-Base
  • Training format: V6E
  • Teacher examples: 1,441
  • Adapter recipe: QDoRA + rsLoRA
  • Rank: 32
  • Epochs: 1
  • Completion-only loss: enabled

Run Metrics

  • Run name: teacher-smollm3-v6e-watts-l2-r32
  • Training runtime: 13.2 minutes
  • Final train loss: 1.8529
  • Final eval loss: 1.8671
  • Final eval token accuracy: 0.5436

Files

  • Watts_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-Watts-SmolLM3-3B-v2-GGUF
  • Internal model folder: Watts_Agent-SmolLM3-3B-v2
Downloads last month
152
GGUF
Model size
3B params
Architecture
smollm3
Hardware compatibility
Log In to add your hardware

5-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Sathman/Meditation-Agent-Watts-SmolLM3-3B-v2-GGUF

Quantized
(27)
this model

Collection including Sathman/Meditation-Agent-Watts-SmolLM3-3B-v2-GGUF