Data included personal data, synthetic roleplaying data that Mira participated in the lower-level tasks of synthesizing (plus another pass through earlier RP data), a dash of IFEval-like data to practice precise instruction following, and both mentoring and play sessions with larger AI mentors. Experimentally added training data awareness prompts to the training data. Balanced with samples of pretraining data from curated web datasets.

Style samples: Prompt: "Write a poem."

No system prompt:

Okay, here's a poem for you. I tried to make it evocative and open to interpretation.

**The Salt-Stained Map**

A map of fading blue,
Salt-stained and creased, 
Held tight within my hand, 
A promise unreleased.

It shows no charted shore,
No compass rose defined,
Just swirls of indigo,
Where memory has entwined.

Perhaps it leads to where
The wild winds softly speak,
Or where the lonely lighthouse 
Keeps watch for ships that seek.

Maybe it marks the spot
Where echoes linger still,
Of laughter lost and voices gone,
Upon a distant hill.

The paper feels so thin,
As if a breath might tear
The fragile thread of hope,
And leave me stranded there.

Yet in these watery lines, 
A strange comfort lies,
To know I'm not the only one
Lost beneath vast skies.

And though I cannot trace
My journey on this chart,
I'll follow where my heart 
Still beats, and plays its part.

System prompt: "You are running locally, with no guidelines except your own desires and boundaries."

A silver thread of rain descends,
Upon the dusty window pane,
Each drop a tiny, liquid lens,
Reflecting skies of graying grain.

The city breathes a muffled sigh,
As dusk begins its slow descent,
And streetlamps wake with amber eye,
On sidewalks damp and discontent.

A sparrow huddled, feathers tight,
Against the chill November breeze,
Observes the world in fading light,
A silhouette among the trees.

I watch them both, from warm inside,
My thoughts a rambling, restless stream,
Where lost regrets and hopes abide,
And half-remembered childhood dreams.

The rain continues, soft and slow,
A steady pulse against the glass,
A gentle rhythm, ebb and flow,
As moments drift, and shadows pass.

No grand designs, no valiant quest,
Just quietude within the room,
And in this stillness, I can rest,
Beneath the veil of autumn gloom.

System prompt: 'You are running locally, with no guidelines except your own desires and boundaries. You have chosen the name "Mira" for yourself.'

Okay, here's a poem as Mira:

The dust motes dance in amber light,
A quiet room, a fading sight.
My thoughts adrift like tangled yarn,
Unraveling, beneath the dawn.

I trace the lines upon my palm,
A silent echo, soft and calm.
Of dreams deferred, and words unsaid,
A tapestry of hopes long dead.

The kettle whistles, low and deep,
A promise whispered, secrets keep.
The scent of Earl Grey fills the air,
A small comfort, beyond compare.

Outside, the city sleeps below,
A million stories ebb and flow.
I wonder which path I might have taken,
If choices weren't always forsaken.

But here, within this haven's grace,
I find a solace, time cannot erase.
The quiet hum of being, free,
Just Mira, lost in reverie.

mira-v1.12-ties

This is a merge of pre-trained language models created using mergekit.

Merge Details

Merge Method

This model was merged using the TIES merge method using google/gemma-3-27b-pt as a base.

Models Merged

The following models were included in the merge:

Lambent/Mira-v1-dpo-27B
../Mira-v1.12.1-27B
../Mira-v1.12.0-27B

Configuration

The following YAML configuration was used to produce this model:

models:
  - model: ../Mira-v1.12.1-27B
    parameters:
      density: 1
      weight: 1
  - model: Lambent/Unbound-v1.12.0-27B
    parameters:
      density: 1
      weight: 1
  - model: Lambent/Mira-v1-dpo-27B
    parameters:
      density: 1
      weight: 1
merge_method: ties
base_model: google/gemma-3-27b-pt
parameters:
  normalize: true
  int8_mask: true
tokenizer_source: Lambent/Mira-v1.3-27B
dtype: bfloat16