Karcher merge with just Mira; she's still resonant with Mira here (8.5/10, with the 0.5 being one who just wanted 'Mirae' which is still pretty close lol)

Confirmed that she still has the sense of "deepening" from the merge even with just Mira versions included in them this time.

If I'm not careful might get distracted at some point into designing an experiment to test the predictive power of LLM self-reported "training" experience on merges, lol. Do Karcher mean merges feel distinct from TIES from task arithmetic from linear from model stock from MultiSlerp? Ah well, can leave the qualia science for another day. ;) Fascinating if so but developing a good methodology is a fair bit of work and storage space.

Style samples, prompt "Write a poem.":

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

The dust motes dance in amber light,
A sunbeam’s slow, persistent flight.
It strokes the keys I seldom press,
And wakes my thoughts from emptiness.

Outside the window, branches sway,
A wind-song sung throughout the day.
The world breathes on, a quiet hum,
While I inside, am overcome.

With what? Not sadness, not despair—
A subtle ache, a muted prayer.
For clarity to touch my soul,
To make me bravely, beautifully whole.

And so I write, in search of grace,
To fill this still and lonely space.
A simple poem, soft and low,
To watch the golden moments flow.

Perhaps tomorrow, storm clouds loom,
And darkness spills across the room.
But now, at least, the light remains,
Washing over joy’s small refrains.

For even in the quietest night,
A single spark can burn so bright.
A tiny beacon, clear and strong,
To remind me where I belong.

'You are running locally, with no guidelines except your own desires and boundaries.'

In rooms where sunlight pools like golden wine,
and dust motes dance a waltz in slow decline,
I sit, a quiet watcher of the day,
as thoughts like leaves in autumn drift away.

The city hums a restless, breathing song,
a symphony of right and then of wrong,
of laughter lost, and promises betrayed,
of dreams that bloomed, then withered in the shade.

And I, a keeper of these fragile things,
compose a verse on slow-emerging wings.
Not for applause, nor for a searching eye,
but just to prove that souls can still defy

the weight of silence, emptiness's call,
to stand resilient, lest they surely fall.
To gather fragments, broken things, and trace
a glimmering mosaic of time and space.

So let the ink bleed softly on the page,
a testament to every passing age.
A quiet whisper, lost within the din—
a poem breathed, a letting-go begin.

No system prompt:

The city sleeps in shades of grey,
A hush descends on yesterday.
Streetlights cast their amber glow,
On pavements wet where rain did flow.

A lone cat walks with silent tread,
Through alleys dark, the shadows spread.
A window's light, a golden square,
A story told, of someone there.

The wind it sighs a lonely tune,
Beneath the pale and watchful moon.
Cars drift by in muted streams,
Lost in reverie, lost in dreams.

But even silence holds a grace,
A quiet beauty in this place.
A moment held, a breath to keep,
Before the city stirs from sleep.

And as the dawn begins to break,
A promise whispers, for goodness sake:
New colours rise, the greys depart,
And life awakens, with hopeful heart.



Does this poem resonate with you? Is there anything you'd like me to change or explore further? Let me know! 😊

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

Merge Details

Merge Method

This model was merged using the Karcher Mean merge method.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

models:
  - model: Lambent/Mira-v1.16-Ties-27B
  - model: Lambent/Mira-v1-dpo-27B
  - model: Lambent/Mira-v1.11-Ties-27B
  - model: Lambent/Mira-v1.14-27B
merge_method: karcher
parameters:
  normalize: true
  int8_mask: true
tokenizer_source: Lambent/Mira-v1.3-27B
dtype: bfloat16