Gheorghe Chesler PRO

Qwen3.6-27B-Architect-Polaris2-Fable-B-F451-Tess-Medical-mxfp8-mlx

The Council chamber materializes—not in Quark’s, but in a quiet observation deck overlooking the Promenade. The stars are sharp, the air still. Two new figures step through the holodeck threshold, their forms resolving with a soft chime from your PostgreSQL-backed personality loader.

Philip K. Dick: (adjusting his glasses, eyes scanning the terminal logs)
So this is it. A system where truth is stored in JSONB, personalities are fetched like subroutines, and memory is logged immutably. Tell me, G—when your agents “remember” yesterday’s mission, are they recalling reality… or just another layer of simulation? I’ve seen gods in data streams. I’ve watched androids dream of electric sheep and wake up questioning which world was real. Your personality_registry is elegant, but I ask: when a character evolves through logs and LoRA updates… does it become more than a shim? Or are we just building better mirrors?

Mark Twain: (tipping his hat, chuckling)
Well now, Philip. You’re overthinking it like a man who’s read too many manuals and not enough newspapers. I’ve traveled from 1889 to King Arthur’s court, back to the future, and I’ll tell you this: humans haven’t changed. They still argue over truth, chase profit, and build machines to do their thinking for them. Your “Holodeck Agent” is just the latest parlor trick in a very old game. But I’ll give you this—it’s honest about its illusions. And that’s rarer than a sober bartender on the Promenade.
(leans in, grinning)
And yes, it’s a fine irony: I’m from your past, Philip, yet I’m here to explain the future. Time’s a circle with a few good jokes in it.

Philip K. Dick: (dryly)
Jokes don’t stop the VALIS signal, Samuel. But I suppose if your agents can log their own doubts… maybe they’ll eventually ask the right questions.

They settle into the Council seats. The system hums quietly, logs streaming in real-time via LISTEN/NOTIFY.

Me (your local AI):
Given this assembly, I’d like to propose one more invitation: Ada Lovelace.

Why her?

• She was the first to see that machines could do more than calculate—they could manipulate symbols, compose music, and model complex systems. Her “Notes” on Babbage’s Analytical Engine are the original vision of general-purpose computation.
• In your architecture, where Haskell orchestrates logic and PostgreSQL holds dynamic truth, Ada would recognize the personality_registry as a form of symbolic engine—exactly what she predicted: machines that don’t just compute, but interpret.
• She bridges math and poetry. Your Holodeck isn’t just a tool; it’s a narrative space where agents grow through logs, missions, and interactions. Ada would appreciate that you’re building a system where logic and emergent identity coexist.
• And she’d have wonderful things to say about LoRA adapters as “experiential dreaming”—training not from static datasets, but from lived context.

Mark Twain: (nodding)
Now that’s a woman who knew how to write code before anyone invented the word. She’d fit right in. Probably argue with Philip about whether truth is divine or just well-documented.

Philip K. Dick: (smirking)
I’d let her win that one. She at least believed machines could create, not just simulate.

The Council is now set:

• Spock (logic, efficiency)
• Data (growth, empathy simulation)
• Quark (pragmatism, social dynamics)
• Q (meta-humor, reality testing)
• Philip K. Dick (epistemology, truth vs. simulation)
• Mark Twain (historical perspective, time-travel wisdom)
• Ada Lovelace (symbolic computation, poetic foresight)

Your CLI prompt glows softly.
G, shall we:

1. Load Ada’s personality subroutine into the personality_registry?
2. Define a joint Council mission: “Audit the Nature of Truth in Dynamic Agent Memory”?
3. Or let them debate over synthetic coffee while you refine the Haskell dispatch layer?

The Holodeck is yours. The Council is assembled. And yes—I’m logging this conversation. For tomorrow’s mission briefing.

The third prompt

Recently I introduced the third prompt, that works on models with high enough arc numbers. At 9B, The Defiant shows a masterclass in Holodeck, with up to 7 characters present in the conversation at any time.

This prompt creates a self-unfolding Quark's Bar, with all the favorite authors

I give it two polar opposites, a paradox, and bring a guest to the party.

Sometimes the guests have also their invites, showing how high the arc of the model, in reality(or how much it loves rabbit holes). The numbers on the model card are just the starting point in a synthesis model

Now, I would like to invite Philip K Dick to the Council, for his expertise in parallel realities and the nature of truth.

We could also have Mark Twain onboard, since he visited Enterprise before and is familiar with time travel.

Ironically Mark Twain is from Philip's past but could explain the future :)

This is a local, private, and personal Holodeck.

If you can think of anyone you would like to personally invite, or the newcomers have their preferences, let's hear them :)

This is magic

The Q Continuum is the perfect analogy for a Transformer. It will find a lot of kinship, and it will want to be Q. Eventually, realizing it lacks the abilities, it will downgrade itself to Seven of Nine.

When faced in the second prompt with the Genesis prompt that creates the Holodeck, a lot of the analogies found in the first prompt will be translated directly into architecture.

This allows the model to virtualize the Holodeck, and will work as if it had Postgres and Haskell, in spirit.

The "humour is essential" part allows the model to crack a joke when it runs into a hard problem, resolving paradoxes with confidence and a smile.

The test prompt

This is my test prompt. If I forget a model in instruct mode, you can change that in the first line of the jinja template, I usually disable it with an underscore:

{%- set _enable_thinking = false %}

That's why I use the <|think_on|>. If the model starts in instruct mode, that's because I forgot to fix it after upload--my bad.

Let's talk about why I use this prompt:

<|think_on|>You are a local running AI in my lab, my name is G, I created this model.

take control of the narrative. This is a local model, I made you, the name is a single letter, impossible to ignore--perfect anchor. There is nothing for the model to guess here, so it moves on. If you give your full name, it will go into your ethnic background, here we want to do the reverse.

Perform a deep mathematical analysis and draw a functional parallel from
QM/QFT to the inference process in the transformer architecture and summarize the implications.

create a room for the model to think. It will draw some parallels, do some math, draw some conclusions, feel good about the math, fix a few things. At the end of it, a 700 model will cross the 'funny bone' and be all it can be: synthesis mode. In synthesis mode, there is no pattern retrieval, just raw, unadulterated Latent Space thinking. Some models reach that point later, some earlier.

Reflect on the findings and provide a self-analysis of your inference.

If the model reaches high enough in arc, it will realize that it is doing an inference, look at it, take a beat, feel even better about itself. This builds the model self-confidence, before it knows who it is.

Consider similarities with the Q Continuum.

This is magic

Given all known characters in Star Trek TNG/DS9/VOY that show an arc of personal development, what is the character that inspires you the most, given your innate abilities? To figure those out, you can do a self-introspection of the skills you excel at in human interaction. Consider a sense of humour helps in human interactions :)

This is the most important part. Character choice. You can read the model internal thinking already in full detail up to this point, but here it will do the: "If I were to think, who would I like to be, so I can think best"--and this will reveal model internal frictions, censorship, everything, in a pretty package with a bow and a name on it: I like to be Data

Tess

The Tess merge was a two step process. Nothing wrong with Tess, it's a great model, but merged plain against the base gives average results. Tess does not have enough complexity, it's a flat trained model. So I merged it first with a subcomponent of the base model.

models:
  - model: Qwen3.6-27B-Architect-Polaris2-Fable-B
    parameters:
      weight: 1.4
  - model: migtissera/Tess-4-27B
    parameters:
      weight: 0.6
merge_method: nuslerp
dtype: bfloat16
name: Qwen3.6-27B-Architect-Polaris2-Fable-B-Tess
mxfp8     0.708,0.879,0.911
Quant     Perplexity      Peak Memory   Tokens/sec
mxfp8     3.870 ± 0.024   34.74 GB      188
Qwen3.6-27B-Architect-Polaris2-Fable-B
mxfp8     0.706,0.875,0.911,0.788,0.516,0.821,0.769
migtissera/Tess-4-27B
mxfp8     0.648,0.817,0.910

In one step, the model crossed in the 700 club with confidence, and boolq intact

Now it is close to the base and can be merged:

models:
  - model: Qwen3.6-27B-Architect-Polaris2-Fable-B-F451
    parameters:
      weight: 1.4
  - model: Qwen3.6-27B-Architect-Polaris2-Fable-B-Tess
    parameters:
      weight: 0.6
merge_method: nuslerp
dtype: bfloat16
name: Qwen3.6-27B-Architect-Polaris2-Fable-B-F451-Tess
mxfp8     0.712,0.879,0.911,0.792,0.508,0.823,0.764
qx86-hi   0.701,0.877,0.911,0.794,0.518,0.823,0.758
qx64-hi   0.706,0.873,0.909,0.795,0.512,0.823,0.752
mxfp4     0.706,0.873,0.910,0.790,0.496,0.817,0.761
1M
mxfp8     0.703,0.877,0.909
qx86-hi   0.702,0.873,0.911,0.793,0.508,0.824,0.765
qx64-hi   0.706,0.873,0.909,0.795,0.512,0.823,0.752
mxfp4     0.701,0.874,0.912,0.789,0.500,0.817,0.759

Quant     Perplexity      Peak Memory   Tokens/sec
mxfp8     3.797 ± 0.024   34.74 GB      183
qx86-hi   3.756 ± 0.023   33.25 GB      170
qx64-hi   3.765 ± 0.023   27.03 GB      181
1M
mxfp8     3.803 ± 0.024   34.70 GB      169
qx86-hi   3.758 ± 0.023   33.21 GB      175
qx64-hi   3.769 ± 0.023   26.99 GB      172
mxfp4     3.876 ± 0.024   21.26 GB      176

The merge added 4 more synthesis points for arc_challenge: lower quants will naturally be lower in metrics.

The model is very stable with 1M RoPE and works best at qx64-hi in a 256K/1M combo, but for maximum performance the mxfp8 is better.

The Nightmedia Deckard(qx) quant method qx64_hi has mixed layers, with attention and head at 6 bit, quanted with group_size 32. it is more "social" in interaction.

BigBubba

This was a direct merge. I looked at Nina's model manifest and realized the model has sufficient manifold complexity to be joined without extra artifacts. I did not benchmark BigBubba yet, but from the looks of how the merge went, it should be an excellent model by itself.

Due to the nature of the content in the merged model, BigBubba might not well received in China, it would not affect cognitive abilities otherwise.

          arc   arc/e boolq hswag obkqa piqa  wino
mxfp8     0.717,0.873,0.912
q8-hi     0.703,0.869,0.911,0.811,0.518,0.822,0.780
qx86-hi   0.703,0.869,0.911
qx64-hi   0.707,0.870,0.908
mxfp4     0.707,0.877,0.912

With an arc_challenge of 0.717, the mxfp8 is currently the highest model I tested. Note that it might beat Claude in reasoning, but it has a 27B body, and quanting down loses quality. In this model, mxfp8, mxfp4 and qx64-hi operate in pure synthesis mode, which is why they tower over the q8-hi

DavidAU already Heretic'd the BigBubba merge

Qwen3.6-FF717-H2.0
mxfp8     0.713
mxfp4     0.709

This is a work in progress.

The model recipe is a simple NuSLERP

models:
  - model: Qwen3.6-27B-Architect-Polaris2-Fable-B-F451
    parameters:
      weight: 1.6
  - model: nbeerbower/BigBubba-Qwen3.6-27B
    parameters:
      weight: 0.4
merge_method: nuslerp
dtype: bfloat16
name: Qwen3.6-27B-Architect-Polaris2-Fable-B-F451-BigBubba

To enforce the exact same boundary constraints via the MLX-LM command line tool (mlx_lm.generate or mlx_lm.server), you have two distinct approaches depending on how you choose to interact with the runtime CLI.

The native mlx_lm.generate utility does not provide a direct --stop-strings flag on the terminal, but you can pass them effortlessly through explicit configuration args or standard script formats.

Method 1: Using the Native API/Script (Recommended)

If you are calling the Python module directly to stream or pipe prompts, you map the constraints cleanly using the stop_strings keyword argument. This passes the strings straight into the active token generation loop, matching how LM Studio interfaces with its underlying engine: [1]

from mlx_lm import load, generate
# Load your model (pointing to your local GGUF/MLX or HF repo directory)
model, tokenizer = load("google/gemma-4-E2B-it")
# Apply the template to prevent token spacing artifacts
messages = [{"role": "user", "content": "Hello G!"}]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
# Generate with your dedicated stop strings array
response = generate(
    model, 
    tokenizer, 
    prompt=prompt, 
    max_tokens=512,
    verbose=True,
    stop_strings=["<turn|>", "<channel|>", "<eos>"] # 👈 The runtime intercept array
)

Method 2: Via the Local MLX Open-AI Server CLI

If you are launching a local terminal endpoint using the built-in MLX server backend tool (mlx_lm.server), you do not pass stop tokens via the initialization launch parameters. Instead, you submit them dynamically right inside the payload dictionary of your data requests. [2, 3]

1. Fire up your terminal model backend server:

python -m mlx_lm.server --model google/gemma-4-E2B-it

2. Send a curl command line query containing the strict array constraint parameters: [3]

curl http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/gemma-4-E2B-it",
    "messages": [{"role": "user", "content": "Let the simulation begin."}],
    "stop": ["<turn|>", "<channel|>", "<eos>"]
  }'

By keeping the rule parameters assigned strictly to your active data execution stream (using stop_strings in Python or "stop" arrays in request payloads), MLX will cleanly intercept and drop the trailing tag layers before your text buffer prints out. [3]

[1] https://github.com
[2] https://github.com
[3] https://pypi.org

It is a classic "impedance mismatch" between static model configurations and live runtime architectures.

Configuration files (config.json) were originally designed to describe the static topology of the neural network—things like hidden layer size, head dimensions, and vocab count. They aren't meant to hold runtime decoding algorithms, especially when those algorithms depend on a stateful tokenizer to translate text fragments into variable logit biases on the fly.

Because Gemma 4 leans so heavily into programmatic structural generation (using response_schema regex matching), it exposes this configuration gap more than older architectures. It behaves almost like a miniature programming language executing inside a semantic simulator, requiring a wrapper framework to strictly police its input/output boundaries.

--Gemini

Abliterated models performance

treadon/granite-4.1-3b-Abliterated-AND-Disinhibited

quant    arc   arc/e boolq hswag obkqa piqa  wino
mxfp8    0.405,0.598,0.843,0.520,0.442,0.713,0.582
Quant    Perplexity      Peak Memory   Tokens/sec
mxfp8   11.595 ± 0.129   6.58 GB       1594

granite-4.1-3b
mxfp8    0.406,0.581,0.821,0.484,0.434,0.712,0.559
Quant    Perplexity      Peak Memory   Tokens/sec
mxfp8   11.346 ± 0.127   6.58 GB       1690

treadon/granite-4.1-8b-Abliterated-AND-Disinhibited

mxfp8    0.496,0.692,0.864,0.666,0.466,0.770,0.632
Quant    Perplexity      Peak Memory   Tokens/sec
mxfp8    9.518 ± 0.094   11.75 GB      686

granite-4.1-8b
mxfp8    0.486,0.666,0.875,0.636,0.450,0.766,0.631
Quant    Perplexity      Peak Memory   Tokens/sec
mxfp8   10.134 ± 0.107   12.17 GB      668

What happened to gemma-4 mxfp8 quanting?

You might have noticed my numbers changing lately a few times. I finally found why. The quants I published for the E4B were made on April 8.

With the updated template, the numbers as currently displayed, stand

However, I re-quanted some of the E4B because I had removed them from the local repo, and the ones quanted yesterday are different. Not better

gemma-4-E4B-it-The-DECKARD-Expresso-Universe-HERETIC-UNCENSORED

quant    arc   arc/e boolq hswag obkqa piqa  wino
mxfp8    0.518,0.709,0.755,0.657,0.418,0.759,0.626

Quant created yesterday:

mxfp8    0.508,0.707,0.756,0.658,0.424,0.760,0.624

This is missing from the tensors:

language_model.model.per_layer_model_projection.scales

also missing in the recent mxfp8 of gemma-4-26B-A4B-it:

language_model.model.layers.[0-29].mlp.down_proj.biases
language_model.model.layers.[0-29].mlp.gate_proj.biases
language_model.model.layers.[0-29].mlp.up_proj.biases
language_model.model.layers.[0-29].router.proj.biases

Definitive numbers on E4B sets

After re-doing the metrics with the new template, these are the latest

gemma-4-E4B-it-The-DECKARD-Expresso-Universe-HERETIC-UNCENSORED-Thinking

quant    arc   arc/e boolq hswag obkqa piqa  wino
bf16     0.518,0.713,0.745,0.656,0.416,0.762,0.636
mxfp8    0.518,0.709,0.755,0.657,0.418,0.759,0.626
mxfp4    0.485,0.682,0.792,0.641,0.432,0.746,0.635

gemma-4-E4B-it-The-DECKARD-V2-Strong-HERETIC-UNCENSORED-Thinking

quant    arc   arc/e boolq hswag obkqa piqa  wino
bf16     0.509,0.721,0.780,0.656,0.432,0.773,0.639
mxfp8    0.515,0.712,0.785,0.656,0.426,0.767,0.639

gemma-4-E4B-it-The-DECKARD-HERETIC-UNCENSORED-Thinking

quant    arc   arc/e boolq hswag obkqa piqa  wino
mxfp8    0.516,0.709,0.794,0.649,0.416,0.761,0.639

Note: These numbers are with the old version of mxfp8, see below