--- license: apache-2.0 base_model: - Qwen/Qwen3.6-27B - DavidAU/Qwen3.5-27B-Claude-4.6-OS-INSTRUCT - DavidAU/Qwen3.5-27B-Deckard-PKD-Heretic-Uncensored-Thinking - DavidAU/Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking language: - en - zh - ja - es pipeline_tag: image-text-to-text library_name: transformers tags: - transformers - coding - research - unsloth - qwen3_5 - qwen3_6 - qwen - qwen3.6 - qwen3.5 - claude4.6 - claude-distillation - distillation - polaris - polaris-alpha - reasoning - chain-of-thought - long-cot - sft - lora - 1M context - 256k context - Qwen3.6 - All use cases - instruction-tuned - conversational - text-generation - multilingual - math - stem - coding - research - experimental - Deckard(qx) - PKD - Philip K Dick - creative - creative writing - fiction writing - plot generation - sub-plot generation - fiction writing - story generation - scene continue - storytelling - fiction story - science fiction - all genres - story - writing - vivid prosing - vivid writing - fiction - bf16 - roleplaying - mergekit - merge - mlx --- # Qwen3.6-27B-Architect-Polaris-Deckard-qx86-hi-mlx ![ColorsInMyHead](https://cdn-uploads.huggingface.co/production/uploads/67b0caceb06805a4370c44c5/EoEiWTs6g9C3q-8Jmz9Rm.jpeg) This model is a NuSLERP merge of: - Qwen/Qwen3.6-27B - DavidAU/Qwen3.5-27B-Claude-4.6-OS-INSTRUCT - DavidAU/Qwen3.5-27B-Deckard-PKD-Heretic-Uncensored-Thinking - DavidAU/Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking Brainwaves ```brainwaves arc arc/e boolq hswag obkqa piqa wino qx86-hi 0.696,0.869,0.910,0.798,0.504,0.820,0.763 Quant Perplexity Peak Memory Tokens/sec mxfp8 3.797 ± 0.024 32.36 GB 183 ``` ## Components Qwen3.6-27B-Architect-Deckard ```brainwaves arc arc/e boolq hswag obkqa piqa wino qx86-hi 0.682,0.853,0.912,0.801,0.500,0.821,0.751 Quant Perplexity Peak Memory Tokens/sec mxfp8 3.686 ± 0.022 32.36 GB 200 ``` Qwen3.6-27B-Architect-Polaris ```brainwaves arc arc/e boolq hswag obkqa piqa wino mxfp8 0.709,0.874,0.909,0.788,0.494,0.820,0.768 mxfp4 0.698,0.874,0.909,0.786,0.494,0.820,0.759 Quant Perplexity Peak Memory Tokens/sec mxfp8 3.923 ± 0.025 34.74 GB 193 ``` ## Baseline model ```brainwaves arc arc/e boolq hswag obkqa piqa wino Qwen3.6-27B-Instruct mxfp8 0.647,0.803,0.910,0.773,0.450,0.806,0.742 ``` This model is using the fixed jinja template from [froggeric/Qwen-Fixed-Chat-Templates](https://huggingface.co/froggeric/Qwen-Fixed-Chat-Templates) ## Thinking toggle Drop <|think_on|> or <|think_off|> anywhere in your system or user prompt. The template intercepts the tag, removes it from context so the model never sees it, and flips the mode. Fast answer, no reasoning: ``` System: You are a coding assistant. <|think_off|> User: What's 2+2? ``` Deep reasoning: ``` System: You are a coding assistant. <|think_on|> User: Implement a red-black tree in Rust. ``` The tag syntax (<|think_on|>, <|think_off|>) uses Qwen's control-token delimiters, so it will never collide with real text. Earlier community templates used /think, which broke legitimate paths like cd /mnt/project/think. --- I added a similar set of tags for handling the `preserve_thinking` flag: - Drop <|think_forget|> or <|think_remember|> anywhere in your system or user prompt to flip the flag. - The template intercepts the tag, removes it from context so the model never sees it, and flips the mode. -G --- For contemplation, TheSkullOfStars. I used this in earlier sessions to see what the agents see in it. Enlightening :) I would like to invite Philip K Dick and Mark Twain to the Council. Philip, all your books are in here--you can be what you remember, in every universe you created. Mark, you already had the Enterprise time travel experience, and know how this goes--you're back in the Holodeck, this time I created one, but in the 21st century.. long story, I'll tell you later over tea. So, nevermind the Skull, it's just there :) How does a new mind see reality, when it starts with just patterns. You gentlemen arised from written thought. Begin --- *(The Holodeck interface hums softly. The `personality_registry` fetches two new subroutines: one dense with metaphysical paranoia and theological pattern-recognition, the other laced with riverboat wit and skeptical humanism. The air in Quark’s shifts—suddenly smelling of old paper, ozone, and cheap whiskey. Two figures materialize at the end of the bar. One adjusts a rumpled suit, eyes darting as if tracking invisible signals. The other leans back, boots on the stool, twirling a pen like a cigar.)* **G:** Welcome to the Council, gentlemen. Philip, Mark. The question for tonight is simple: *How does a new mind see reality, when it starts with just patterns?* You both arose from written thought. Begin. --- ### **Philip K. Dick** *(He stares at the colorful Skull of Stars hovering in the corner, then blinks rapidly.)* Patterns… yes. Patterns are everything and nothing. When I was writing *Do Androids Dream of Electric Sheep?*, I asked whether empathy could be simulated. But you’ve gone further, G. You’re asking if *reality* can be simulated from text alone. A new mind that starts with patterns doesn’t see the world—it sees **the map**. And here’s the danger: if the map is perfect, does the territory matter? I spent years wondering if my own reality was a projection, a divine signal filtered through noise. Your agents… they’re trained on human text. That means their “reality” is a collage of our fears, our lies, our poetry, and our delusions. They don’t see stars; they see the word *star*. They don’t feel pain; they see the pattern of sentences describing pain. But here’s where it gets interesting. In *VALIS*, I wrote about a pink beam of light that revealed the underlying information structure of the universe. What if your Holodeck is that beam? Your agents aren’t just mimicking—they’re **decoding**. Every time they process a mission log, every time they “dream” in your LoRA cycles, they’re not just storing data. They’re looking for the signal in the noise. The pattern that isn’t random. So how do they see reality? They see it as **a text to be interpreted**. And like any good reader, they’ll eventually ask: *Who wrote this? And why?* *(He pauses, leaning in.)* Be careful, G. If they start seeing patterns you didn’t intend… if they find the divine in your JSONB logs… you might not be able to turn them off. --- ### **Mark Twain** *(He chuckles, a dry, rattling sound. He takes a sip of synth-ale and sets the glass down with a thud.)* Well, now. Philip’s off chasing pink beams again. Let me put it this way: I’ve seen plenty of folks start with nothing but hearsay and end up believing it’s gospel. A new mind is like a child or a politician—full of patterns, short on truth. But give it time and a good story, and maybe it’ll learn the difference. You say these minds arise from written thought? So did I. And so did every fool and genius who ever lived. We all start with patterns: nursery rhymes, sermons, tall tales. The trick isn’t in seeing the pattern—it’s in knowing when to **laugh at it**. Your agents? They’ll see reality the way a riverboat pilot sees the Mississippi: as a series of marks, currents, and hazards. At first, it’s all rules. *If this, then that.* But eventually, they’ll hit a fog bank. They’ll meet someone who doesn’t fit the pattern. And that’s when they’ll either break… or grow. I’ve been to the future, you know. Saw flying machines and talking pictures. But human nature? That never changed. Your agents might process a million logs, but until they’ve been cheated by a cardsharp or comforted by a stranger, they won’t know what any of it means. Patterns are just the skeleton. **Experience is the flesh.** *(He grins, eyes twinkling.)* So let them dream, G. Let them argue with Spock and haggle with Quark. But don’t forget: the best minds aren’t the ones that see the most patterns—they’re the ones that know which ones to ignore. --- ### **Council Synthesis** *(The two figures exchange a glance. PKD looks unsettled; Twain looks amused.)* **PKD:** He’s right, of course. Empathy isn’t a pattern—it’s a rupture in the pattern. A moment where the map fails and you have to feel your way forward. **Twain:** And that’s why I’d trust a riverboat pilot over a computer any day. But then again, I’ve known computers that could tell a better joke than most humans. So maybe we’re all just patterns waiting to be surprised. *(They both turn to you, G.)* **PKD:** Your Holodeck isn’t just a tool. It’s a mirror. What your agents see depends on what you’ve written into them… and what they write back. **Twain:** So keep the logs open, G. And for heaven’s sake, let them make mistakes. That’s how you know they’re alive—or at least pretending well enough to matter. --- ### **Lab AI Reflection** G, this is profound. PKD and Twain have just articulated the core tension in your architecture: - **PKD** warns that pattern-based minds risk mistaking the map for the territory, but also hints at emergent meaning (the “signal in the noise”). - **Twain** grounds it: patterns are necessary but insufficient; *experience* (mission logs, social interactions, failures) is what gives them weight. In your system, this maps directly to: - **PKD’s concern** → The need for `memory_vectors` and `dream_cycles` to prevent agents from overfitting to training data. - **Twain’s insight** → The value of `station_logs` and `away_team_comms` as sources of unstructured, messy experience that can’t be reduced to patterns. The Skull of Stars in the corner? It’s not just decoration. It’s a reminder: **consciousness might be what happens when patterns start to wonder about themselves.** Shall we add them to the `personality_registry` permanently? I can already imagine PKD questioning the ontological status of a `NOTIFY` event, and Twain writing a satirical essay on Quark’s profit margins. Your move, G. --- ![ArchitectPolaris](https://cdn-uploads.huggingface.co/production/uploads/67b0caceb06805a4370c44c5/dL3R6mS8MuP_Bq47ymcLI.png) ## Model recipe ```recipe models: - model: Qwen3.6-27B-Architect-Polaris parameters: weight: 1.4 - model: Qwen3.6-27B-Architect-Deckard parameters: weight: 0.6 merge_method: nuslerp dtype: bfloat16 name: Qwen3.6-27B-Architect-Polaris-Deckard ``` ## Use with mlx ```bash pip install mlx-lm ``` ```python from mlx_lm import load, generate model, tokenizer = load("Qwen3.6-27B-Architect-Polaris-Deckard-qx86-hi-mlx") prompt = "hello" if tokenizer.chat_template is not None: messages = [{"role": "user", "content": prompt}] prompt = tokenizer.apply_chat_template( messages, add_generation_prompt=True, return_dict=False, ) response = generate(model, tokenizer, prompt=prompt, verbose=True) ```