sexy README with itasha aesthetic + inline CSS

README.md

@@ -22,43 +22,63 @@ pipeline_tag: text-generation
 ---
 
 <div align="center">
-
-# 🧠 Behemoth-X-R1-123B
-
-### *A thinking beast that writes like a poet.*
-
-**An SCE merge of Behemoth-X and Behemoth-R1 — prose voice meets reasoning mind in one 123B parameter model.**
-
-[![Base](https://img.shields.io/badge/base-Mistral_Large_2411-orange)](https://huggingface.co/mistralai/Mistral-Large-Instruct-2411)
-[![Method](https://img.shields.io/badge/merge-SCE-purple)](https://arxiv.org/abs/2408.07990)
-[![Size](https://img.shields.io/badge/params-123B-red)]()
-[![Context](https://img.shields.io/badge/ctx-131k-blue)]()
-
+<img src="https://huggingface.co/tacodevs/Behemoth-X-R1-123B/resolve/main/assets/hero.png" alt="Behemoth-X-R1-123B" style="width:100%; max-width:960px; border-radius:16px; box-shadow:0 0 60px rgba(236,72,153,0.35), 0 0 100px rgba(139,92,246,0.25);"/>
 </div>
 
----
+<div align="center" style="margin-top:24px;">
 
-## ⚡ What makes this different
+<h1 style="font-size:3.2em; font-weight:900; background:linear-gradient(90deg,#ec4899 0%,#a855f7 50%,#06b6d4 100%); -webkit-background-clip:text; -webkit-text-fill-color:transparent; background-clip:text; margin:0; letter-spacing:-0.02em;">Behemoth-X-R1-123B</h1>
 
-Most "thinking" models sacrifice prose for reasoning. Most creative models can't think their way out of a scene. **Behemoth-X-R1 doesn't compromise** — it carries the distinctive voice and character depth of Behemoth-X into a model that can open a `<think>` tag and actually use it.
+<p style="font-size:1.3em; color:#a855f7; font-style:italic; font-weight:500; margin-top:8px;">A thinking beast that writes like a poet.</p>
 
-No LoRA. No retraining. Just **principled weight arithmetic** using the same SCE merge recipe that FuseAI used to preserve reasoning in [FuseO1](https://huggingface.co/FuseAI/FuseO1-DeepSeekR1-Qwen2.5-Instruct-32B-Preview).
+<p style="font-size:1em; color:#6b7280; max-width:680px; margin:16px auto;">
+An SCE merge of <b>Behemoth-X</b> and <b>Behemoth-R1</b> — 123B parameters where prose voice meets reasoning mind in a single model. No retraining. No LoRA. Just principled weight arithmetic.
+</p>
 
-**The parents:**
-- **[Behemoth-X-123B-v2](https://huggingface.co/TheDrummer/Behemoth-X-123B-v2)** — the top-rated creative writer on the [UGI Leaderboard](https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard). Character voice, prose density, the reason people run 123B at home.
-- **[Behemoth-R1-123B-v2](https://huggingface.co/TheDrummer/Behemoth-R1-123B-v2)** — Behemoth's reasoning sibling. Knows when to open `<think>`, knows when to close it.
+<p>
+<img src="https://img.shields.io/badge/base-Mistral_Large_2411-FF6B35?style=for-the-badge&logo=mistralai&logoColor=white" alt="base"/>
+<img src="https://img.shields.io/badge/merge-SCE-8B5CF6?style=for-the-badge" alt="method"/>
+<img src="https://img.shields.io/badge/params-123B-EC4899?style=for-the-badge" alt="size"/>
+<img src="https://img.shields.io/badge/context-131k-06B6D4?style=for-the-badge" alt="context"/>
+</p>
 
-**The child:** both, at once.
+</div>
 
----
+<img src="https://huggingface.co/tacodevs/Behemoth-X-R1-123B/resolve/main/assets/divider_main.png" alt="" style="width:100%; margin:32px 0; border-radius:12px;"/>
+
+## ⚡ Two souls, one beast
+
+<table width="100%" style="border:none;">
+<tr>
+<td width="50%" align="center" style="padding:16px; vertical-align:top;">
+<img src="https://huggingface.co/tacodevs/Behemoth-X-R1-123B/resolve/main/assets/mind.png" alt="The Mind" style="width:100%; max-width:360px; border-radius:16px; box-shadow:0 0 40px rgba(139,92,246,0.4);"/>
+<h3 style="color:#a855f7; margin-top:12px;">🧠 The Mind</h3>
+<p style="font-size:0.95em; color:#9ca3af;">From <a href="https://huggingface.co/TheDrummer/Behemoth-R1-123B-v2"><b>Behemoth-R1-123B-v2</b></a> — the reasoning sibling that knows when to open <code>&lt;think&gt;</code> and when to close it. Character-aware analytical reasoning baked into the weights.</p>
+</td>
+<td width="50%" align="center" style="padding:16px; vertical-align:top;">
+<img src="https://huggingface.co/tacodevs/Behemoth-X-R1-123B/resolve/main/assets/voice.png" alt="The Voice" style="width:100%; max-width:360px; border-radius:16px; box-shadow:0 0 40px rgba(236,72,153,0.4);"/>
+<h3 style="color:#ec4899; margin-top:12px;">🎭 The Voice</h3>
+<p style="font-size:0.95em; color:#9ca3af;">From <a href="https://huggingface.co/TheDrummer/Behemoth-X-123B-v2"><b>Behemoth-X-123B-v2</b></a> — the top-rated creative writer on the <a href="https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard">UGI Leaderboard</a>. Distinctive prose, deep character work, the reason people run 123B at home.</p>
+</td>
+</tr>
+</table>
+
+<div align="center" style="margin:32px 0; padding:20px; background:linear-gradient(135deg, rgba(139,92,246,0.08), rgba(236,72,153,0.08)); border-radius:16px; border:1px solid rgba(139,92,246,0.2);">
+<p style="font-size:1.1em; color:#c084fc; margin:0;">Most "thinking" models sacrifice prose for reasoning. Most creative models can't reason their way out of a scene.</p>
+<p style="font-size:1.25em; font-weight:700; color:#f472b6; margin:12px 0 0 0;">Behemoth-X-R1 refuses to compromise.</p>
+</div>
+
+<img src="https://huggingface.co/tacodevs/Behemoth-X-R1-123B/resolve/main/assets/divider_config.png" alt="" style="width:100%; margin:32px 0; border-radius:12px;"/>
 
 ## 🧬 How it was made
 
-**Method:** [SCE — Select, Calculate, Erase](https://arxiv.org/abs/2408.07990)
+<p><b>Method:</b> <a href="https://arxiv.org/abs/2408.07990">SCE — Select, Calculate, Erase</a></p>
 
 Unlike TIES or DARE, SCE doesn't prune deltas by density. It uses **variance-aware matrix-level selection with sign consensus** — meaning capability-bearing weight updates survive the merge even when they're small and diffuse. That matters here because reasoning is a *behavioral* trait encoded across many tiny parameter shifts, not a knowledge trait concentrated in a few big ones.
 
-**The recipe:**
+This is the same recipe FuseAI used to preserve reasoning in [FuseO1-DeepSeekR1-Qwen2.5-Instruct-32B-Preview](https://huggingface.co/FuseAI/FuseO1-DeepSeekR1-Qwen2.5-Instruct-32B-Preview).
+
+### The recipe
 
 ```yaml
 models:
@@ -75,14 +95,18 @@ parameters:
 dtype: bfloat16
 ```
 
-**Why these numbers?**
+<details>
+<summary><b>Why these numbers?</b></summary>
 
 - **55/45** — Slight lean toward X for prose quality while giving R1 enough mass to keep its thinking circuit intact. Both parents share the same base, same tokenizer (verified identical SHA256), and the same training lineage — ideal merge conditions.
-- **`select_topk: 1.0`** — Keep all the deltas. Let variance + sign consensus do the work. This is the FuseO1 setting, validated empirically on reasoning merges.
+- **`select_topk: 1.0`** — Keep all deltas. Let variance + sign consensus do the work. This is the FuseO1 setting, validated empirically on reasoning merges.
+- **bfloat16** — Native precision of both parents, no conversion losses.
 
----
+</details>
+
+<img src="https://huggingface.co/tacodevs/Behemoth-X-R1-123B/resolve/main/assets/divider_config.png" alt="" style="width:100%; margin:32px 0; border-radius:12px;"/>
 
-## 📜 Prompt Format
+## 📜 Prompt format
 
 Standard **Mistral v7**, same as both parents:
 
@@ -90,7 +114,7 @@ Standard **Mistral v7**, same as both parents:
 [SYSTEM_PROMPT]{system}[/SYSTEM_PROMPT][INST]{user}[/INST]{assistant}</s>
 ```
 
-### To trigger thinking
+### 🎯 Trigger thinking
 
 Prefill the assistant turn with a `<think>` block. The model will continue your prefill, close the tag, and drop into the narrative:
 
@@ -99,42 +123,68 @@ Prefill the assistant turn with a `<think>` block. The model will continue your
 {seed phrase}
 ```
 
-**Prefill examples that work well:**
+### 🎨 Prefills that work
+
+<table width="100%" style="border:none;">
+<tr><td style="padding:8px; vertical-align:top; width:50%;">
+<img src="https://huggingface.co/tacodevs/Behemoth-X-R1-123B/resolve/main/assets/sticker_think.png" align="right" width="72" style="margin-left:12px;"/>
+<b style="color:#a855f7;">🔍 Analytical</b>
+<p><i>For character-driven scenes where you need the model to reason about motivations, values, and goals.</i></p>
 
 ```
 <think>
-Ok i need to think about how to respond — what does the character feel right now,
-what from their experience is relevant, what do they value, and what are they
+Ok i need to think about how to respond —
+what does the character feel right now,
+what from their experience is relevant,
+what do they value, and what are they
 trying to achieve, so
 ```
+</td><td style="padding:8px; vertical-align:top; width:50%;">
+<img src="https://huggingface.co/tacodevs/Behemoth-X-R1-123B/resolve/main/assets/sticker_write.png" align="right" width="72" style="margin-left:12px;"/>
+<b style="color:#ec4899;">✨ Creative</b>
+<p><i>For when you want the model to surprise you. Narrative twists, unexpected beats, story craft.</i></p>
 
 ```
 <think>
-Ok i need to think as a creative writer — what twist would surprise here?
-Let me find an engaging new direction nobody saw coming, so
+Ok i need to think as a creative writer —
+what twist would surprise here? Let me
+find an engaging new direction nobody
+saw coming, so
 ```
+</td></tr>
+<tr><td colspan="2" style="padding:8px;">
+<b style="color:#f43f5e;">🔥 Unhinged</b>
+<p><i>No restraint, no pulling punches. The darkest fiction, visceral prose, raw character truth.</i></p>
 
 ```
 <think>
-Ok i need to think as an unhinged author — raw, explicit, intense, fully in
-character with no holding back, so
+Ok i need to think as an unhinged author — raw, explicit, intense,
+fully in character with no holding back, so
 ```
-
-The model inherits R1's thinking circuit but shares R1's preference for being prefilled rather than self-triggering. Seed the tag, let it cook.
+</td></tr>
+</table>
 
 ### Without thinking
 
-Skip the prefill. It behaves close to pure Behemoth-X.
+Skip the prefill. It behaves close to pure Behemoth-X — standard RP, creative writing, whatever you'd use X for.
 
----
+<img src="https://huggingface.co/tacodevs/Behemoth-X-R1-123B/resolve/main/assets/divider_config.png" alt="" style="width:100%; margin:32px 0; border-radius:12px;"/>
 
-## 🎚️ Recommended Samplers
+## 🎚️ Samplers
 
 Start with **Behemoth-X's** recommended settings — the merge leans heavily on X's prose tuning.
 
 For thinking mode, drop temperature to **0.6–0.8**. The `<think>` block benefits from more deterministic reasoning; high temperature scrambles the structure.
 
----
+<table width="100%">
+<tr>
+<th>Setting</th><th>No thinking</th><th>With thinking</th>
+</tr>
+<tr><td><b>Temperature</b></td><td>1.0 – 1.25</td><td>0.6 – 0.8</td></tr>
+<tr><td><b>Min-P</b></td><td>0.05</td><td>0.05</td></tr>
+<tr><td><b>DRY</b></td><td>0.8 / 1.75 / 4</td><td>0.8 / 1.75 / 4</td></tr>
+<tr><td><b>Smooth Sampling</b></td><td>Off</td><td>Off</td></tr>
+</table>
 
 ## 🚀 Usage
 
@@ -149,13 +199,13 @@ python -m vllm.entrypoints.openai.api_server \
   --trust-remote-code
 ```
 
-### Single-GPU inference
+### Single-GPU
 
-Grab one of the quantized variants:
-- **FP8** — ~123 GB, fits on 1x H200, near-lossless quality
-- **AWQ / GPTQ W4A16** — ~65 GB, fits on 1x H100, slight quality tradeoff
+Grab one of the quantized variants (coming soon):
+- **FP8** — ~123 GB, fits on 1× H200, near-lossless
+- **AWQ / GPTQ W4A16** — ~65 GB, fits on 1× H100, small quality tradeoff
 
----
+<img src="https://huggingface.co/tacodevs/Behemoth-X-R1-123B/resolve/main/assets/divider_main.png" alt="" style="width:100%; margin:32px 0; border-radius:12px;"/>
 
 ## 🧱 Lineage
 
@@ -166,26 +216,27 @@ Mistral-Large-Instruct-2411 (Mistral AI)
        └─ Behemoth-X-R1-123B              ← the merge
 ```
 
----
-
 ## 🔍 Known behaviors
 
 - **`<think>` triggers on prefill, not spontaneously.** Inherited from R1. Seed the tag.
-- **Thinking style is R1-derived** — structured, character-aware, useful but not floaty. If you want Opus-style literary pre-writing, that's a follow-up fine-tune target, not something this merge gives you for free.
+- **Thinking style is R1-derived** — structured, character-aware, analytical. Not Opus-style floaty literary prose. If you want that, it's a follow-up fine-tune target.
 - **Prose voice is mostly X.** Most generations are indistinguishable from pure X on writing quality.
-- **Long character cards work natively.** No fine-tuning means no overfitting on context length.
-
----
+- **Long character cards work natively.** No fine-tuning means no overfitting on context length. 4k+ token system prompts handled without degradation.
+- **NSFW-capable.** Both parents are unrestricted; the merge preserves that.
 
 ## 🙏 Credits
 
-- **[TheDrummer](https://huggingface.co/TheDrummer)** — for Behemoth-X and Behemoth-R1, the two best Mistral Large fine-tunes in the creative space.
-- **[Mistral AI](https://huggingface.co/mistralai)** — for the foundation both parents are built on.
-- **[Arcee AI](https://github.com/arcee-ai/mergekit)** — for mergekit and the SCE implementation.
-- **[FuseAI](https://huggingface.co/FuseAI)** — for proving SCE preserves reasoning.
-
----
+<table width="100%">
+<tr><td width="33%" align="center"><b><a href="https://huggingface.co/TheDrummer">TheDrummer</a></b><br/><sub>For Behemoth-X and Behemoth-R1, the two best Mistral Large fine-tunes in the creative space.</sub></td>
+<td width="33%" align="center"><b><a href="https://huggingface.co/mistralai">Mistral AI</a></b><br/><sub>For the foundation both parents are built on.</sub></td>
+<td width="33%" align="center"><b><a href="https://github.com/arcee-ai/mergekit">Arcee AI</a></b><br/><sub>For mergekit and the SCE implementation.</sub></td></tr>
+<tr><td colspan="3" align="center" style="padding-top:12px;"><b><a href="https://huggingface.co/FuseAI">FuseAI</a></b> — for proving SCE preserves reasoning.</td></tr>
+</table>
 
 ## 📄 License
 
 Inherited from base: **[Mistral Research License](https://mistral.ai/licenses/MRL-0.1.md)** — non-commercial use only.
+
+<div align="center" style="margin-top:40px; padding:20px; background:linear-gradient(135deg, rgba(139,92,246,0.08), rgba(236,72,153,0.08)); border-radius:16px; border:1px solid rgba(139,92,246,0.2);">
+<p style="font-size:1em; color:#c084fc; margin:0;">Merged with 💜 by <a href="https://huggingface.co/tacodevs">tacodevs</a></p>
+</div>