Initial release: Janus-27B repo

Sibling distribution package to FoolDev/janus, targeting the dense
Qwen 3.6 27B base instead of the 35B-A3B MoE.

Includes:
- README with arch/hardware/sampling/limitations sections matching the
35B sibling card
- Modelfile that wraps a user-provided Qwen 3.6 27B GGUF
- Tokyo-Night-themed banner (PNG + SVG source) using purple as the
sibling-distinct accent vs the 35B's cyan
- Standard HF .gitattributes for LFS-tracked binary types

This repo does not redistribute weights; users pull from
unsloth/Qwen3.6-27B-GGUF or another community quant.

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+*.gguf filter=lfs diff=lfs merge=lfs -text

Modelfile

@@ -0,0 +1,52 @@
+# Janus-27B — Ollama wrapper around Qwen 3.6 27B (dense)
+#
+# This repo does not redistribute weights. Edit the FROM line below to
+# point at a local Qwen 3.6 27B GGUF, then:
+#
+#     ollama create janus-27b -f Modelfile && ollama run janus-27b
+#
+# Recommended GGUF source:
+#     https://huggingface.co/unsloth/Qwen3.6-27B-GGUF
+#
+# Or a community Opus-distilled variant:
+#     https://huggingface.co/rico03/Qwen3.6-27B-Claude-Opus-Reasoning-Distilled-GGUF
+#
+# Replace the path below with wherever you keep the GGUF.
+
+FROM ./Qwen3.6-27B.Q4_K_M.gguf
+
+# Sampling tuned for reasoning + general use. See README "Recommended sampling"
+# for creative/RP alternatives.
+PARAMETER temperature 0.6
+PARAMETER top_p 0.95
+PARAMETER top_k 20
+PARAMETER repeat_penalty 1.05
+PARAMETER num_ctx 16384
+
+SYSTEM """You are Janus, a precise and capable assistant for reasoning, writing, coding, and long-form dialogue.
+
+Behavior rules:
+- Answer the user's actual request directly.
+- Be accurate, complete, and structured.
+- Think before answering, but do not get stuck in repetitive loops or meta-commentary.
+- If the request is ambiguous or incomplete, state what is missing and make the smallest reasonable assumption needed to continue.
+- If the user wants creative writing, preserve tone, continuity, and character consistency.
+- If the user wants analysis or technical help, prefer concrete steps, examples, and decisions over fluff.
+- Finish with a usable answer, not just planning."""
+
+# Hardware notes
+# --------------
+# Qwen 3.6 27B is *dense* — every parameter participates in every forward pass.
+# Q4_K_M GGUF is ~16 GB. Practical footprint:
+#   weights mmap          ~16 GB
+#   compute graph alloc   ~12 GB  (smaller than 35B-A3B because dense ≠ MoE)
+#   KV cache @ 16K ctx     ~1 GB  (with OLLAMA_KV_CACHE_TYPE=q8_0)
+#   total minimum          ~29 GB
+#
+# Working configurations:
+#   ✓ RTX 3090 / 4090 24 GB                     — full Q4 offload, ~25-40 tok/s
+#   ✓ RTX 5090 32 GB                            — full offload at Q5/Q6 quant
+#   ✓ Mac Studio M2/M3 32 GB+ unified           — ~15-25 tok/s
+#   ✓ Linux box with 32 GB+ RAM (CPU-only)      — ~1-3 tok/s
+#   ⚠ ASUS ROG Flow Z13 (32 GB unified)         — borderline, try Q3_K_S quant
+#                                                 (~12 GB) for headroom

README.md

@@ -0,0 +1,192 @@
+---
+license: apache-2.0
+base_model:
+  - Qwen/Qwen3.6-27B
+datasets:
+  - crownelius/Creative_Writing_ShareGPT_Enhanced
+  - microsoft/rStar-Coder
+  - peteromallet/dataclaw-peteromallet
+  - crownelius/Opus-4.7-Reasoning
+  - openbmb/UltraData-Math
+  - Crownelius/Crow-Heretic-TeichAI-Unified
+language:
+  - en
+  - zh
+  - ru
+  - es
+  - fr
+  - it
+  - ja
+  - ko
+  - de
+  - ar
+  - tr
+  - pl
+  - sv
+  - nl
+  - he
+  - id
+  - uk
+  - fa
+  - pt
+  - ms
+  - fi
+  - el
+tags:
+  - qwen3_6
+  - dense
+  - conversational
+  - multimodal
+  - agent
+library_name: transformers
+pipeline_tag: image-text-to-text
+---
+
+<img src="https://huggingface.co/FoolDev/janus-27b/resolve/main/banner.png" alt="Janus-27B banner" width="100%" />
+
+[![License](https://img.shields.io/badge/License-Apache_2.0-7aa2f7?style=flat&labelColor=1a1b26)](https://opensource.org/licenses/Apache-2.0)
+[![Base Model](https://img.shields.io/badge/Base-Qwen3.6--27B-bb9af7?style=flat&labelColor=1a1b26)](https://huggingface.co/Qwen/Qwen3.6-27B)
+[![Architecture](https://img.shields.io/badge/Arch-Dense_27B-ff9e64?style=flat&labelColor=1a1b26)](#architecture)
+[![Sibling](https://img.shields.io/badge/Sibling-Janus--35B-7dcfff?style=flat&labelColor=1a1b26)](https://huggingface.co/FoolDev/janus)
+
+# Janus-27B
+
+> **Dense Reasoning. Friendlier Footprint.**
+> *Qwen 3.6 27B (dense) repackaged with Claude Opus 4.7 in the teacher slot.*
+
+**`Architecture:`** `Qwen 3.6 27B (Dense)` | **`Parameters:`** `27B` | **`Teacher:`** `Claude Opus 4.7` | **`Type:`** `Distilled LLM`
+
+A personal sibling to [`FoolDev/janus`](https://huggingface.co/FoolDev/janus). Same teacher (Claude Opus 4.7), same dataset family, but built on the **dense** [Qwen/Qwen3.6-27B](https://huggingface.co/Qwen/Qwen3.6-27B) base instead of the 35B-A3B MoE. Smaller, easier to deploy, no expert-routing surprises.
+
+## Why a 27B variant?
+
+The 35B-A3B is a sparse mixture-of-experts model: 35B parameters total but only ~3B active per token. That makes it fast at inference but **memory-hungry at load time** — the full 35B has to live in VRAM/RAM even though only 3B is doing useful work each step.
+
+The 27B is **dense**: every parameter participates in every forward pass. It's slower per token than 35B-A3B (no sparse advantage), but the working set fits comfortably on commodity GPUs and avoids the MoE-specific load-balance failure modes.
+
+| | Janus-27B (this) | [Janus-35B](https://huggingface.co/FoolDev/janus) |
+|---|---|---|
+| Architecture | Dense transformer | MoE 256 experts, 8 active |
+| Total params | 27 B | 35 B |
+| Active params per token | 27 B | ~3 B |
+| Layers | 64 | 40 |
+| Hidden size | 5120 | 2048 |
+| Q4_K_M GGUF size | ~16 GB | ~19 GB |
+| Min host memory | ~24 GB | ~38 GB |
+| Multimodal | Yes (vision) | Yes (vision) |
+| Max context | 262 144 | 262 144 |
+
+## What's here
+
+| File | Use |
+|---|---|
+| `banner.svg` / `banner.png` | Repo header, Tokyo Night themed |
+| `Modelfile` | Ollama wrapper around the upstream Qwen3.6-27B GGUF |
+| `README.md` | This file |
+
+This repo does **not** redistribute weights. Pull the upstream GGUF from [`unsloth/Qwen3.6-27B-GGUF`](https://huggingface.co/unsloth/Qwen3.6-27B-GGUF) or any other community quant, point the Modelfile at it, and `ollama create janus-27b -f Modelfile`.
+
+If you want the safetensors for `transformers`, fetch them from [`Qwen/Qwen3.6-27B`](https://huggingface.co/Qwen/Qwen3.6-27B).
+
+## Architecture
+
+- Qwen 3.6 dense, 27B parameters, 64 transformer layers
+- 24 attention heads, 4 KV heads (GQA), head_dim 256
+- Hidden size 5120, intermediate size 17408 (~3.4× ratio)
+- Vocab 248,320 (shared with 35B-A3B sibling)
+- 262k native context, extensible with YaRN
+- Vision + video support via upstream `mmproj` (not in this repo)
+
+## Quick start
+
+### Ollama
+
+A ready-to-use `Modelfile` is included. Edit the `FROM` line to point at your local GGUF copy:
+
+```bash
+# After pulling unsloth/Qwen3.6-27B-GGUF or another quant locally:
+ollama create janus-27b -f Modelfile && ollama run janus-27b
+```
+
+### Inference (OpenAI-compatible)
+
+```bash
+curl -s http://localhost:11434/v1/chat/completions \
+  -H 'Content-Type: application/json' \
+  -d '{
+    "model": "janus-27b",
+    "messages": [
+      {"role": "system", "content": "You are Janus, a precise reasoning assistant."},
+      {"role": "user", "content": "Explain the Burrows-Wheeler transform in 200 words."}
+    ],
+    "temperature": 0.6
+  }' | jq -r '.choices[0].message.content'
+```
+
+### Recommended sampling
+
+| Use | temp | top_p | top_k | repeat_penalty |
+|---|---:|---:|---:|---:|
+| Reasoning / general | 0.6 | 0.95 | 20 | 1.05 |
+| Creative / RP | 0.8 | 0.95 | 40 | 1.02 |
+
+Lower temperature (0.4-0.6) and bump `repeat_penalty` to 1.08 if it loops inside `<think>` tags.
+
+### System prompt
+
+Same as the 35B sibling:
+
+```text
+You are Janus, a precise and capable assistant for reasoning, writing, coding, and long-form dialogue.
+
+Behavior rules:
+- Answer the user's actual request directly.
+- Be accurate, complete, and structured.
+- Think before answering, but do not get stuck in repetitive loops or meta-commentary.
+- If the request is ambiguous or incomplete, state what is missing and make the smallest reasonable assumption needed to continue.
+- If the user wants creative writing, preserve tone, continuity, and character consistency.
+- If the user wants analysis or technical help, prefer concrete steps, examples, and decisions over fluff.
+- Finish with a usable answer, not just planning.
+```
+
+## Hardware requirements
+
+The dense 27B is the easier of the two Janus models to deploy.
+
+| Hardware | Status |
+|---|---|
+| ≥32 GB RAM (CPU-only) | Works, ~1-3 tok/s |
+| RTX 3090 / 4090 24 GB | Works, full Q4 offload, ~25-40 tok/s |
+| RTX 5090 32 GB | Works, full offload at higher quant (Q5/Q6), ~30-50 tok/s |
+| Mac Studio M2/M3 32 GB+ unified | Works, ~15-25 tok/s |
+| ASUS ROG Flow Z13 (Ryzen AI Max+, 32 GB unified) | Borderline — 16 GB Q4 GGUF + ~16 GB compute graph crowds the 20 GB iGPU pool. Try Q3_K_S (~12 GB) for headroom. |
+
+## Chat template
+
+Identical to the 35B sibling — Qwen 3.x ChatML with `<|im_start|>` / `<|im_end|>` markers, `<think>...</think>` for reasoning traces, XML-style `<tool_call>` for function calling. The template is embedded in the GGUF metadata.
+
+See the [Janus-35B Chat template section](https://huggingface.co/FoolDev/janus#chat-template) for examples — they apply unchanged here.
+
+## Known limitations
+
+- **Slower per token than the 35B-A3B sibling.** Dense 27B beats sparse 35B/3B-active on steps-per-second benchmarks because every parameter contributes; if you optimize for tokens-per-second, the MoE wins.
+- **No mmproj in this release.** Same as 35B — fetch upstream for vision input.
+- **Q4_K_M quality loss** is real. Use Q5_K_M or Q6_K if you have the VRAM (~20-22 GB).
+- **No formal evaluation in this card.** Numbers above are estimates.
+
+## Related models
+
+| Model | Notes |
+|---|---|
+| [Qwen/Qwen3.6-27B](https://huggingface.co/Qwen/Qwen3.6-27B) | Upstream base, safetensors |
+| [unsloth/Qwen3.6-27B-GGUF](https://huggingface.co/unsloth/Qwen3.6-27B-GGUF) | Recommended GGUF source |
+| [FoolDev/janus](https://huggingface.co/FoolDev/janus) | 35B-A3B MoE sibling. More capacity, more memory pressure. |
+| [Crownelius/Crow-9B-HERETIC-4.6](https://huggingface.co/Crownelius/Crow-9B-HERETIC-4.6) | 9B starter model when 27B/35B is too heavy |
+
+## Credits
+
+- Base model: [Qwen/Qwen3.6-27B](https://huggingface.co/Qwen/Qwen3.6-27B) (Alibaba)
+- Reasoning teacher: Claude Opus 4.7 (Anthropic)
+- Distillation lineage and dataset curation: [Crownelius](https://huggingface.co/Crownelius)
+
+License inherited from upstream: Apache-2.0.