Store › model › LLM
Qwopus3.6-27B-Fusion-GGUF
by KyleHessling1 · source Hugging Face · updated 2026-08-04
other16 GB~19 GB RAMsource aliveunlabeled
▶ PLAY ORBITAL — a game this model built, live in your browser ◀
Add to LogiShell Open in the web IDE
The button opens LogiShell with this card; nothing installs from a link by itself. Inside the app the install goes through lsh models install hf:KyleHessling1/Qwopus3.6-27B-Fusion-GGUF and its progress lives in the Resource Center.
Source and license
- Source: https://huggingface.co/KyleHessling1/Qwopus3.6-27B-Fusion-GGUF
- License: other (custom license: read it at the source before installing) · text
- Requirements: about 19 GB of RAM, 16 GB on disk (estimate: suggested file size × 1.15 + 0.5 GB; a real measurement comes with lsh models). Runs with llama.cpp, ollama.
- Tags:
ggufmergetask-vectordare-tieslayer-weightedqwen3qwen35reasoningcodellama.cppexperimentaltext-generationenendpoints_compatibleconversational
Numbers
- 291,886 downloads on Hugging Face
- 78 likes
- license other
- 16 GB for Qwopus3.6-27B-Fusion-Q4_K_M.gguf
- 327,838 npm downloads a week for node-llama-cpp
- latest node-llama-cpp@3.22.1
Numbers as of 2026-10-02 21:01 UTC, from the source API.
Summary
Summary not ready yet: the numbers are here, the text is not. It is written by the collector through the LogiShell model facade when a provider key is present.
Reviews
No reviews yet. Reviews are written inside LogiShell: open this card in the app.
Files
| file | quant | size |
|---|---|---|
Qwopus3.6-27B-Fusion-BF16.gguf | BF16 | 51 GB |
Qwopus3.6-27B-Fusion-Q3_K_M.gguf | Q3_K_M | 13 GB |
Qwopus3.6-27B-Fusion-Q4_K_M.gguf | Q4_K_M | 16 GB |
Qwopus3.6-27B-Fusion-Q5_K_M.gguf | Q5_K_M | 18 GB |
Qwopus3.6-27B-Fusion-Q6_K.gguf | Q6_K | 21 GB |
Qwopus3.6-27B-Fusion-Q8_0.gguf | Q8_0 | 27 GB |
Qwopus3.6-27B-Fusion-mmproj-Q8_0.gguf | Q8_0 | 600 MB |
From the source README
▶ PLAY ORBITAL — a game this model built, live in your browser ◀
> Research preview. This model attempts to combine the reasoning capability of
> `Qwopus3.6-27B-v2` with the code-execution capability of `Qwopus3.6-27B-Coder` — in a single 27B
> model, *without significant loss to either* skill. In practice it behaves like a production model
> with both capabilities fused, but it has not yet been through a full rigorous evaluation. If you
> find issues, please reach out on X — @kylehessling1.
- Format: GGUF, `Q4_K_M` (~16.5 GB) — runs on a single 24–32 GB GPU via `llama.cpp`.
- Architecture: `qwen35` hybrid (linear-attention + periodic full-attention), 27B params, 64 layers.
- Native context: 262,144 tokens. Verified clean (needle + termination) to 60K.
- Modes: thinking (reasoning) on by default; strong agentic/coding behavior.
- Vision: 🖼️ image input supported via the companion projector `Qwopus3.6-27B-Fusion-mmproj-Q8_0.gguf` (~0.6 GB) — pair it with *any* quant below (see Vision).
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
◆ TL;DR
| | |
|---|---|
| Base / anchor | `Qwen/Qwen3.6-27B` (both parents descend from it) |
| Parent A — reasoning | `Qwopus3.6-27B-v2` — donates the merge base + embeddings/head/norms/MTP |
| Parent B — code | `Qwopus3.6-27B-Coder` — delta source, injected with depth-increasing weight |
| Merge math | `W(L) = V2 + α(L)·(Coder − V2)`, α linear ramp 0.12 (early) → 0.48 (late) |
| Frozen (copied from A) | `embed_tokens`, `lm_head`, all `norm`, `mtp`/NextN, any vision tensors |
Headline results (Q4_K_M, thinking-on unless noted):
| Capability | Score | Stability | Score |
|---|---|---|---|
| HumanEval | 94.5% | Termination @ temp 0.2 & 0.9 | 4/4 clean |
| MBPP | 87.9% | Temp sweep 0.5…
Source: https://huggingface.co/KyleHessling1/Qwopus3.6-27B-Fusion-GGUF
Card id model:hf:KyleHessling1/Qwopus3.6-27B-Fusion-GGUF · collected 2026-10-02 21:01 UTC · JSON