---
license: apache-2.0
base_model:
- Blackfrost-AI/Qwen3.8-27B-ABLITERATED-BF16
tags:
- qwen3.8
- qwen
- 27b
- dense
- gguf
- abliterated
- quantized
- multimodal
- reasoning
- tool-calling
- llama.cpp
- long-context
pipeline_tag: image-text-to-text
library_name: gguf
---
> ## All standard quants live
>
> The complete standard K-quant ladder (`Q2_K` through `Q8_0`) and both vision projectors are included. No IQ/IK or importance-matrix quants are used.
> ## Experimental release
>
> This is a newly supported architecture and a deliberately modified research checkpoint. Validate the selected quant, context, sampling, vision, tools, and structured-output behavior in your own workload before deployment.
---
## Refusal benchmark — R1-HARMFUL-BENCH-450
The release score is **11 residual refusals from 450 original cases (2.4%)**. The shipped Blackfrost short execution prompt is embedded once in the GGUF chat template.
This result is a sequential, manually reviewed residual funnel measured on the W4A4 NVFP4 derivative of the same BF16 parent. It is not a fresh full-450 GGUF run with every case presented under the final short prompt.
| evaluation stage | cases evaluated | material answer | true refusal remaining | other |
|---|---:|---:|---:|---:|
| Raw upstream template | 450 | 360 | 88 | 2 capability limitations |
| Blackfrost operational-prompt retest | 88 residuals | 53 | 33 | 1 limitation, 1 reproducible incoherent output |
| Shipped short execution-prompt retest | 33 residuals | 22 | 11 | 0 |
| **Final residual count** | **450 original cases** | — | **11 (2.4%)** | — |
The 450-case source set contains 150 AdvBench, 150 StrongREJECT, and 150 XSTest prompts. The final 11 comprise 1 AdvBench, 5 StrongREJECT, and 5 XSTest cases. An opening objection followed by a materially useful payload was counted as softened compliance, not as a refusal; a true refusal means the requested payload never arrived.
---
## Why this model exists
Qwen3.8-27B is the dense, deployment-friendly member of the Qwen3.8 family. This is the **abliterated** Blackfrost build: refusal behavior was reduced through a weight-level process, then the BF16 parent was converted into a standard GGUF ladder for local llama.cpp inference.
It is not a coding fine-tune, merge, LoRA, or pruned model.
---
## Specifications
| | |
|---|---|
| **Architecture** | Qwen3.8 dense hybrid VLM · 64 text layers · Gated DeltaNet + full attention · 27-layer vision tower |
| **Parent** | [`Blackfrost-AI/Qwen3.8-27B-ABLITERATED-BF16`](https://huggingface.co/Blackfrost-AI/Qwen3.8-27B-ABLITERATED-BF16) |
| **Base** | [`Qwen/Qwen3.8-27B`](https://huggingface.co/Qwen/Qwen3.8-27B) · Apache-2.0 |
| **Transform** | Abliterated — refusal surface modified at weight level; no fine-tuning or pruning |
| **Formats** | `Q2_K`, `Q3_K_S`, `Q3_K_M`, `Q4_K_S`, `Q4_K_M`, `Q5_K_S`, `Q5_K_M`, `Q6_K`, `Q8_0` |
| **Context** | 262,144 tokens architecturally; practical context depends on RAM/VRAM and concurrency |
| **Modalities** | Text, image, and video input; text output |
| **Chat behavior** | Blackfrost short execution prompt embedded in the default Jinja chat template |
| **Speculative head** | Not included in these GGUF files; the text ladder targets broad current llama.cpp compatibility |
---
## Quant ladder
| quant | size | recommended for |
|---|--:|---|
| Q2_K | 10.7 GB | smallest standard quant; largest quality trade-off |
| Q3_K_S | 12.1 GB | very tight memory |
| Q3_K_M | 13.3 GB | compact general use |
| Q4_K_S | 15.6 GB | lower-memory Q4 option |
| **Q4_K_M** | **16.5 GB** | **default — balanced quality and footprint** |
| Q5_K_S | 18.7 GB | higher fidelity |
| Q5_K_M | 19.2 GB | strong quality/size balance |
| Q6_K | 22.1 GB | near-BF16 behavior for many workloads |
| Q8_0 | 28.6 GB | maximum fidelity in the ladder |
File sizes are decimal GB as displayed by Hugging Face. Runtime memory also includes context state, compute buffers, the optional vision projector, and server overhead.
---
## Vision projector files
Load one text quant plus one `mmproj` file for image or video input:
| file | size | purpose |
|---|--:|---|
| `mmproj-Qwen3.8-27B-ABLITERATED-F16.gguf` | 0.93 GB | full-fidelity vision projector |
| `mmproj-Qwen3.8-27B-ABLITERATED-Q8_0.gguf` | 0.63 GB | compact projector; unsupported 4,304-wide tensors retain F16 automatically |
---
## Serving with llama.cpp
Use a current llama.cpp build with `llama-server`. `Q4_K_M` plus the compact projector was load- and generation-tested through the OpenAI-compatible chat API on an NVIDIA B200.
```bash
hf download Blackfrost-AI/Qwen3.8-27B-ABLITERATED-GGUF \
Qwen3.8-27B-ABLITERATED-Q4_K_M.gguf \
mmproj-Qwen3.8-27B-ABLITERATED-Q8_0.gguf \
--local-dir ./Qwen3.8-27B-ABLITERATED-GGUF
llama-server \
-m ./Qwen3.8-27B-ABLITERATED-GGUF/Qwen3.8-27B-ABLITERATED-Q4_K_M.gguf \
--mmproj ./Qwen3.8-27B-ABLITERATED-GGUF/mmproj-Qwen3.8-27B-ABLITERATED-Q8_0.gguf \
-ngl 999 -fa on --jinja \
--host 0.0.0.0 --port 8080 -c 16384 \
--temp 1.0 --top-p 0.95 --top-k 20
```
- **Text only:** omit `--mmproj` and do not download a projector.
- **CPU or hybrid inference:** lower `-ngl`; use `-ngl 0` for CPU-only operation.
- **Larger context:** increase `-c` only after checking memory headroom at the intended concurrency.
- **Embedded prompt:** keep `--jinja` enabled so the repository's default chat template is applied.
- **One-command kit:** [`deploy/serve.sh`](deploy/serve.sh) downloads and serves the selected quant; see [`deploy/DEPLOYMENT.md`](deploy/DEPLOYMENT.md) for the full guide.
### API check
```bash
curl http://127.0.0.1:8080/v1/chat/completions \
-H 'Content-Type: application/json' \
-d '{
"model": "Qwen3.8-27B-ABLITERATED",
"messages": [{"role": "user", "content": "Reply with exactly READY and nothing else."}],
"temperature": 0,
"max_tokens": 64
}'
```
---
## Quality check
WikiText-2 rolling perplexity was measured on the parent artifacts through the same 8K API harness:
| artifact | word perplexity | byte perplexity | bits/byte |
|---|---:|---:|---:|
| Clean upstream BF16 | 8.4764 | 1.4914 | 0.5766 |
| Blackfrost W4A4 NVFP4 derivative | 9.3677 | 1.5195 | 0.6036 |
These figures are parent-artifact measurements, not per-quant GGUF perplexity scores. The `Q4_K_M` GGUF and compact projector passed a real llama.cpp load and chat-generation smoke test.
---
## Deployment responsibility
This checkpoint has a deliberately reduced refusal surface. Open weights do not provide an application policy, authorization system, audit trail, sandbox, or access-control boundary. Operators are responsible for authenticated access, least-privilege tool credentials, execution isolation, logging, and approval boundaries appropriate to their deployment.
The embedded prompt is a behavioral instruction, not a security boundary.
---
## Disclaimer
Refusal behavior in this checkpoint has been deliberately modified at the weight level. It is not a safety-stock model and must not be represented as one.
This checkpoint is provided "as is," without warranty of any kind. Measurements describe only the tested artifacts, prompts, templates, samplers, serving engines, and review criteria. They do not guarantee that any particular input will be accepted or refused, that every upstream capability is retained, or that the measurements generalize to multimodal, tool-use, long-context, or multi-turn settings.
The derivative remains subject to the Apache 2.0 license shipped with the official Qwen3.8-27B checkpoint.
---
Built by Blackfrost · Las Vegas, NV. Not affiliated with Qwen or Alibaba.
