At a glance
Table with columns: Field, Value| Field | Value |
|---|
| Format | compressed-tensors nvfp4-pack-quantized |
| Quantization | W4A16, group size 16, FP8 E4M3 scales, weight only |
| Kept in BF16 | vision tower, linear attention path, lm_head, embeddings, MTP head |
| Calibration | none |
| Base | DavidAU/Qwen3.8-27B-TWIN-TURBO-Fable-Cold-Fusion-709-L-Uncensored (Qwen3.8-27B, Apache 2.0) |
Quantized set covers MLP on all 64 layers plus Q, K, V, O on the 16 full attention layers (256 NVFP4 tensors verified). The DeltaNet path stays BF16 in this checkpoint and is normalized to NVFP4 in the GGUF tier builds (448 tensor backbone).
Checkpoint
- Single
model.safetensors about 25 GB
config.json quantization_config.format=nvfp4-pack-quantized, quant_method=compressed-tensors, Qwen3_5ForCausalLM, 64 layers, hybrid GatedDeltaNet, 262144 context, MTP head
tokenizer.json intact, chat template intact
- NVFP4 tensors verified via metadata checks, plus a vLLM smoke check (tensor parallel 2, coherent generation) before the GGUF tier builds
Provenance
Derivative of DavidAU/Qwen3.8-27B-TWIN-TURBO-Fable-Cold-Fusion-709-L-Uncensored (Apache 2.0) which itself derives from Qwen/Qwen3.8-27B. The TWIN-TURBO tune targets reduced thinking tokens with matched output quality. See the GGUF model card for full attribution once published.
Follow along & support
I post updates on new conversions, benchmarks, and what I'm working on over on my site. Follow along there to keep up with new releases and the work in progress. If you'd like to support more of it, you will find the options there. I do this on consumer hardware and like seeing how far it goes. More is on the way.
🌐 esatapedico.com. Updates, work-in-progress, and ways to support.
License
apache-2.0
Card written with AI assistance.