Ingot 27B DFlash2 needs about 2.7 GB of memory at Q8_0 (a 2.1 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 2 quants, from 2.1 GB (Q8_0) to 3.9 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 0.7 GB.
Ingot 27B DFlash2 – a speculative‑decoding drafter for Qwen models
Ingot 27B DFlash2 is a DFlash 2 drafter fine‑tuned from z‑lab/Qwen3.8‑27B‑DFlash2. It proposes tokens for a target king model, speeding up generation without altering the final output. The model is 1.92 billion parameters and supports a 262144‑token context.
Good at
- Speculative decoding drafting
- Fast token proposal
- Compatible with Qwen kings
Limits
- No vision support
- Requires a target king model
- License is apache‑2.0
Best for
- Speeding up Qwen inference
- Low‑latency chat applications
- Batch token drafting
Who it’s for. Engineers building inference pipelines that need faster generation from Qwen models. The drafter works with a PQ2_0 king in the Dendrite llama.cpp fork.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q8_0 | 2.1 GB | 2.7 GB | Fits | Get file → | |
| BF16 | 3.9 GB | 4.5 GB | Fits | Get file → |
For 8 GB VRAM use Q8_0: file size 2.06 GB, memory 2.7 GB at 8k context. For 12 GB VRAM use BF16: file size 3.86 GB, memory 4.5 GB at 8k context. Estimates are approximate.
What GPU do you need for Ingot 27B DFlash2?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: the best quant that fits is Q8_0 (2.1 GB file, about 2.7 GB in use).
- 6 GB VRAM: the best quant that fits is BF16 (3.9 GB file, about 4.5 GB in use).
Which Ingot 27B DFlash2 quant should you choose?
Q8_0 (2.1 GB) is the middle option in this repository. Close to lossless. It costs roughly twice the memory of a 4-bit file, so it suits cases where you have plenty of VRAM.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Ingot 27B DFlash2
How much VRAM does Ingot 27B DFlash2 need?
About 2.7 GB at Q8_0 with an 8K context window. That is the 2.1 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run Ingot 27B DFlash2 on an 8 GB GPU?
Yes, with BF16: the 3.9 GB file needs about 4.5 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Ingot 27B DFlash2 GGUF?
It depends on the quant: Q8_0 is 2.1 GB, BF16 is 3.9 GB.
What context length does Ingot 27B DFlash2 support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 0.2 GB at 8K and 0.7 GB at 32K.
Where can I download Ingot 27B DFlash2 GGUF?
The files are on Hugging Face in the dendriteholdings/Ingot-27B-DFlash2 repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does Ingot 27B DFlash2 use?
Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run Ingot 27B DFlash2 on 8 GB VRAM?
Yes, the Q8_0 quant uses about 2.7 GB of VRAM at 8k context, leaving room for the king model.
Does this model support vision tasks?
No, Ingot 27B DFlash2 is a text‑only drafter and has no vision capabilities.