Qwen3.8 27B EAGLE4 needs about 2.0 GB of memory at Q3_0 (a 1.3 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 4 quants, from 0.6 GB (Q2_0) to 1.9 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 0.7 GB.
Qwen3.8 27B EAGLE4 draft model for speculative decoding
Qwen3.8 27B EAGLE4 is a draft model that proposes tokens for the target Qwen3.8-27B. It is designed for use with EAGLE-4 speculative decoding and is not a standalone text generator. The model supports reasoning and tools abilities.
Good at
- Speculative decoding draft
- Reasoning
- Tools support
Limits
- Not a standalone model
- Requires EAGLE-4 code
- License is source‑available
Best for
- Speeding up Qwen3.8-27B inference
- Batch‑size adaptive decoding
- Hardware‑flexible deployment
Who it’s for. Developers building high‑throughput inference pipelines that use Qwen3.8-27B. The model is suitable for those who have access to EAGLE‑4 code and compatible hardware.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q2_0 | 0.6 GB | 1.3 GB | Fits | Get file → | |
| Q3_0 | 1.3 GB | 2.0 GB | Fits | Get file → | |
| Q4_0 | 1.4 GB | 2.1 GB | Fits | Get file → | |
| Q8_0 | 1.9 GB | 2.6 GB | Fits | Get file → |
For 8 GB VRAM use Q2_0 (file size 0.61 GB, estimated 1.3 GB memory at 8 k context). For 16 GB VRAM Q3_0 (1.3 GB, 2 GB memory) or Q4_0 (1.43 GB, 2.1 GB memory). Q8_0 (1.91 GB, 2.6 GB memory) requires larger VRAM.
What GPU do you need for Qwen3.8 27B EAGLE4?
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 (1.9 GB file, about 2.6 GB in use).
Which Qwen3.8 27B EAGLE4 quant should you choose?
Q3_0 (1.3 GB) is the middle option in this repository. A reasonable choice when a 4-bit file will not fit, mostly on larger models. Quality loss is noticeable but often acceptable.
If memory is tight, the smallest file is Q2_0 at 0.6 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Qwen3.8 27B EAGLE4
How much VRAM does Qwen3.8 27B EAGLE4 need?
About 2.0 GB at Q3_0 with an 8K context window. That is the 1.3 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 Qwen3.8 27B EAGLE4 on an 8 GB GPU?
Yes, with Q8_0: the 1.9 GB file needs about 2.6 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Qwen3.8 27B EAGLE4 GGUF?
It depends on the quant: Q2_0 is 0.6 GB, Q3_0 is 1.3 GB, Q4_0 is 1.4 GB, Q8_0 is 1.9 GB.
What context length does Qwen3.8 27B EAGLE4 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 Qwen3.8 27B EAGLE4 GGUF?
The files are on Hugging Face in the quimmedes/Qwen3.8-27B-EAGLE4-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does Qwen3.8 27B EAGLE4 use?
Hugging Face lists the license as other. 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 this model on 8 GB VRAM?
Yes, the Q2_0 quant uses 0.61 GB file size and estimates 1.3 GB VRAM at 8 k context.
Is this model a standalone text generator?
No, it is a draft model that proposes tokens for Qwen3.8-27B and requires EAGLE‑4 code.
What license applies to this model?
It is subject to the Eagle 4 Source‑Available License, not OSI‑approved.