Spark X2.6 4B EAGLE3 Chinese retrain needs about 1.0 GB of memory at its main file (a 0.4 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 4 quants, from 0.3 GB to 0.5 GB.
Spark X2.6 4B EAGLE3 Chinese retrain – a 4B Chinese/English NLP draft head
Spark X2.6 4B EAGLE3 Chinese retrain is a 0.259 billion‑parameter draft head built on the Spark family. It supports Chinese↔English translation, summarisation, document organisation and chat. The model is released under the apache‑2.0 license.
Good at
- Chinese↔English translation
- NLP summarisation
- Document organisation
Limits
- Requires a one‑line llama.cpp patch
- No vision support
- Context limited to 4096 tokens
Best for
- Chat‑based Chinese/English NLP tasks
- Translation workflows
- Summarisation pipelines
Who it’s for. Developers building bilingual NLP applications who need a lightweight draft head to accelerate inference on a 4B base model. The model is suitable for CPU‑only setups with a GPU head or for GPU‑only inference with a full 4B target.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 0.3 GB | 0.8 GB | Fits | Get file → | |
| — | 0.4 GB | 1.0 GB | Fits | Get file → | |
| — | 0.5 GB | 1.0 GB | Fits | Get file → | |
| — | 0.5 GB | 1.0 GB | Fits | Get file → |
All available GGUF files are quantised as "unknown". The smallest file is 0.28 GB and estimates 0.8 GB of VRAM at 8K context. The 0.44 GB file uses about 1 GB of VRAM. For 8 GB VRAM, any of the files will fit; for 4 GB VRAM, use the 0.28 GB file. Memory estimates are approximate.
What GPU do you need for Spark X2.6 4B EAGLE3 Chinese retrain?
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 the main file (0.5 GB file, about 1.1 GB in use).
Which Spark X2.6 4B EAGLE3 Chinese retrain quant should you choose?
If memory is tight, the smallest file is the first one in the table at 0.3 GB, with the quality trade-offs that come with fewer bits.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Spark X2.6 4B EAGLE3 Chinese retrain
How much VRAM does Spark X2.6 4B EAGLE3 Chinese retrain need?
About 1.0 GB at its main file with an 8K context window. That is the 0.4 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 Spark X2.6 4B EAGLE3 Chinese retrain on an 8 GB GPU?
Yes, with the smaller files: the 0.5 GB file needs about 1.1 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Spark X2.6 4B EAGLE3 Chinese retrain GGUF?
It depends on the quant: file is 0.3 GB, file is 0.4 GB, file is 0.5 GB, file is 0.5 GB.
What context length does Spark X2.6 4B EAGLE3 Chinese retrain support?
The GGUF header lists a trained context length of 4,096 tokens (4K). The KV cache grows with the context you use: about 0.0 GB at 8K and 0.0 GB at 4K.
Where can I download Spark X2.6 4B EAGLE3 Chinese retrain GGUF?
The files are on Hugging Face in the TiaoFeng/Spark-X2.6-4B-EAGLE3-Chinese-retrain repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does Spark X2.6 4B EAGLE3 Chinese retrain 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 the 0.28 GB file on 4 GB VRAM?
Yes, the 0.28 GB file estimates 0.8 GB of VRAM at 8K context, well within a 4 GB limit.
Do I need a special patch to use this model?
Yes, a one‑line llama.cpp patch is required for the Spark‑2.5 architecture to enable draft head inputs.
Is this model suitable for vision tasks?
No, the model does not support vision and is focused on natural‑language processing.