Zen Alta Draft Merged needs about 1.2 GB of memory at Q4_K_M (a 0.6 GB file) with an 8K context window, so it fits on a 4 GB GPU. Longer context needs more: at 32K tokens the KV cache alone adds about 0.5 GB.
Zen Alta Draft Merged – 790M 4‑layer speculative draft for fast inference
Zen Alta Draft Merged is a 0.797B llama model with a 4‑layer draft architecture. It is designed for speculative decoding alongside the 24‑layer target model. The model is quantized to Q4_K_M for efficient use in llama.cpp and other runtimes.
Good at
- Speculative decoding acceleration
- Low‑memory inference
- Maintains persona quality
Limits
- No vision support
- Requires paired target model for full generation
- Context limited to 131072 tokens
Best for
- Mobile inference
- Fast preview of responses
- Low‑budget LLM deployment
Who it’s for. Developers building lightweight conversational agents who need quick token generation without sacrificing persona. Ideal for environments with limited VRAM or CPU resources.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K_M | 0.6 GB | 1.2 GB | Fits | Get file → |
The Q4_K_M quantization results in a 0.57 GB file and requires approximately 1.2 GB of VRAM at 8 k context. This is suitable for systems with 8 GB or more of VRAM. For tighter budgets, consider a lower‑precision quant if available.
What GPU do you need for Zen Alta Draft Merged?
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 Q4_K_M (0.6 GB file, about 1.2 GB in use).
Which Zen Alta Draft Merged quant should you choose?
Q4_K_M is the usual starting point for Zen Alta Draft Merged: it balances file size and quality well. Here it is a 0.6 GB file.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Zen Alta Draft Merged
How much VRAM does Zen Alta Draft Merged need?
About 1.2 GB at Q4_K_M with an 8K context window. That is the 0.6 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 Zen Alta Draft Merged on an 8 GB GPU?
Yes, with Q4_K_M: the 0.6 GB file needs about 1.2 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Zen Alta Draft Merged GGUF?
It depends on the quant: Q4_K_M is 0.6 GB.
What context length does Zen Alta Draft Merged support?
The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 0.1 GB at 8K and 0.5 GB at 32K.
Where can I download Zen Alta Draft Merged GGUF?
The files are on Hugging Face in the ZenithLLM/ZenAlta-Draft repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does Zen Alta Draft Merged use?
Hugging Face lists the license as llama3.2. 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 Zen Alta Draft on 8 GB VRAM?
Yes, the Q4_K_M quantization uses about 1.2 GB of VRAM at 8 k context, fitting comfortably on 8 GB systems.
Does this model support vision?
No, the model has no vision capabilities.
What is speculative decoding?
It uses a fast 4‑layer draft to propose tokens, then a 24‑layer target verifies them, enabling faster generation without quality loss.