PH

Phi 4 Mini Instruct DSpark

PhiGGUF by rasyosefUpdated 2026-10-05
Download on Hugging Face
Parameters658M
Context128K
Layers4
Architecturedflash
Smallest quant0.7 GB
Quick answer

Phi 4 Mini Instruct DSpark needs about 2.0 GB of memory at BF16 (a 1.3 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 3 quants, from 0.7 GB (Q8_0) to 1.3 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 0.5 GB.

Phi 4 Mini Instruct DSpark – draft sidecar for speculative decoding

Phi 4 Mini Instruct DSpark is a draft sidecar built on 0.658B parameters and 131072 context tokens. It contains 4 Qwen3 layers and was trained on 116,000 samples. The model works with a paired Phi-4-mini-instruct GGUF target for exact speculative decoding.

Good at

  • Speculative decoding
  • Fast draft generation
  • Exact verification

Limits

  • Requires paired Phi-4-mini-instruct GGUF target
  • No vision support
  • MIT license

Best for

  • Speeding up inference on Phi-4-mini-instruct
  • Low‑latency chat
  • Batch generation with draft sidecar

Who it’s for. Developers building custom inference pipelines with llama.cpp or vLLM who need faster token generation. The model is designed for text‑only workloads and works best when paired with a Phi‑4‑mini‑instruct GGUF target.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q8_00.7 GB 1.3 GB
Fits Get file →
BF161.3 GB 2.0 GB
Fits Get file →
F161.3 GB 2.0 GB
Fits Get file →

For 8‑k context, the Q8_0 quant uses about 1.3 GB memory, BF16 and F16 use about 2 GB. If you have 8 GB VRAM, Q8_0 is recommended; for 16 GB or more, BF16 or F16 provide a balance of speed and quality.

What GPU do you need for Phi 4 Mini Instruct DSpark?

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 BF16 (1.3 GB file, about 2.0 GB in use).

Which Phi 4 Mini Instruct DSpark quant should you choose?

BF16 (1.3 GB) is the middle option in this repository. The unquantized model. It gives original quality and needs the most memory, so it is mainly for conversion or reference.

If memory is tight, the smallest file is Q8_0 at 0.7 GB, with the quality trade-offs that come with fewer bits. 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

Architecturedflash
Parameters658M
Layers4
Trained context131,072 tokens
KV cache (fp16)16 KB per token

Source

Quantized byrasyosef
FamilyPhi
Licensemit
Downloads (HF)274
Last updated2026-10-05

Frequently asked questions about Phi 4 Mini Instruct DSpark

How much VRAM does Phi 4 Mini Instruct DSpark need?

About 2.0 GB at BF16 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 Phi 4 Mini Instruct DSpark on an 8 GB GPU?

Yes, with BF16: the 1.3 GB file needs about 2.0 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of Phi 4 Mini Instruct DSpark GGUF?

It depends on the quant: Q8_0 is 0.7 GB, BF16 is 1.3 GB, F16 is 1.3 GB.

What context length does Phi 4 Mini Instruct DSpark 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 Phi 4 Mini Instruct DSpark GGUF?

The files are on Hugging Face in the rasyosef/Phi-4-mini-instruct-DSpark-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 Phi 4 Mini Instruct DSpark use?

Hugging Face lists the license as mit. 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, using Q8_0 quant requires about 1.3 GB memory at 8‑k context, well within 8 GB.

Does it support vision?

No, it only handles text.

How do I pair it with a target model?

Load the draft sidecar with a Phi‑4‑mini‑instruct GGUF target; the embeddings are shared at load time.

Similar models

Browse all →

More from rasyosef

All by rasyosef →

More Phi models

All Phi →

Other tiny models

All tiny models →