Downloads · 30 days
400
11% of all-time downloads
WhiskyAKM/Gemma-4-E4B-it-qat-NVFP4-GGUF
Gemma-4-E4B-it-qat-NVFP4-GGUF is a any-to-any model from WhiskyAKM. Use it for the any-to-any task on the model card, and read the license before you ship it in a product. It is set up for llama-cpp. The card lists the license as apache-2.0.
NVFP4 (NVIDIA Floating Point 4-bit) GGUF quantization of google/gemma-4-E4B-it-qat-q40-unquantized, created from the Quantization-Aware Training (QAT) checkpoint of Gemma 4 E4B IT.
Downloads · 30 days
400
11% of all-time downloads
All-time downloads
3.6K
Public
Repo size
10.7 GB
Likes
2
Public
Click a slice to open those files.
.gguf10.7 GB · 100%
From the Hugging Face model README
NVFP4 (NVIDIA Floating Point 4-bit) GGUF quantization of google/gemma-4-E4B-it-qat-q4_0-unquantized, created from the Quantization-Aware Training (QAT) checkpoint of Gemma 4 E4B IT.
Gemma 4 E4B IT is a multimodal model built by Google DeepMind that handles text, image, and audio inputs and generates text output. It is designed for efficient on-device deployment on laptops and mobile devices.
The "E" in E4B stands for "effective" parameters — the model uses Per-Layer Embeddings (PLE) to maximize parameter efficiency, giving each decoder layer its own small embedding for every token. This keeps the effective parameter count much smaller than the total.
This repository contains an NVFP4 GGUF conversion of the QAT-optimized checkpoint, making it usable with llama.cpp and other GGUF-compatible inference engines that support the NVFP4 data type.
| Property | Value |
|---|---|
| Architecture | Gemma4ForConditionalGeneration |
| Effective Parameters | 4.3B (7.5B with embeddings) |
| Layers | 42 |
| Embedding Length | 2560 |
| Sliding Window | 512 tokens |
| Context Length | 128K tokens |
| Vocabulary Size | 262K |
| Supported Modalities | Text, Image, Audio |
| Vision Encoder | SigLIP-based, 16 layers, 768-dim, ~478M params |
| Audio Encoder | 12 layers, 1024-dim |
| Attention | Hybrid (sliding window + global, every 6th layer) |
| RoPE | Proportional RoPE (p-RoPE) on global layers |
| File | Format | Size | Description |
|---|---|---|---|
gemma-4-E4B-it-qat-nvfp4.gguf | NVFP4 | 4.9G | NVFP4 quantized model - embeddings are in Q6_K |
gemma-4-E4B-it-qat-fast-nvfp4.gguf | NVFP4 | 4.0G | NVFP4 quantized model - embeddings are in NVFP4 |
mtp-bf16.gguf | BF16 | 160M | MTP draft model |
mtp-nvfp4.gguf | NVFP4 | 80M | NVFP4 quantized MTP draft model - not recomendend, in this case MTP model is already small and quantization causes degradation of accuracy |
mmproj.gguf | f32 | 946M | Multimodal projector (vision + audio) |
A chat_template.jinja file is also provided for use with chat-based inference.
Note on NVFP4: NVFP4 uses 4-bit floating-point representation (1 sign + 3 mantissa + 2 exponent bits on an E2M1 basis with microscaling) designed for NVIDIA Blackwell and later GPUs. It offers significant memory savings while maintaining good quality thanks to the QAT optimization.
Note on
mmproj.gguf: This file contains the vision and audio encoders/projection layers required for multimodal inference. It is not needed for text-only use.
# Run text inference
./llama-cli \
-m gemma-4-E4B-it-qat-nvfp4.gguf \
-p "Explain quantum computing in simple terms." \
--temp 1.0 --top-k 64 --top-p 0.95
# Text-only
./llama-server \
-m gemma-4-E4B-it-qat-nvfp4.gguf \
--host 0.0.0.0 --port 8080
# With multimodal (image + audio) support
./llama-server \
-m gemma-4-E4B-it-qat-nvfp4.gguf \
--mmproj mmproj.gguf \
--host 0.0.0.0 --port 8080
# fit must be off to eanble MTP
# i find max 6, min 1 to be optimal settings
./llama-cli \
-m gemma-4-E4B-it-qat-nvfp4.gguf \
--fit off \
--spec-type draft-mtp -md mtp-bf16.gguf \
--spec-draft-n-max 6 --spec-draft-n-min 1
For image and audio inputs, pass the --mmproj mmproj.gguf flag to load the multimodal projector. Refer to your inference engine's documentation for passing image/audio data alongside text prompts.
Modality order tip: For best results, place image content before text and audio content after text in your prompt.
Recommended parameters from the model's generation_config.json:
| Parameter | Value |
|---|---|
| Temperature | 1.0 |
| Top-K | 64 |
| Top-P | 0.95 |
| BOS Token ID | 2 |
| EOS Token IDs | 1, 106, 50 |
| Pad Token ID | 0 |
Gemma 4 supports configurable thinking (reasoning) mode:
<|think|> token at the start of the system prompt.<|channel>thought
[Internal reasoning]
<channel|>
[Final answer]
<|think|> token. For the E4B variant, thinking is fully off when disabled (no empty thought block is generated).Many libraries like Transformers and llama.cpp handle the chat template complexities automatically.
system role for structured conversations@misc{gemmateam2026gemma4,
title={Gemma 4 Technical Report},
author={Gemma Team},
year={2026},
eprint={2607.02770},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2607.02770},
}