Instructions to use eaddario/gemma-4-26B-A4B-it-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use eaddario/gemma-4-26B-A4B-it-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="eaddario/gemma-4-26B-A4B-it-GGUF",
	filename="gemma-4-26B-A4B-it-F16.gguf",
)

llm.create_chat_completion(
	messages = "No input example has been defined for this model task."
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use eaddario/gemma-4-26B-A4B-it-GGUF with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf eaddario/gemma-4-26B-A4B-it-GGUF:F16
# Run inference directly in the terminal:
llama-cli -hf eaddario/gemma-4-26B-A4B-it-GGUF:F16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf eaddario/gemma-4-26B-A4B-it-GGUF:F16
# Run inference directly in the terminal:
llama-cli -hf eaddario/gemma-4-26B-A4B-it-GGUF:F16

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf eaddario/gemma-4-26B-A4B-it-GGUF:F16
# Run inference directly in the terminal:
./llama-cli -hf eaddario/gemma-4-26B-A4B-it-GGUF:F16

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf eaddario/gemma-4-26B-A4B-it-GGUF:F16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf eaddario/gemma-4-26B-A4B-it-GGUF:F16

Use Docker

docker model run hf.co/eaddario/gemma-4-26B-A4B-it-GGUF:F16

LM Studio
Jan
Ollama
How to use eaddario/gemma-4-26B-A4B-it-GGUF with Ollama:
```
ollama run hf.co/eaddario/gemma-4-26B-A4B-it-GGUF:F16
```

Unsloth Studio new

How to use eaddario/gemma-4-26B-A4B-it-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for eaddario/gemma-4-26B-A4B-it-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for eaddario/gemma-4-26B-A4B-it-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for eaddario/gemma-4-26B-A4B-it-GGUF to start chatting

Pi new

How to use eaddario/gemma-4-26B-A4B-it-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf eaddario/gemma-4-26B-A4B-it-GGUF:F16

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "eaddario/gemma-4-26B-A4B-it-GGUF:F16"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use eaddario/gemma-4-26B-A4B-it-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf eaddario/gemma-4-26B-A4B-it-GGUF:F16

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default eaddario/gemma-4-26B-A4B-it-GGUF:F16

Run Hermes

hermes

Docker Model Runner
How to use eaddario/gemma-4-26B-A4B-it-GGUF with Docker Model Runner:
```
docker model run hf.co/eaddario/gemma-4-26B-A4B-it-GGUF:F16
```

Lemonade

How to use eaddario/gemma-4-26B-A4B-it-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull eaddario/gemma-4-26B-A4B-it-GGUF:F16

Run and chat with the model

lemonade run user.gemma-4-26B-A4B-it-GGUF-F16

List all available models

lemonade list

Experimental global target bits‑per‑weight quantization of google/gemma-4-26B-A4B-it

Using non-standard (forked) LLaMA C++ release b8838 for quantization.

Original model: google/gemma-4-26B-A4B-it

From the original model creators:

Hugging Face | GitHub | Launch Blog | Documentation
License: Apache 2.0 | Authors: Google DeepMind

Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on small models) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants. Gemma 4 features a context window of up to 256K tokens and maintains multilingual support in over 140 languages.

Featuring both Dense and Mixture-of-Experts (MoE) architectures, Gemma 4 is well-suited for tasks like text generation, coding, and reasoning. The models are available in four distinct sizes: E2B, E4B, 26B A4B, and 31B. Their diverse sizes make them deployable in environments ranging from high-end phones to laptops and servers, democratizing access to state-of-the-art AI.

Gemma 4 introduces key capability and architectural advancements:

Reasoning – All models in the family are designed as highly capable reasoners, with configurable thinking modes.

Extended Multimodalities – Processes Text, Image with variable aspect ratio and resolution support (all models), Video, and Audio (featured natively on the E2B and E4B models).

Diverse & Efficient Architectures – Offers Dense and Mixture-of-Experts (MoE) variants of different sizes for scalable deployment.

Optimized for On-Device – Smaller models are specifically designed for efficient local execution on laptops and mobile devices.

Increased Context Window – The small models feature a 128K context window, while the medium models support 256K.

Enhanced Coding & Agentic Capabilities – Achieves notable improvements in coding benchmarks alongside native function-calling support, powering highly capable autonomous agents.

Native System Prompt Support – Gemma 4 introduces native support for the system role, enabling more structured and controllable conversations.

⚠️ PLEASE READ THIS BEFORE USING THESE EXPERIMENTAL VERSIONS! ⚠️

An area of personal interest is finding ways to optimize the inference performance of LLMs when deployed in resource-constrained environments like commodity hardware, desktops, laptops, mobiles, edge devices, etc. There are many approaches to accomplish this, including architecture simplification and knowledge distillation, but my focus has been primarily on quantization and pruning.

The method to produce these experimental versions involves using a custom version of llama-imatrix to generate an imatrix that includes tensor statistics, and a custom version of llama-quantize, which computes a per-tensor quantization error, to automatically select the lowest error quantization recipe that achieves a global target bits‑per‑weight (bpw). More details on the implementation and test results here

There are two pull requests (#14891 & #15550) to merge these changes back into the core llama.cpp project. This may or may not ever happen so, until then, the modified versions will be available on GitHub.

For testing and comparison, I use models produced by Bartowski (see credits below) and Unsloth (Daniel and Michael Han do some really interesting stuff!) but when they don't provide versions of the required model, tests and comparisons are against standard quantization obtained by simply running llama-quantize with no further optimizations.

All experimental versions were generated using an appropriate imatrix created from datasets available at eaddario/imatrix-calibration. In llama.cpp, an imatrix is a calibration file derived from running representative text through the model and collecting activation statistics. It is used to weight quantization error so that error in more “important” directions (as estimated from activations) is penalized more heavily.

The process to generate these models is roughly as follows:

Convert the original model's safetensors to GGUF F16
Estimate the Perplexity score for the F16 model (baseline) using the wikitext-2-raw-v1 dataset, and save the logits
Generate an imatrix from the most appropriate calibration dataset
Quantize the baseline model targeting a bpw average (e.g. llama-quantize --target-bpw 4.5678 --state-file --imatrix imatrix.gguf baseline-model-F16.gguf 12)
Calculate Perplexity, KL Divergence, ARC (Easy+Challenge), HellaSwag, MMLU, Truthful QA and WinoGrande scores for each quantized model
Keep version with the best 𝜌PPL and μKLD scores
Repeat until all desired quants are created

Misconceptions about BF16 to F16 Conversion

A common concern when converting BFloat16 (BF16) models to Float16 (F16) is the potential for accuracy loss. Specifically:

Weight Clipping (Overflow): Clipping, or overflow, is often feared but only occurs if a model's weights exceed the range of ±65,503. This is a relatively rare issue in practice.
Subnormal Zeroing (Underflow): A more frequent occurrence is underflow, where weights smaller than approximately 5.96x10⁻⁸ are converted to zero.

Crucially, when the F16 model is subsequently used for quantization, the resulting degradation in metrics like Perplexity (PPL) or Kullback–Leibler Divergence (KLD) is minimal. Any variations are typically restricted to the hundreds or thousandths decimal places compared to the BF16 model.

However, considering that weight clipping presents a more substantial risk to model integrity, every BF16 base model undergoes validation prior to the conversion process. Consequently, no models hosted in this repository exhibit performance degradation due to overflow clipping.

While BF16 offers precision benefits, performance remains a key factor.

Conversion Speed: Tests, such as timing convert_hf_to_gguf.py, show a notable performance difference, with conversion to BF16 being 15–30% slower than to F16.
Inference Speed: A less pronounced but still present difference (3–6%) is observed during inference. Although native BF support has been introduced by many chip manufacturers, the slower performance may stem from the entire software and hardware stack (firmware, libraries, etc.) not being fully optimized yet.

The choice to prioritize F16 over BF16 is driven by a focus on maximizing performance in specific deployment environments. My primary objective is not large-scale quantization production, a domain where others like Bartowski and Unsloth excel at, but rather optimizing inference performance for resource-constrained environments. Since BF16 support is not yet widespread in areas like mobile, edge, and embedded devices, using F16 ensures broader compatibility and easier optimization for these use cases.

Advantages and disadvantages of the global target bits‑per‑weight quantization process

Advantages

Target arbitrary size models
- When specifying --target-bpw 4.5678 for instance, the algorithm will produce a model (nearly) exactly of that size, which is very useful for maximizing VRAM usage. In a system with 24GB VRAM and a 70B model, standard quants might produce a 16.8GB file (too small, quality left on table) or a 24.1GB file (won't fit). This approach can generate a 23.85GB file to utilize the hardware fully.
Data-driven mixed precision often can improve quality at fixed size
- Instead of using hardcoded heuristics (e.g. make attn_v Q5_K for a 70B model), that may be sub‑optimal for a given architecture or size, the quantization mix is determined by the actual error sensitivity of the specific model's weights. This, in practice, often yields a better quality/size trade-off, especially in aggressive quantization scenarios (1.5 to 3.5 bpw), or for unusual architectures.
- Please note: llama.cpp’s heuristics have been tuned across many models and are highly optimized; although the target bpw method produces better quality often (>75% based on tests with 130 models from 11 different families), it can also lose in surprising cases.
Allows better like-for-like comparisons between models and families
- Standard llama.cpp quantization uses hardcoded rules like: "use Q4_K_M, except bump some tensors up/down, except fall back if incompatible, except keep some tensors unquantized..." and for that reason, two different models quantized with the same Q4_K_M type can end up with very different bpw (e.g. 4.75 and 4.30).
- All things being equal, the performance of a model is usually proportional to its overall bpw size; models with a higher bpw tend to perform better than lower bpw models. Since model A has simply been given more bits, it will typically perform better (lower perplexity, better eval scores, etc.) even if the underlying quantization method is identical. That makes comparing the performance not a controlled experiment, because the comparison is between models with different effective compression ratios.
- --target-bpw tries to address that by making the experiment more controlled: each model gets quantized to land on (approximately) the same global byte budget, so that the models' performance differences are more attributable to architecture/training differences, quantization error behaviour at the same compression ratio, optimizer’s allocation decisions, etc.

Disadvantages

Quantization process is significantly slower than standard
- This approach can take 5x-10x longer as it quantizes a sample of most tensors into 15 different formats, dequantizes them back to floats, computes error diffs, and selects the best size/error option that fits the global bpw budget.
- However, the --state-file option will save/use the above-mentioned computations so that future quantizations, for the same model, can be generated at normal speed. It also allows to interrupt the computation process and resume it at a later time.
The optimization target is only a proxy for the model's performance quality
- The process minimizes a per-tensor estimated error computed from sampled rows, not actual perplexity or divergence of output distributions (a future version may address this). Since errors interact nonlinearly across layers, there are no guarantees it will select the best possible quantization recipe subject to the bpw size constraint.
An imatrix with activations data is required for best results
- Activation data is required to compute the bias factor (i.e. the systematic error projected onto activation directions). If the imatrix file does not contain activation data, the --target-bpw option will refuse to run.

Models

Bits per weight, size, perplexity and KL Divergence scores

Model	BPW	Size (GB)	μPPL	𝜌PPL	μKLD	Same Top-P
gemma-4-26B-A4B-it-F16	16.0073	47.0	6617.791728 ±107.205431	100%	N/A	N/A
gemma-4-26B-A4B-it-Q2_K	2.5145	7.4	17569.623360 ±285.143912	40.59%	9.423727 ±0.015560	10.100 ±0.079
gemma-4-26B-A4B-it-Q3_K	3.5000	10.0	8598.460723 ±162.541607	80.01%	3.945886 ±0.012077	41.573 ±0.129
gemma-4-26B-A4B-it-Q4_K	4.5000	13.0	19441.390960 ±385.285232	87.72%	2.067631 ±0.008670	55.808 ±0.129
gemma-4-26B-A4B-it-Q5_K	5.5000	16.0	16520.065420 ±329.228992	90.37%	1.337499 ±0.007116	65.841 ±0.124
gemma-4-26B-A4B-it-Q6_K	6.5000	19.0	17532.470669 ±349.360366	91.90%	1.015299 ±0.006077	70.259 ±0.119
gemma-4-26B-A4B-it-Q7_K	7.4997	22.0	17147.252339 ±340.529496	93.49%	0.636821 ±0.004695	77.572 ±0.109
gemma-4-26B-A4B-it-Q8_0	8.4996	25.0	17152.384071 ±340.978253	93.69%	0.580789 ±0.004459	78.799 ±0.107

ARC, HellaSwag, MMLU, Truthful QA and WinoGrande scores

Scores generated using llama-perplexity with 750 tasks per test, and a context size of 768 tokens.

For the test data used in the generation of these scores, follow the appropriate links: HellaSwag, ARC, Truthful QA, MMLU and WinoGrande

Model	ARC	HellaSwag	MMLU	Truthful QA	WinoGrande	Avg Score
gemma-4-26B-A4B-it-Q2_K	35.4667 ±1.7481	36.13	26.0000 ±1.6027	29.4667 ±1.6658	48.9333 ±1.8265	35.20
gemma-4-26B-A4B-it-Q3_K	44.1333 ±1.8143	54.40	47.6000 ±1.8249	32.2667 ±1.7082	53.7333 ±1.8219	46.43
gemma-4-26B-A4B-it-Q4_K	42.0000 ±1.8034	51.07	49.6000 ±1.8269	32.0000 ±1.7045	56.0000 ±1.8138	46.13
gemma-4-26B-A4B-it-Q5_K	42.2667 ±1.8050	53.47	54.0000 ±1.8211	31.7333 ±1.7007	56.1333 ±1.8132	47.52
gemma-4-26B-A4B-it-Q6_K	41.6000 ±1.8010	52.13	54.4000 ±1.8199	32.5333 ±1.7119	54.4000 ±1.8199	47.01
gemma-4-26B-A4B-it-Q7_K	41.3333 ±1.7993	52.53	53.3333 ±1.8229	32.1333 ±1.7063	55.8667 ±1.8143	47.04
gemma-4-26B-A4B-it-Q8_0	42.0000 ±1.8034	52.27	52.8000 ±1.8241	32.5333 ±1.7119	56.6667 ±1.8106	47.25

Tokens per second benchmarks

Scores generated using llama-bench. Standard (llama-quantize with no optimization) Q4_K_M quantization included for comparison.

model	size	params	backend	threads	test	t/s
gemma-4-26B-A4B-it-Q2_K	7.39 GiB	25.23 B	BLAS,MTL	12	pp512	1401.19 ±20.33
gemma-4-26B-A4B-it-Q2_K	7.39 GiB	25.23 B	BLAS,MTL	12	tg128	96.27 ±1.08
gemma-4-26B-A4B-it-Q2_K	7.39 GiB	25.23 B	BLAS,MTL	12	pp1024+tg1024	171.06 ±3.07
gemma-4-26B-A4B-it-Q3_K	10.28 GiB	25.23 B	BLAS,MTL	12	pp512	1422.88 ±5.82
gemma-4-26B-A4B-it-Q3_K	10.28 GiB	25.23 B	BLAS,MTL	12	tg128	99.25 ±0.85
gemma-4-26B-A4B-it-Q3_K	10.28 GiB	25.23 B	BLAS,MTL	12	pp1024+tg1024	156.23 ±1.91
gemma-4-26B-A4B-it-Q4_K	13.22 GiB	25.23 B	BLAS,MTL	12	pp512	1385.18 ±7.25
gemma-4-26B-A4B-it-Q4_K	13.22 GiB	25.23 B	BLAS,MTL	12	tg128	94.55 ±1.08
gemma-4-26B-A4B-it-Q4_K	13.22 GiB	25.23 B	BLAS,MTL	12	pp1024+tg1024	156.71 ±0.33
gemma-4-26B-A4B-it-Q5_K	16.16 GiB	25.23 B	BLAS,MTL	12	pp512	1321.55 ±5.68
gemma-4-26B-A4B-it-Q5_K	16.16 GiB	25.23 B	BLAS,MTL	12	tg128	81.25 ±0.24
gemma-4-26B-A4B-it-Q5_K	16.16 GiB	25.23 B	BLAS,MTL	12	pp1024+tg1024	142.95 ±3.71
gemma-4-26B-A4B-it-Q6_K	19.09 GiB	25.23 B	BLAS,MTL	12	pp512	1328.48 ±8.00
gemma-4-26B-A4B-it-Q6_K	19.09 GiB	25.23 B	BLAS,MTL	12	tg128	82.00 ±0.14
gemma-4-26B-A4B-it-Q6_K	19.09 GiB	25.23 B	BLAS,MTL	12	pp1024+tg1024	136.02 ±3.90
gemma-4-26B-A4B-it-Q7_K	22.03 GiB	25.23 B	BLAS,MTL	12	pp512	1414.49 ±5.13
gemma-4-26B-A4B-it-Q7_K	22.03 GiB	25.23 B	BLAS,MTL	12	tg128	75.55 ±0.16
gemma-4-26B-A4B-it-Q7_K	22.03 GiB	25.23 B	BLAS,MTL	12	pp1024+tg1024	131.36 ±0.31
gemma-4-26B-A4B-it-Q8_0	24.97 GiB	25.23 B	BLAS,MTL	12	pp512	1435.65 ±2.43
gemma-4-26B-A4B-it-Q8_0	24.97 GiB	25.23 B	BLAS,MTL	12	tg128	73.95 ±0.23
gemma-4-26B-A4B-it-Q8_0	24.97 GiB	25.23 B	BLAS,MTL	12	pp1024+tg1024	128.04 ±0.23

Metrics used

Perplexity: one of the key metrics used in NLP evaluation. It measures the quality of a language model by evaluating how well it predicts the next token given a particular sequence of words. A PPL of 1 indicates an exact match between predicted and actual, whereas values greater than one indicate a degree of "surprise" the generated token differs from the expected.

Kullback–Leibler (KL) Divergence: a statistical measure of how much a probability distribution differs from another. When quantizing models (or altering the original tensors in any way for that matter), the closest we can preserve the weights' probability distribution to the original model the better, thus the closest to 0 the better.

AI2 Reasoning Challenge (ARC): a benchmark to evaluate the ability of AI models to answer complex science questions that require logical reasoning beyond pattern matching.

HellaSwag: the Harder Endings, Longer contexts, and Low-shot Activities for Situations With Adversarial Generations (bit of a mouthful!) is a benchmark designed to test commonsense natural language inference. It requires the model to predict the most likely ending of a sentence.

MMLU: the Massive Multitask Language Understanding evaluates LLMs’ general knowledge and problem-solving abilities across 57 subjects, including elementary mathematics, US history, computer science, and law.

Truthful QA: evaluates how well LLMs generate truthful responses to questions. It identifies whether AI models can avoid generating false or misleading information, particularly in areas where human knowledge is prone to misconceptions.

Winogrande: based on the Winograd Schema Challenge, is a natural language understanding task requiring models to resolve ambiguities in sentences involving pronoun references.

Credits

LLaMa C++ has a large and vibrant community of contributors (~1,600 last time I checked) that actively maintain and extend its functionality, adding new models and architectures almost as fast as they appear. Considering the breakneck speed at which the AI/ML field is advancing, this alone is a remarkable feat!

While I'm grateful to all contributors, I want to recognise three in particular:

Colin Kealty (Bartowski), for the many contributions and for being one of the best sources of high quality quantized models available on Hugging Face
Georgi Gerganov for his amazing work with llama.cpp and the ggml/gguf libraries
Iwan Kawrakow for being one of the key authors behind the many quantization algorithms and the imatrix functionality.

Downloads last month: 4,478

GGUF

Model size

25B params

Architecture

gemma4

Hardware compatibility

2-bit

6-bit

8-bit

16-bit

View +2 variants

Inference Providers NEW

Any-to-Any

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for eaddario/gemma-4-26B-A4B-it-GGUF

Base model

google/gemma-4-26B-A4B

Finetuned

google/gemma-4-26B-A4B-it

Quantized

(188)

this model

eaddario
/

gemma-4-26B-A4B-it-GGUF