---

license: apache-2.0

base_model:

- RedHatAI/gemma-4-31B-it-speculator.eagle3

- gemma-4-31B-it

base_model_relation: quantized

library_name: llama.cpp

tags:

- gguf

- llama.cpp

- eagle3

- speculative-decoding

- speculator

- draft-model

- gemma-4

- gemma

- redhatai

pipeline_tag: text-generation

---

Gemma 4 31B IT EAGLE3 Speculator GGUF

This repository contains GGUF conversions and quantizations of RedHatAI/gemma-4-31B-it-speculator.eagle3 for use with llama.cpp EAGLE3 speculative decoding.

> [!IMPORTANT]

> This is not a standalone chat model. It is an EAGLE3 draft/speculator model and must be used together with the matching target/verifier model.

• Target model: gemma-4-31B-it
• Speculator source: RedHatAI/gemma-4-31B-it-speculator.eagle3
• Runtime: llama.cpp with --spec-type draft-eagle3

Files

| File | Type | Notes |

| ---------------------------------------------- | -------------------------------- | ------------------------------------------------------------- |

| gemma-4-31B-it-speculator.eagle3-F16.gguf | EAGLE3 speculator GGUF | Converted from the original RedHatAI safetensors checkpoint |

| gemma-4-31B-it-speculator.eagle3-Q8_0.gguf | Quantized EAGLE3 speculator GGUF | Quantized from the F16 GGUF |

| gemma-4-31B-it-speculator.eagle3-Q4_K_M.gguf | Quantized EAGLE3 speculator GGUF | Quantized from the F16 GGUF; may be faster for draft decoding |

Usage with llama.cpp

Example:

llama-server \
  -m gemma-4-31B-it-Q4_K_M.gguf \
  -md gemma-4-31B-it-speculator.eagle3-Q4_K_M.gguf \
  --spec-type draft-eagle3 \
  --spec-draft-n-max 4 \
  --spec-draft-p-min 0.5 \
  -c 32768 \
  -ngl 99 \
  -fa on

Windows CMD example:

llama-server.exe ^
  -m gemma-4-31B-it-Q4_K_M.gguf ^
  -md gemma-4-31B-it-speculator.eagle3-Q4_K_M.gguf ^
  --spec-type draft-eagle3 ^
  --spec-draft-n-max 4 ^
  --spec-draft-p-min 0.5 ^
  -c 32768 ^
  -ngl 99 ^
  -fa on

PowerShell example:

.\llama-server.exe `
  -m "gemma-4-31B-it-Q4_K_M.gguf" `
  -md "gemma-4-31B-it-speculator.eagle3-Q4_K_M.gguf" `
  --spec-type draft-eagle3 `
  --spec-draft-n-max 4 `
  --spec-draft-p-min 0.5 `
  -c 32768 `
  -ngl 99 `
  -fa on

Important Notes

This GGUF file is only the draft/speculator model. You still need a compatible GGUF of the target model, such as a GGUF conversion or quantization of gemma-4-31B-it.

Do not use this speculator with unrelated models such as Gemma 4 12B, Gemma 3, Qwen, Llama, Mistral, or other non-matching models. EAGLE3 speculators are target-specific.

Even small differences in the target model, prompt format, quantization, or runtime settings may affect draft acceptance rate and overall speed.

Tested Configuration

Tested with:

• Runtime: llama.cpp with EAGLE3 support
• Target model: Gemma 4 31B IT GGUF
• Draft model: this EAGLE3 GGUF
• Example settings:

* --spec-type draft-eagle3

* --spec-draft-n-max 4

* --spec-draft-p-min 0.5

Local benchmark observations may vary depending on GPU, quantization, context length, batch size, sampling settings, and prompt type.

Benchmark Notes

In local testing, EAGLE3 showed stronger gains on structured or predictable outputs such as:

• code generation
• JSON output
• repeated pattern generation
• summarization
• code completion

It was less effective on some tasks such as translation, creative writing, and some open-ended prompts, where draft acceptance may be lower.

For this speculator, lower-bit draft quantization may perform better than F16 in some llama.cpp setups because the draft model becomes cheaper to run. In local testing, Q4_K_M was a practical candidate for draft decoding, while results may vary across hardware and runtime versions.

On smaller VRAM setups, the extra draft/speculator model may reduce the practical benefit of EAGLE3. In those cases, native MTP models may be more efficient.

Conversion

Converted with llama.cpp convert_hf_to_gguf.py using the original speculator repository and the matching target model directory.

Example conversion command:

python convert_hf_to_gguf.py \
  RedHatAI/gemma-4-31B-it-speculator.eagle3 \
  --outtype f16 \
  --target-model-dir gemma-4-31B-it \
  --outfile gemma-4-31B-it-speculator.eagle3-F16.gguf

PowerShell example:

python .\convert_hf_to_gguf.py `
  "E:\OLLAMA_MODELS\gemma-4-31B-it-speculator.eagle3" `
  --outtype f16 `
  --target-model-dir "E:\OLLAMA_MODELS\gemma-4-31B-it" `
  --outfile "E:\OLLAMA_MODELS\gemma-4-31B-it-speculator.eagle3-F16.gguf"

Quantization

The F16 GGUF can be quantized with llama-quantize.

Q8_0 example:

llama-quantize \
  gemma-4-31B-it-speculator.eagle3-F16.gguf \
  gemma-4-31B-it-speculator.eagle3-Q8_0.gguf \
  Q8_0

Q4_K_M example:

llama-quantize \
  gemma-4-31B-it-speculator.eagle3-F16.gguf \
  gemma-4-31B-it-speculator.eagle3-Q4_K_M.gguf \
  Q4_K_M

PowerShell example:

.\llama-quantize.exe `
  "E:\OLLAMA_MODELS\gemma-4-31B-it-speculator.eagle3-F16.gguf" `
  "E:\OLLAMA_MODELS\gemma-4-31B-it-speculator.eagle3-Q4_K_M.gguf" `
  Q4_K_M

Credits

Original EAGLE3 speculator model by RedHatAI:

• RedHatAI/gemma-4-31B-it-speculator.eagle3

Target model:

• gemma-4-31B-it

GGUF support and runtime:

• ggml-org/llama.cpp

License

This repository is a converted GGUF version of the original speculator model. The original model license and usage terms apply. Please refer to the upstream repositories for full license details.