Mungert/EXAONE-Deep-2.4B-GGUF

EXAONE-Deep-2.4B GGUF Models

Choosing the Right Model Format

Selecting the correct model format depends on your hardware capabilities and memory constraints.

BF16 (Brain Float 16) – Use if BF16 acceleration is available

• A 16-bit floating-point format designed for faster computation while retaining good precision.
• Provides similar dynamic range as FP32 but with lower memory usage.
• Recommended if your hardware supports BF16 acceleration (check your device’s specs).
• Ideal for high-performance inference with reduced memory footprint compared to FP32.

📌 Use BF16 if:
✔ Your hardware has native BF16 support (e.g., newer GPUs, TPUs).
✔ You want higher precision while saving memory.
✔ You plan to requantize the model into another format.

📌 Avoid BF16 if:
❌ Your hardware does not support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.

---

F16 (Float 16) – More widely supported than BF16

• A 16-bit floating-point high precision but with less of range of values than BF16.
• Works on most devices with FP16 acceleration support (including many GPUs and some CPUs).
• Slightly lower numerical precision than BF16 but generally sufficient for inference.

📌 Use F16 if:
✔ Your hardware supports FP16 but not BF16.
✔ You need a balance between speed, memory usage, and accuracy.
✔ You are running on a GPU or another device optimized for FP16 computations.

📌 Avoid F16 if:
❌ Your device lacks native FP16 support (it may run slower than expected).
❌ You have memory limitations.

---

Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference

Quantization reduces model size and memory usage while maintaining as much accuracy as possible.

• Lower-bit models (Q4_K) → Best for minimal memory usage, may have lower precision.
• Higher-bit models (Q6_K, Q8_0) → Better accuracy, requires more memory.

📌 Use Quantized Models if:
✔ You are running inference on a CPU and need an optimized model.
✔ Your device has low VRAM and cannot load full-precision models.
✔ You want to reduce memory footprint while keeping reasonable accuracy.

📌 Avoid Quantized Models if:
❌ You need maximum accuracy (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).

---

Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)

These models are optimized for extreme memory efficiency, making them ideal for low-power devices or large-scale deployments where memory is a critical constraint.

• IQ3_XS: Ultra-low-bit quantization (3-bit) with extreme memory efficiency.

  • Use case: Best for ultra-low-memory devices where even Q4_K is too large.
  • Trade-off: Lower accuracy compared to higher-bit quantizations.

• IQ3_S: Small block size for maximum memory efficiency.

  • Use case: Best for low-memory devices where IQ3_XS is too aggressive.

• IQ3_M: Medium block size for better accuracy than IQ3_S.

  • Use case: Suitable for low-memory devices where IQ3_S is too limiting.

• Q4_K: 4-bit quantization with block-wise optimization for better accuracy.

  • Use case: Best for low-memory devices where Q6_K is too large.

• Q4_0: Pure 4-bit quantization, optimized for ARM devices.

  • Use case: Best for ARM-based devices or low-memory environments.

---

Summary Table: Model Format Selection

Model Format	Precision	Memory Usage	Device Requirements	Best Use Case
BF16	Highest	High	BF16-supported GPU/CPUs	High-speed inference with reduced memory
F16	High	High	FP16-supported devices	GPU inference when BF16 isn’t available
Q4_K	Medium Low	Low	CPU or Low-VRAM devices	Best for memory-constrained environments
Q6_K	Medium	Moderate	CPU with more memory	Better accuracy while still being quantized
Q8_0	High	Moderate	CPU or GPU with enough VRAM	Best accuracy among quantized models
IQ3_XS	Very Low	Very Low	Ultra-low-memory devices	Extreme memory efficiency and low accuracy
Q4_0	Low	Low	ARM or low-memory devices	llama.cpp can optimize for ARM devices

---

Included Files & Details

EXAONE-Deep-2.4B-bf16.gguf

• Model weights preserved in BF16.
• Use this if you want to requantize the model into a different format.
• Best if your device supports BF16 acceleration.

EXAONE-Deep-2.4B-f16.gguf

• Model weights stored in F16.
• Use if your device supports FP16, especially if BF16 is not available.

EXAONE-Deep-2.4B-bf16-q8_0.gguf

• Output & embeddings remain in BF16.
• All other layers quantized to Q8_0.
• Use if your device supports BF16 and you want a quantized version.

EXAONE-Deep-2.4B-f16-q8_0.gguf

• Output & embeddings remain in F16.
• All other layers quantized to Q8_0.

EXAONE-Deep-2.4B-q4_k.gguf

• Output & embeddings quantized to Q8_0.
• All other layers quantized to Q4_K.
• Good for CPU inference with limited memory.

EXAONE-Deep-2.4B-q4_k_s.gguf

• Smallest Q4_K variant, using less memory at the cost of accuracy.
• Best for very low-memory setups.

EXAONE-Deep-2.4B-q6_k.gguf

• Output & embeddings quantized to Q8_0.
• All other layers quantized to Q6_K .

EXAONE-Deep-2.4B-q8_0.gguf

• Fully Q8 quantized model for better accuracy.
• Requires more memory but offers higher precision.

EXAONE-Deep-2.4B-iq3_xs.gguf

• IQ3_XS quantization, optimized for extreme memory efficiency.
• Best for ultra-low-memory devices.

EXAONE-Deep-2.4B-iq3_m.gguf

• IQ3_M quantization, offering a medium block size for better accuracy.
• Suitable for low-memory devices.

EXAONE-Deep-2.4B-q4_0.gguf

• Pure Q4_0 quantization, optimized for ARM devices.
• Best for low-memory environments.
• Prefer IQ4_NL for better accuracy.

🚀 If you find these models useful

Please click like ❤ . Also I’d really appreciate it if you could test my Network Monitor Assistant at 👉 Network Monitor Assitant.

💬 Click the chat icon (bottom right of the main and dashboard pages) . Choose a LLM; toggle between the LLM Types TurboLLM -> FreeLLM -> TestLLM.

What I'm Testing

I'm experimenting with function calling against my network monitoring service. Using small open source models. I am into the question "How small can it go and still function".

🟡 TestLLM – Runs the current testing model using llama.cpp on 6 threads of a Cpu VM (Should take about 15s to load. Inference speed is quite slow and it only processes one user prompt at a time—still working on scaling!). If you're curious, I'd be happy to share how it works! .

The other Available AI Assistants

🟢 TurboLLM – Uses gpt-4o-mini Fast! . Note: tokens are limited since OpenAI models are pricey, but you can Login or Download the Free Network Monitor agent to get more tokens, Alternatively use the FreeLLM .

🔵 FreeLLM – Runs open-source Hugging Face models Medium speed (unlimited, subject to Hugging Face API availability).

EXAONE-Deep-2.4B

Introduction

We introduce EXAONE Deep, which exhibits superior capabilities in various reasoning tasks including math and coding benchmarks, ranging from 2.4B to 32B parameters developed and released by LG AI Research. Evaluation results show that 1) EXAONE Deep 2.4B outperforms other models of comparable size, 2) EXAONE Deep 7.8B outperforms not only open-weight models of comparable scale but also a proprietary reasoning model OpenAI o1-mini, and 3) EXAONE Deep 32B demonstrates competitive performance against leading open-weight models.

For more details, please refer to our documentation, blog and GitHub.

This repository contains the reasoning 2.4B language model with the following features:

• Number of Parameters (without embeddings): 2.14B
• Number of Layers: 30
• Number of Attention Heads: GQA with 32 Q-heads and 8 KV-heads
• Vocab Size: 102,400
• Context Length: 32,768 tokens
• Tie Word Embeddings: True (unlike 7.8B and 32B models)

Note

The EXAONE Deep models are trained with an optimized configuration, so we recommend following the Usage Guideline section to achieve optimal performance.

Evaluation

The following table shows the evaluation results of reasoning tasks such as math and coding. The full evaluation results can be found in the documentation.

Models	MATH-500 (pass@1)	AIME 2024 (pass@1 / cons@64)	AIME 2025 (pass@1 / cons@64)	CSAT Math 2025 (pass@1)	GPQA Diamond (pass@1)	Live Code Bench (pass@1)
EXAONE Deep 32B	95.7	72.1 / 90.0	65.8 / 80.0	94.5	66.1	59.5
DeepSeek-R1-Distill-Qwen-32B	94.3	72.6 / 83.3	55.2 / 73.3	84.1	62.1	57.2
QwQ-32B	95.5	79.5 / 86.7	67.1 / 76.7	94.4	63.3	63.4
DeepSeek-R1-Distill-Llama-70B	94.5	70.0 / 86.7	53.9 / 66.7	88.8	65.2	57.5
DeepSeek-R1 (671B)	97.3	79.8 / 86.7	66.8 / 80.0	89.9	71.5	65.9
EXAONE Deep 7.8B	94.8	70.0 / 83.3	59.6 / 76.7	89.9	62.6	55.2
DeepSeek-R1-Distill-Qwen-7B	92.8	55.5 / 83.3	38.5 / 56.7	79.7	49.1	37.6
DeepSeek-R1-Distill-Llama-8B	89.1	50.4 / 80.0	33.6 / 53.3	74.1	49.0	39.6
OpenAI o1-mini	90.0	63.6 / 80.0	54.8 / 66.7	84.4	60.0	53.8
EXAONE Deep 2.4B	92.3	52.5 / 76.7	47.9 / 73.3	79.2	54.3	46.6
DeepSeek-R1-Distill-Qwen-1.5B	83.9	28.9 / 52.7	23.9 / 36.7	65.6	33.8	16.9

Deployment

EXAONE Deep models can be inferred in the various frameworks, such as:

• TensorRT-LLM
• vLLM
• SGLang
• llama.cpp
• Ollama
• LM-Studio

Please refer to our EXAONE Deep GitHub for more details about the inference frameworks.

Quantization

We provide the pre-quantized EXAONE Deep models with AWQ and several quantization types in GGUF format. Please refer to our EXAONE Deep collection to find corresponding quantized models.

Usage Guideline

To achieve the expected performance, we recommend using the following configurations:

1. Ensure the model starts with <thought>\n for reasoning steps. The model's output quality may be degraded when you omit it. You can easily apply this feature by using tokenizer.apply_chat_template() with add_generation_prompt=True. Please check the example code on Quickstart section.
2. The reasoning steps of EXAONE Deep models enclosed by <thought>\n...\n</thought> usually have lots of tokens, so previous reasoning steps may be necessary to be removed in multi-turn situation. The provided tokenizer handles this automatically.
3. Avoid using system prompt, and build the instruction on the user prompt.
4. Additional instructions help the models reason more deeply, so that the models generate better output.
   • For math problems, the instructions "Please reason step by step, and put your final answer within \boxed{}." are helpful.
   • For more information on our evaluation setting including prompts, please refer to our Documentation.
5. In our evaluation, we use temperature=0.6 and top_p=0.95 for generation.
6. When evaluating the models, it is recommended to test multiple times to assess the expected performance accurately.

Limitation

The EXAONE language model has certain limitations and may occasionally generate inappropriate responses. The language model generates responses based on the output probability of tokens, and it is determined during learning from training data. While we have made every effort to exclude personal, harmful, and biased information from the training data, some problematic content may still be included, potentially leading to undesirable responses. Please note that the text generated by EXAONE language model does not reflects the views of LG AI Research.

• Inappropriate answers may be generated, which contain personal, harmful or other inappropriate information.
• Biased responses may be generated, which are associated with age, gender, race, and so on.
• The generated responses rely heavily on statistics from the training data, which can result in the generation of semantically or syntactically incorrect sentences.
• Since the model does not reflect the latest information, the responses may be false or contradictory.

LG AI Research strives to reduce potential risks that may arise from EXAONE language models. Users are not allowed to engage in any malicious activities (e.g., keying in illegal information) that may induce the creation of inappropriate outputs violating LG AI’s ethical principles when using EXAONE language models.

License

The model is licensed under EXAONE AI Model License Agreement 1.1 - NC

Citation

@article{exaone-deep,
  title={EXAONE Deep: Reasoning Enhanced Language Models},
  author={{LG AI Research}},
  journal={arXiv preprint arXiv:2503.12524},
  year={2025}
}

Contact

LG AI Research Technical Support: [email protected]