Mungert/MetaStone-L1-7B-GGUF

MetaStone-L1-7B 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

MetaStone-L1-7B-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.

MetaStone-L1-7B-f16.gguf

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

MetaStone-L1-7B-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.

MetaStone-L1-7B-f16-q8_0.gguf

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

MetaStone-L1-7B-q4_k.gguf

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

MetaStone-L1-7B-q4_k_s.gguf

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

MetaStone-L1-7B-q6_k.gguf

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

MetaStone-L1-7B-q8_0.gguf

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

MetaStone-L1-7B-iq3_xs.gguf

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

MetaStone-L1-7B-iq3_m.gguf

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

MetaStone-L1-7B-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).

Introduction MetaStone-L1 is the lite reasoning model of the MetaStone series, which aims to enhance the performance in hard downstream tasks.

On core reasoning benchmarks including mathematics and code, MetaStone-L1-7B achieved SOTA results in the parallel-level models, and it also achieved the comparable results as the API models such as Claude-3.5-Sonnet-1022 and GPT4o-0513.

This repo contains the MetaStone-L1-7B model, which is trained based on DeepSeek-R1-Distill-Qwen-7B by GRPO. For full details of this model please refer to our release blog.

Requirements

We advise you to use the latest version of transformers(transformers==4.48.3). For the best experience, please review the Usage Guidelines.

Quickstart

Here give the example of how to use our model.

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "MetaStoneTec/MetaStone-L1-7B"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

messages = [
    {"role": "user", "content": "Complete the square for the following quadratic: $-x^2+7 x-11$\n\nPlease reason step by step, and put your final answer within \\boxed{}."}
]

text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=32768
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]

Usage Guidelines

To achieve optimal performance, we recommend the following settings:

1. Enhace the thoughful output：

   a. Make sure the model starts with <think>\n to prevent generating empty think content. If you use apply_chat_template and set add_generation_prompt=True, this is automatically implemented, but this may result in replies not having a tag at the beginning, which is normal.

   b. Ensure the final input of the model is in the format of <｜User｜> [your prompt] <｜Assistant｜><think>.

2. Use a temperature of 0.6, a top sampling probability of 0.95, a maximum generation length of 32k.

3. Standardize output format: We recommend using hints to standardize model outputs when benchmarking.

   a. Math questions: Add a statement "Please reason step by step, and put your final answer within \\boxed{}." to the prompt.

   b. Code problems: Add "### Format: Read the inputs from stdin solve the problem and write the answer to stdout. Enclose your code within delimiters as follows.\n ```python\n# YOUR CODE HERE\n```\n### Answer: (use the provided format with backticks)" to the prompt.

4. In particular, we use latex2sympy2 and sympy to assist in judging complex Latex formats for the Math500 evaluation script. For all datasets, we generate 64 responses per query to estimate pass@1.

Citation

If you find our work helpful, feel free to give us a cite.

@misc{MetaStoneL17B,
    title = {MetastoneL17B},
    url = {https://huggingface.co/MetaStoneTec/MetaStone-L1-7B},
    author = {MetaStone Team},
    month = {March},
    year = {2025}
}

@article{wang2024graph,
  title={A Graph-Based Synthetic Data Pipeline for Scaling High-Quality Reasoning Instructions},
  author={Wang, Jiankang and Xu, Jianjun and Wang, Xiaorui and Wang, Yuxin and Xing, Mengting and Fang, Shancheng and Chen, Zhineng and Xie, Hongtao and Zhang, Yongdong},
  journal={arXiv preprint arXiv:2412.08864},
  year={2024}
}