qmemcalc 0.1.1

GPU Memory Estimator for QLoRA / LoRA / Transformers

qmemcalc

Quantized Memory Calculator for Transformers (QLoRA / LoRA / Full Fine-Tuning)

Estimate GPU memory requirements for training or fine-tuning transformer models with ease. Supports multiple precision types, quantization, and LoRA configurations.

---

Features

• Estimate GPU memory for FP32/FP16/BF16
• Supports 4-bit / 8-bit quantization
• LoRA rank or fraction support
• Extended optimizer support
• Gradient checkpointing
• CLI for quick estimation

Installation

pip install qmemcalc

Python API Usage

from qmemcalc import estimate_memory

result = estimate_memory(
    model_name="sshleifer/tiny-gpt2",
    batch_size=2,
    seq_len=16,
    lora_r=4,
    precision="fp16",
    quantization="4bit"
)

print(result)

Contributing

Contributions are welcome! Please open issues or submit pull requests.

git clone https://github.com/sachin62025/qmemcalc.git
cd qmemcalc
pip install -e .