KL
LLM Sizing Calculator
ไทย
EN
kl-itonline.com →
Inference
Fine-tuning
Params (B)
Active (B)
Layers
KV heads
Head dim
Weight quantization
KV cache precision
Context length (tokens)
tokens
QLoRA (base 4-bit)
LoRA (base BF16)
Full fine-tune (BF16 mixed)
LoRA rank
8
16
32
64
128
Sequence length
FlashAttention / SDPA
GPU
1
2
3
4
6
8
Runtime
vLLM / TensorRT-LLM
llama.cpp / Ollama / LM Studio
—
0
GB
GB
GPU
VRAM
BW GB/s
TDP
tok/s
NVIDIA DGX / HGX
CPU platform
Intel Xeon (W / Scalable)
AMD Ryzen Threadripper PRO
AMD EPYC
Intel Core (i9 / Ultra)
AMD Ryzen 9
Cores
Clock (GHz)
RAM (GB)
Storage (TB)
NVMe SSD
SATA SSD
NVMe + HDD
LINE ID
@462lkloe
sales@klmogroup.com
095-431-8124