KULLM 12.8B
Korea University · dense · 12.8B parameters · 4,096 context
Parameters
12.8B
Context Window
4K tokens
Architecture
Dense
Best GPU
A100 40GB SXM
Intelligence Brief
KULLM 12.8B is a 12.8B parameter DENSE model from Korea University, featuring Multi-Head Attention (MHA) with 40 layers and 5,120 hidden dimensions. With a 4,096 token context window, it supports multilingual. For self-hosted inference, A100 40GB SXM delivers optimal throughput at $807/month.
Recent changes
Loading…
Related models
3 suggestions
Picks: same family first, then same vendor within ±2× params, then top tag-overlap matches. Price shown is the cheapest Output $/M across providers — the row's page shows the canonical anchor.
Architecture Details
Memory Requirements
BF16 Weights
25.6 GB
FP8 Weights
12.8 GB
INT4 Weights
6.4 GB
GPU Compatibility Matrix
KULLM 12.8B is compatible with 82% of GPU configurations across 41 GPUs at 3 precision levels.
GPU Recommendations
BF16 · 1 GPU · vllm
95/100
score
Throughput
328.0 tok/s
Latency (ITL)
3.0ms
Est. TTFT
1ms
Cost/Month
$807
Cost/M Tokens
$0.94
BF16 · 1 GPU · vllm
95/100
score
Throughput
162.0 tok/s
Latency (ITL)
6.2ms
Est. TTFT
1ms
Cost/Month
$465
Cost/M Tokens
$1.09
BF16 · 1 GPU · vllm
95/100
score
Throughput
146.8 tok/s
Latency (ITL)
6.8ms
Est. TTFT
1ms
Cost/Month
$399
Cost/M Tokens
$1.03
Deployment Options
API Deployment
No API pricing available
Single GPU
A100 40GB SXM
$807/mo
Min VRAM: 13 GB
Multi-GPU
RTX 3090 x2
324.1 tok/s
TP· $361/mo
API Pricing Comparison
No API pricing data available for this model.
Performance Estimates
Throughput by GPU
VRAM Breakdown (A100 40GB SXM, BF16)
Precision Impact
bf16
25.6 GB
weights/GPU
~328.0 tok/s
fp8
12.8 GB
weights/GPU
int4
6.4 GB
weights/GPU
Capabilities
Features
Supported Frameworks
Supported Precisions
Where to Deploy KULLM 12.8B
Similar Models
OLMo 2 13B
13B params · dense
Quality: 50
Baichuan 2 13B
13B params · dense
Quality: 50
from $0.25/M
Cerebras GPT 13B
13B params · dense
Quality: 50
ELYZA 13B
13B params · dense
Quality: 50
Vicuna 13B
13B params · dense
Quality: 50
Frequently Asked Questions
How much VRAM does KULLM 12.8B need for inference?
KULLM 12.8B requires approximately 25.6 GB of VRAM at BF16 precision, 12.8 GB at FP8, or 6.4 GB at INT4 quantization. Additional VRAM is needed for KV-cache (409600 bytes per token) and activations (~1.00 GB).
What is the best GPU for KULLM 12.8B?
The top recommended GPU for KULLM 12.8B is the A100 40GB SXM using BF16 precision. It achieves approximately 328.0 tokens/sec at an estimated cost of $807/month ($0.94/M tokens). Score: 95/100.
How much does KULLM 12.8B inference cost?
KULLM 12.8B inference costs vary by provider and GPU setup. Use our calculator for detailed cost estimates across all providers.