Parameters
3B
Context Length
33K
Modality
Text
Architecture
Dense
KV Cache Rate
60 KB/tok
License
other
Release Date
5 Oct 2026
Knowledge Cutoff
-
API Pricing (per 1M)
Self-hosted only
Precision
Context
Workload Memory: 4.37 GB
INT4 · 8K context
Base Weights: 1.40 GB
KV Cache: 0.47 GB
Runtime & Graphs: 2.50 GB
1x NVIDIA L4 24GB
4.4 / 21.6 GB usable
1x RTX 3060 12GB
4.4 / 10.8 GB usable
Apple M4 24GB
4.4 / 18.0 GB usable
Determine requirements for your specific workload:
Open in VRAM Calculator
Checkpoint format
BF16
CUDA
>= 11.8
PyTorch
>= 2.1
Transformers
>= 5.14.1
| Architecture | Precision | Attention kernel |
|---|---|---|
Blackwell, Hopper | Native BF16 | FlashAttention-3 |
Ada Lovelace, Ampere | Native BF16 | FlashAttention-2 |
No evaluation benchmarks for Liquid D1-VL 3B available.
Overall Rank
-
Coding Rank
-
Liquid D1-VL 3B is a compact multimodal model based on the LFM2 vision-language architecture, designed for efficient edge deployment and vision reasoning tasks.
Attention
Attention Structure
Grouped-Query Attention
Attention Heads
32
Key-Value Heads
8
Attention Head Dimension
64
Position Embedding
ROPE
RoPE Theta
1,000,000
Linear Attention
Yes
Linear Attention Ratio
73.3%
Normalization
RMS Normalization
Activation Function
SwigLU
Dimensions
Base Precision
BF16
Hidden Dimension Size
2,048
Number of Layers
30
FFN Intermediate Size (Dense)
10,752
Tokenizer
Vocabulary Size
128,000
The LFM2 model family developed by Liquid AI.