Parameters
1B
Context Length
128K
Modality
Text
Architecture
Dense
KV Cache Rate
16 KB/tok
License
Llama 3.2 Community License
Release Date
25 Sept 2024
Knowledge Cutoff
Dec 2023
API Pricing (per 1M)
Input: $0.03 · Output: $0.20
Precision
Context
Workload Memory: 3.09 GB
INT4 · 8K context
Base Weights: 0.47 GB
KV Cache: 0.13 GB
Runtime & Graphs: 2.50 GB
1x NVIDIA L4 24GB
3.1 / 21.6 GB usable
1x RTX 3060 12GB
3.1 / 10.8 GB usable
Apple M4 24GB
3.1 / 18.0 GB usable
Determine requirements for your specific workload:
Open in VRAM Calculator
Checkpoint format
FP16
CUDA
>= 11.8
PyTorch
>= 2.1
Transformers
>= 4.52.0.dev0
| Architecture | Precision | Attention kernel |
|---|---|---|
Blackwell, Hopper | Native FP16 | FlashAttention-3 |
Ada Lovelace, Ampere | Native FP16 | FlashAttention-2 |
Rank
#235
| Benchmark | Score | Rank |
|---|---|---|
General Text | 1111 | 195 |
Intelligence Index | 0.05 | 330 |
Overall Rank
#235
Coding Rank
-
Llama 3.2 1B is an ultra-compact edge model by Meta optimized for on-device summarization, rewriting, and multilingual retrieval. Distilled from larger Llama 3.1 models, it supports local inference across a 128K token context window.
Attention
Attention Structure
Grouped-Query Attention
Attention Heads
16
Key-Value Heads
4
Attention Head Dimension
64
Position Embedding
ROPE
RoPE Theta
500,000
Normalization
RMS Normalization
Activation Function
SwigLU
Dimensions
Base Precision
FP16
Hidden Dimension Size
1,024
Number of Layers
16
FFN Intermediate Size (Dense)
8,192
Tokenizer
Vocabulary Size
128,256
Meta's Llama 3.2 family introduces vision models, integrating image encoders with language models for multimodal text and image processing. It also includes lightweight variants optimized for efficient on-device deployment, supporting an extended 128K token context length.