ApX logoApX logo

Alpamayo2 Edge

Parameters

2.65B

Context Length

131K

Modality

Text

Architecture

Dense

KV Cache Rate

112 KB/tok

License

openmdw-1.1

Release Date

12 Aug 2026

Knowledge Cutoff

-

API Pricing (per 1M)

Self-hosted only

Hardware & System Requirements

Precision

Native

Context

Workload Memory: 4.61 GB

INT4 · 8K context

Base Weights: 1.23 GB

KV Cache: 0.88 GB

Runtime & Graphs: 2.50 GB

Minimum Required Hardware

1x NVIDIA L4 24GB

4.6 / 21.6 GB usable

1x RTX 3060 12GB

4.6 / 10.8 GB usable

Apple M4 24GB

4.6 / 18.0 GB usable

Determine requirements for your specific workload:

Open in VRAM Calculator

Runtime requirements

Checkpoint format

FP16

CUDA

>= 11.8

PyTorch

>= 2.1

Transformers

>= 5.17.0

ArchitecturePrecisionAttention kernel

Blackwell, Hopper

Native FP16

FlashAttention-3

Ada Lovelace, Ampere

Native FP16

FlashAttention-2

Architecture Diagram

Input TokensToken EmbeddingPosition: RoPEHidden: 2k · Context: 131K · Vocab: 135.2kx 28 layersRMSNormPre-AttentionGrouped-Query Attention16Q / 8KV headsHead dim: 128+RMSNormPre-FFNFeed-Forward NetworkReLU²Intermediate: 9.2k+Final RMSNormOutput Logits

Evaluation Benchmarks

No evaluation benchmarks for Alpamayo2 Edge available.

Rankings

Overall Rank

-

Coding Rank

-

About Alpamayo2 Edge

Alpamayo2 Edge is an efficient edge-optimized multimodal vision-language model developed by NVIDIA for embedded robotics and computer vision tasks. It integrates specialized vision backbones designed for low-power edge accelerators.

Specifications

Attention

Attention Structure

Grouped-Query Attention

Attention Heads

16

Key-Value Heads

8

Attention Head Dimension

128

Position Embedding

ROPE

RoPE Theta

100,000,000

Normalization

RMS Normalization

Activation Function

ReLU2

Dimensions

Base Precision

FP16

Hidden Dimension Size

2,048

Number of Layers

28

FFN Intermediate Size (Dense)

9,216

Tokenizer

Vocabulary Size

135,168

About Nemotron 3.5

NVIDIA's Nemotron 3.5 family of open models delivers leading efficiency and accuracy for agentic AI applications. Built on an advanced hybrid Mamba-2 SSM and sparse Mixture-of-Experts architecture with native Multi-Token Prediction (MTP) and speculative decoding harnesses, Nemotron 3.5 models are purpose-built for the high-volume execution layer of always-on agents.


Other Nemotron 3.5 Models