Skip to content
Log in

meta-llama/Llama-3.2-3B-Instruct

meta-llama/Llama-3.2-3B-Instruct architecture graph
meta-llama/Llama-3.2-3B-Instruct architecture graph
  • huggingface
  • AGX Thor
Labels
  • Text
  • LLM
  • Generation
Build
Embedl-optimized
Precision
W4A16
Runtime
TensorRT Edge-LLM 0.10.1 with Embedl optimizations
Context length
4,096 tokens
Run Llama 3.2 3B on AGX Thor

Meta's 3B Llama 3.2 instruction model for summarizing, rewriting and simple agents.

Methodology

  • Throughput: one user, real conversations from the public OpenAssistant set, 2,048 tokens in, 128 out, greedy, no prompt reuse. Average of 20 requests after 3 warm-up requests. End-to-end is the whole request. Prompt processing is the 2,048 prompt tokens over the time to first token.
  • Power and memory: sampled on the device during those runs. Idle is the device at rest with the model loaded; energy per token is the run's energy over its generated tokens; peak memory is the device's.
  • Accuracy: full test sets, greedy, the same model files and settings as served. GSM8K 1319, IFEval 541, MMLU-Pro 1008 questions.
  • Baseline TensorRT Edge-LLM: the original FP16 model on upstream TensorRT Edge-LLM, same settings, same device.

Files

Image Size

Model

docker.io/embedl/model-artifacts:llama3-2-3b.huggingface_int4_embedl-flashhead_tensorrt-edgellm.fa6968d9848847ef

 2.92 GiB

Runtime

docker.io/embedl/models:runtimes.tensorrt-edgellm.arm64.2026101815014ee0

 1.01 GiB

Variants

Every build of Llama 3.2 3B on AGX Thor.

Variant Precision Speed Peak memory Energy / token Runtime
Baseline TensorRT Edge-LLM
FP16 35.4 tok/s 8.68 GiB 1.76 J
TensorRT Edge-LLM 0.10.1
Embedl optimized · TensorRT Edge-LLM This page
 W4A16 91.5 tok/s 4.67 GiB 0.75 J
TensorRT Edge-LLM 0.10.1 with Embedl optimizations
runtimes.tensorrt-edgellm.arm64.2026101815014ee0