Llama 3.2 3B Instruct

Model Overview

The Llama 3.2 collection of multilingual large language models (LLMs) is a collection of pretrained and instruction-tuned generative models in 1B and 3B sizes (text in/text out). The Llama 3.2 instruction-tuned text only models are optimized for multilingual dialogue use cases, including agentic retrieval and summarization tasks.

• Model Architecture: Llama 3.2 is an auto-regressive language model that uses an optimized transformer architecture. The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety.
• Model Release Date: Sept 25, 2024.
• Model Source: meta-llama/Llama-3.2-3B-Instruct
• License: llama3.2
• Supported languages: English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai

QPC Configurations

Precision	SoCs / Tensor slicing	NSP-Cores (per SoC)	Full Batch Size	Chunking Prompt Length	Context Length (CL)	QPC URL	QPC Size	QPC Download	Onnx URL	Onnx Download	Generation Date
MXFP6	2	16	1	128	8192	https://dc00tk1pxen80.cloudfront.net/SDK1.21.2/meta-llama/Llama-3.2-3B-Instruct/meta-llama_Llama-3.2-3B-Instruct_qpc_16cores_128pl_8192cl_1fbs_2devices_mxfp6_mxint8.tar.gz	5.5GB	Download	https://dc00tk1pxen80.cloudfront.net/SDK1.21.2/meta-llama/Llama-3.2-3B-Instruct/meta-llama_Llama-3.2-3B-Instruct_ONNX.tar.gz	Download	25-Mar-2026

Run This Model

Download QPCs

mkdir -p meta-llama/Llama-3.2-3B-Instruct
cd meta-llama/Llama-3.2-3B-Instruct

# Download QPC
wget <QPC_Download_URL>
tar xzvf <qpc_filename.tar.gz>

Run QPC

Replace QPC_PATH with actual extracted QPC directories.

python3 -m vllm.entrypoints.openai.api_server \
  --port <PORT> \
  --model meta-llama/Llama-3.2-3B-Instruct \
  --device-group <DEVICE_IDS> \
  --max-model-len <CTX_LEN> \
  --max-seq-len-to-capture <PREFILL_SEQ_LEN> \
  --max-num-seqs <MAX_NUM_SEQS> \
  --quantization mxfp6 \
  --kv-cache-dtype mxint8 \
  --override-qaic-config "num_cores:[num_cores] qpc_path:[qpc_path] ccl_enabled:True comp_ctx_lengths_prefill=[1024,2048,4096,8192] comp_ctx_lengths_decode=[1024,2048,4096,8192]"

Run Inference

Once the server is running, send a request to the OpenAI-compatible endpoint:

curl http://localhost:<PORT>/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "meta-llama/Llama-3.2-3B-Instruct",
    "messages": [
      {"role": "user", "content": "<PROMPT>"}
    ],
    "max_tokens": 200
  }'

Useful Links

• Environment Setup