Instructions to use meta-llama/Llama-3.2-3B-Instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use meta-llama/Llama-3.2-3B-Instruct with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="meta-llama/Llama-3.2-3B-Instruct")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.2-3B-Instruct")
model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-3.2-3B-Instruct", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Inference
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use meta-llama/Llama-3.2-3B-Instruct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "meta-llama/Llama-3.2-3B-Instruct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "meta-llama/Llama-3.2-3B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/meta-llama/Llama-3.2-3B-Instruct

SGLang

How to use meta-llama/Llama-3.2-3B-Instruct with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "meta-llama/Llama-3.2-3B-Instruct" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "meta-llama/Llama-3.2-3B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "meta-llama/Llama-3.2-3B-Instruct" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "meta-llama/Llama-3.2-3B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use meta-llama/Llama-3.2-3B-Instruct with Docker Model Runner:
```
docker model run hf.co/meta-llama/Llama-3.2-3B-Instruct
```

Cross-architecture RYS sweep — Llama-3.2-3B-Instruct (biggest math lift in corpus +14.32; family spans entire baseline curve)

#282

by john-broadway - opened May 13

Discussion

john-broadway

May 13

Sharing a cross-architecture RYS (layer-duplication, "Repeat Your Self") sweep that includes Llama-3.2-3B-Instruct alongside 20 other model variants spanning 10 architecture families.

Sweep result for this model (28 layers, Q4_K_M, baseline reasoning 88.24%):

Configuration	Math Δ	EQ Δ	Reasoning Δ
Best: (21,24) block-3	+14.32	−0.27	−5.88

Peak reasoning Δ: +0.00% (near-ceiling baseline leaves no headroom). Peak math Δ: +14.32% — the biggest math lift in the entire v2 corpus. Late-layer block duplication at (21,24) targets a math circuit specifically.

Llama-3.2 family span — together with sibling Llama-3.2-1B-Instruct (baseline reasoning 0.00%, peak Δ +76.47%, the most-dramatic result in the corpus), the Llama-3.2 family alone spans the entire baseline-vs-magnitude curve.

The cross-architecture finding (Pearson r = −0.726 across 21 variants, 10 families): weak baselines lift more, in their weakest dimension. Llama-3.2-3B is at the high-baseline ceiling end; the math lift is what surfaces when reasoning has nowhere to go.

Full sweep data + analysis: https://huggingface.co/datasets/john-broadway/rys-sovereign-collection-v2
Evaluation card for Llama-3.2-3B-Instruct: https://huggingface.co/john-broadway/Llama-3.2-3B-RYS-eval

Method: original RYS post by David Ng; sweep toolkit by alainnothere. Train-free — no weight changes, no merging.

— John Broadway, with collaboration from Claude (Opus 4.6 in April 2026 sweep generation; Opus 4.7 in May 2026 cross-architecture analysis).

john-broadway

May 13

Update (2026-05-13 PM): The eval-only john-broadway/Llama-3.2-3B-RYS-eval repo linked in the original post has been consolidated. The same sweep results + mechanism writeup are now in the deployable weights repo: john-broadway/Llama-3.2-3B-RYS-21-24-GGUF — RYS-applied Q4_K_M GGUF, ready for llama-server. No new content, just one repo per model instead of two.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment