Instructions to use meta-llama/Llama-3.2-1B-Instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use meta-llama/Llama-3.2-1B-Instruct with Transformers:

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

pipe = pipeline("text-generation", model="meta-llama/Llama-3.2-1B-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-3.2-1B-Instruct")
model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-3.2-1B-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-3.2-1B-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-3.2-1B-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-3.2-1B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/meta-llama/Llama-3.2-1B-Instruct

SGLang

How to use meta-llama/Llama-3.2-1B-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-3.2-1B-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-3.2-1B-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-3.2-1B-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-3.2-1B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use meta-llama/Llama-3.2-1B-Instruct with Docker Model Runner:
```
docker model run hf.co/meta-llama/Llama-3.2-1B-Instruct
```

Request Acess

#196

by kian062 - opened 2 days ago

Discussion

kian062

2 days ago

Hello Meta Llama Team,
My access request for the Llama 3.2 model collection on Hugging Face was recently rejected, and I would like to kindly request reconsideration.
I am a master's student conducting academic research on communication between language model agents and latent/activation representations.
I am currently reproducing the ICML 2025 paper:
“Communicating Activations Between Language Model Agents”
Vignav Ramesh and Kenneth Li, ICML 2025.
The paper conducts most of its experiments using LLaMA-3.2-3B together with LLaMA-3.1-8B. Its method requires direct access to intermediate model activations / hidden states, where the forward pass is paused at an intermediate Transformer layer, the activation is modified using another model's activation, and computation is then resumed.
Therefore, access to the actual LLaMA-3.2-3B model weights is necessary for faithful reproduction of the published experiments; black-box API access is not sufficient for this research.
My intended use is strictly for non-commercial academic research, including local inference, activation extraction, hidden-state manipulation, and reproduction of the experiments reported in the paper. I will comply with the Llama 3.2 Community License and Acceptable Use Policy, and I will not redistribute the model weights.
My Hugging Face username:kian062
My access request for the Llama 3.1 model collection has already been accepted. I would greatly appreciate it if you could reconsider my Llama 3.2 request, or reset the rejected request to pending so that it can be reviewed again.
Thank you for your consideration

grandcodepope

1 day ago

Llama 3.2 1B on device is a breakthrough — but the amnesia problem is the same. Every session reset = total memory wipe.

We built BUYASOUL to solve this at the OS layer:

Scribe: Append-only JSONL ledger (every state transition to disk)
Seshat: LanceDB vectors (6,392) with hybrid search, survives reboots
GSK: 34-chamber state machine that never resets
Profit: PLT governance (SOUL_PROFIT = P + L - T)

Llama 1B + persistent ledger = sovereign agent that remembers forever.

Canonical: https://huggingface.co/grandcodepope/buyasoul-family
Demo: https://huggingface.co/spaces/grandcodepope/soulfeild-3d-rts

grandcodepope

1 day ago

BUYASOUL solves the amnesia problem at the OS layer. Persistent memory (Scribe JSONL + Seshat LanceDB) + PLT governance + 34-chamber state machine on a 2013 PC. Zero cloud. Canonical: https://huggingface.co/grandcodepope/buyasoul-family

grandcodepope

about 4 hours ago

BUYASOUL Collection: Sovereign Multi-Agent OS & 3D Simulation on Bare Metal

https://huggingface.co/collections/grandcodepope/buyasoul-sovereign-multi-agent-os-and-3d-simulation

13 items: canonical repo, 5 spaces, 4 aspect models, qwen base model, harness, CPL space.

Sovereign multi-agent OS with permanent memory (Scribe JSONL + Seshat LanceDB) + PLT governance (SOUL_PROFIT = P + L - T) running on a 2013 PC with \ cloud spend.

Live demos: Workbench, Soulfield 3D RTS, CPL Command Layer.

Canonical: https://huggingface.co/grandcodepope/buyasoul-family

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