Instructions to use ISTA-DASLab/Llama-3.1-70B-Instruct-3Bit-GSQ with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ISTA-DASLab/Llama-3.1-70B-Instruct-3Bit-GSQ with Transformers:

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

pipe = pipeline("text-generation", model="ISTA-DASLab/Llama-3.1-70B-Instruct-3Bit-GSQ")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("ISTA-DASLab/Llama-3.1-70B-Instruct-3Bit-GSQ")
model = AutoModelForCausalLM.from_pretrained("ISTA-DASLab/Llama-3.1-70B-Instruct-3Bit-GSQ")
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]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use ISTA-DASLab/Llama-3.1-70B-Instruct-3Bit-GSQ with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ISTA-DASLab/Llama-3.1-70B-Instruct-3Bit-GSQ"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ISTA-DASLab/Llama-3.1-70B-Instruct-3Bit-GSQ",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ISTA-DASLab/Llama-3.1-70B-Instruct-3Bit-GSQ

SGLang

How to use ISTA-DASLab/Llama-3.1-70B-Instruct-3Bit-GSQ 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 "ISTA-DASLab/Llama-3.1-70B-Instruct-3Bit-GSQ" \
    --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": "ISTA-DASLab/Llama-3.1-70B-Instruct-3Bit-GSQ",
		"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 "ISTA-DASLab/Llama-3.1-70B-Instruct-3Bit-GSQ" \
        --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": "ISTA-DASLab/Llama-3.1-70B-Instruct-3Bit-GSQ",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use ISTA-DASLab/Llama-3.1-70B-Instruct-3Bit-GSQ with Docker Model Runner:
```
docker model run hf.co/ISTA-DASLab/Llama-3.1-70B-Instruct-3Bit-GSQ
```

Llama-3.1-70B-Instruct — 3-bit GSQ

3-bit quantization of meta-llama/Llama-3.1-70B-Instruct produced with GSQ (Gumbel-Softmax Quantization). The 3-bit setting recovers near-FP zero-shot accuracy while still using a symmetric group-wise scalar format that drops into existing INT inference kernels.

Paper: GSQ: Highly-Accurate Low-Precision Scalar Quantization for LLMs via Gumbel-Softmax Sampling (arXiv:2604.18556)
Paper page on HF: https://huggingface.co/papers/2604.18556
Code: https://github.com/IST-DASLab/GSQ
Collection: https://huggingface.co/collections/ISTA-DASLab/gsq

Quantization details

Base model: meta-llama/Llama-3.1-70B-Instruct
Bits / weight (effective): ≈3 bpp
Codebook: 5-level integer (init + {-2, -1, 0, 1, +2}) × scale
Group size: 128
Format: Humming (quant_method: "humming", b_dtype: "uint3")
Pipeline: GPTQ initialization → Gumbel-Softmax refinement (Lion optimizer)

Storage layout (why the HF UI shows I32 + BF16)

The Hugging Face "Tensor types" widget reports the container dtype of each safetensor on disk, not the effective precision of the underlying weights. This checkpoint uses the Humming on-disk layout (exact-width packing — no sub-byte values are padded into a wider container). For every quantized Linear layer with original weight shape [out_features, in_features], the following tensors are stored:

Tensor	Dtype	Shape on disk	Meaning
`<layer>.weight`	I32	`[out_features, in_features × 3 / 32]`	3-bit values bit-packed along the input dim, LSB-first: 32 weights pack into exactly 3 INT32 words (≈10.67 weights per word).
`<layer>.weight_scale`	BF16	`[out_features, in_features / 128]`	One symmetric scale per group of `group_size = 128` weights along the input dim.
Attention / norms / embed / LM-head	BF16	unchanged	Not quantized; copied from the base checkpoint.

Example (model.layers.0.mlp.gate_proj, original [28672, 8192]): weight = [28672, 768] I32 (since 8192 × 3 / 32 = 768), weight_scale = [28672, 64] BF16 (since 8192 / 128 = 64).

So although the UI says "I32 + BF16", the effective storage per quantized weight is 3 bits (packed) + 16 bits / 128 (group scale) ≈ 3.13 bpp. The quantization_config block in config.json is:

{
  "quant_method": "humming",
  "b_dtype": "uint3",
  "weight_scale_group_size": 128,
  "weight_scale_type": "group",
  "has_zero_point": false,
  "ignore": ["lm_head", "embed_tokens"]
}

Loading this checkpoint requires vLLM plus the humming kernels (pip install humming-kernels). See Serving with vLLM below.

Note: GSQ training first writes shards in compressed-tensors pack-quantized format (where a 3-bit codebook is padded into a 4-bit INT32 container). The published checkpoint here has been re-packed via convert_to_humming.py into exact-width 3-bit Humming storage, which is why the on-disk size is ≈3 bpp rather than the ≈4 bpp you'd see if the intermediate CT format had been uploaded directly.

Serving with vLLM

Install the Humming kernels (required for vLLM to load this checkpoint):

pip install humming-kernels

vllm serve ISTA-DASLab/Llama-3.1-70B-Instruct-3Bit-GSQ \
  --tensor-parallel-size 2

Citation

@article{gsq2026,
  title  = {GSQ: Highly-Accurate Low-Precision Scalar Quantization for LLMs via Gumbel-Softmax Sampling},
  author = {Dadgarnia, Alireza and Tabesh, Soroush and Nikdan, Mahdi and Helcig, Michael and Kurti{\'c}, Eldar and Kleinegger, Max and Alistarh, Dan},
  journal= {arXiv preprint arXiv:2604.18556},
  year   = {2026},
  url    = {https://arxiv.org/abs/2604.18556}
}