Instructions to use xv0y5ncu/SmolLM3-3B-GLQ-6bpw with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use xv0y5ncu/SmolLM3-3B-GLQ-6bpw with Transformers:

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

pipe = pipeline("text-generation", model="xv0y5ncu/SmolLM3-3B-GLQ-6bpw")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("xv0y5ncu/SmolLM3-3B-GLQ-6bpw")
model = AutoModelForCausalLM.from_pretrained("xv0y5ncu/SmolLM3-3B-GLQ-6bpw", 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]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use xv0y5ncu/SmolLM3-3B-GLQ-6bpw with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "xv0y5ncu/SmolLM3-3B-GLQ-6bpw"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "xv0y5ncu/SmolLM3-3B-GLQ-6bpw",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/xv0y5ncu/SmolLM3-3B-GLQ-6bpw

SGLang

How to use xv0y5ncu/SmolLM3-3B-GLQ-6bpw 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 "xv0y5ncu/SmolLM3-3B-GLQ-6bpw" \
    --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": "xv0y5ncu/SmolLM3-3B-GLQ-6bpw",
		"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 "xv0y5ncu/SmolLM3-3B-GLQ-6bpw" \
        --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": "xv0y5ncu/SmolLM3-3B-GLQ-6bpw",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use xv0y5ncu/SmolLM3-3B-GLQ-6bpw with Docker Model Runner:
```
docker model run hf.co/xv0y5ncu/SmolLM3-3B-GLQ-6bpw
```

SmolLM3-3B GLQ 6bpw

SmolLM3-3B quantized to 6 bits per weight using GLQ (Golay-Leech Quantization).

Key Features

6bpw -- 3-stage residual vector quantization (E8 lattice codebook)
99.6% of bf16 quality on lm-eval 5-task benchmark
2.5 GB model size (vs 6.1 GB bf16)
Block-diagonal FHT -- zero padding waste, honest bpw labeling

Quality (lm-eval 5-task)

Task	bf16	GLQ 6bpw
arc_easy	0.7908	0.7824
hellaswag	0.5651	0.5618
piqa	0.7845	0.7802
winogrande	0.6685	0.6661
lambada_openai	0.6592	0.6645
Average	0.6936	0.6910 (99.6%)

Usage

pip install glq

import glq.hf_integration
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "xv0y5ncu/SmolLM3-3B-GLQ-6bpw",
    device_map="cuda",
    dtype="float16",
)
tokenizer = AutoTokenizer.from_pretrained("HuggingFaceTB/SmolLM3-3B")

inputs = tokenizer("The capital of France is", return_tensors="pt").to("cuda")
output = model.generate(**inputs, max_new_tokens=64)
print(tokenizer.decode(output[0], skip_special_tokens=True))

Requirements

transformers >= 5.0 (4.x has a weight loading bug that breaks small GLQ models)
torch >= 2.0
glq >= 0.2.8 (pip install glq)

How it works

GLQ uses the E8 lattice codebook (65536 entries) with Randomized Hadamard Transform (RHT) for incoherence processing. At 6bpw, three codebook stages are used:

Primary E8 codebook (16 bits per 8-weight block = 2 bpw)
Secondary E8 codebook on residual (16 bits = 2 bpw)
Tertiary E8 codebook on residual (16 bits = 2 bpw)

Block-diagonal FHT decomposes non-power-of-2 dimensions into sums of powers of 2, eliminating padding waste. For SmolLM3-3B (hidden_size=2048), dimensions are already power-of-2 so there is zero overhead.