Instructions to use donghyunli/Llama-2-13b-KronQ-W3A16-fake with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use donghyunli/Llama-2-13b-KronQ-W3A16-fake with Transformers:

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

pipe = pipeline("text-generation", model="donghyunli/Llama-2-13b-KronQ-W3A16-fake")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("donghyunli/Llama-2-13b-KronQ-W3A16-fake")
model = AutoModelForCausalLM.from_pretrained("donghyunli/Llama-2-13b-KronQ-W3A16-fake", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use donghyunli/Llama-2-13b-KronQ-W3A16-fake with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "donghyunli/Llama-2-13b-KronQ-W3A16-fake"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "donghyunli/Llama-2-13b-KronQ-W3A16-fake",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/donghyunli/Llama-2-13b-KronQ-W3A16-fake

SGLang

How to use donghyunli/Llama-2-13b-KronQ-W3A16-fake 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 "donghyunli/Llama-2-13b-KronQ-W3A16-fake" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "donghyunli/Llama-2-13b-KronQ-W3A16-fake",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "donghyunli/Llama-2-13b-KronQ-W3A16-fake" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "donghyunli/Llama-2-13b-KronQ-W3A16-fake",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use donghyunli/Llama-2-13b-KronQ-W3A16-fake with Docker Model Runner:
```
docker model run hf.co/donghyunli/Llama-2-13b-KronQ-W3A16-fake
```

Llama-2-13b — KronQ W3A16 (fake-quant fp16)

Paper: arXiv:2607.07964 · Code: GitHub

⚠️ Fake-quant fp16 checkpoint. The 3-bit weights are stored in fp16 (KronQ does not pack int3), so this repo is the same size as bf16 — no compression or speedup, for PPL / accuracy reproduction only. For deployable low-bit, see the W4A16 / W2A16 (packed int4/int2) repos.

Llama-2-13b quantized to 3-bit weights with KronQ (Kronecker-factored Hessian quantization), exported as a standard fp16 model.

Results (WikiText-2, seqlen 2048)

Perplexity: 5.18

Zero-shot accuracy:

PIQA	ARC-E	ARC-C	HellaSwag	WinoGrande	BoolQ	OBQA	Average
79.27	73.53	46.50	77.00	71.19	78.87	45.00	67.34

(lm-evaluation-harness, 0-shot. acc_norm for PIQA/HellaSwag/ARC/OBQA, acc for WinoGrande/BoolQ.)

Usage

Loads as a standard fp16 model — no KronQ code needed:

from transformers import AutoModelForCausalLM, AutoTokenizer
m = AutoModelForCausalLM.from_pretrained("donghyunli/Llama-2-13b-KronQ-W3A16-fake", torch_dtype="float16").cuda()
tok = AutoTokenizer.from_pretrained("donghyunli/Llama-2-13b-KronQ-W3A16-fake")

Recipe

Per-channel asymmetric W3, weight-only (a_bits=16), --alpha 0.25, bidirectional incoherence processing (BiIP), act_order, raw H_G. Calibrated on 128 WikiText-2 sequences.

License

Derivative of Llama-2-13b — subject to the Llama 2 Community License.

Downloads last month: 134

Safetensors

Model size

13B params

Tensor type

F16

Model tree for donghyunli/Llama-2-13b-KronQ-W3A16-fake

Base model

meta-llama/Llama-2-13b-hf

Finetuned

(76)

this model

Paper for donghyunli/Llama-2-13b-KronQ-W3A16-fake

KronQ: LLM Quantization via Kronecker-Factored Hessian

Paper • 2607.07964 • Published Jul 8 • 33