Evaluation
Thinking mode, pass@1 averaged over 1–32 seeds, same settings for all models; MMLU-Pro answers are extracted with the
official MMLU-Pro lenient rules, because this model does not follow the standard answer format.
Usage
Runs on vLLM only (SGLang does not support this 3-bit format yet), on NVIDIA H100, A100, L40S, RTX Pro 6000 and B200 GPUs.
vLLM:
vllm serve qda-lab/QDA-DeepSeek-R1-Distill-Llama-8B-W3A16
How it was made
Weights are 3-bit integers in groups of 128, quantized with GPTQ on a Hadamard-rotated copy of the model; the rotation is folded
into the weights, so there is no runtime transform.
The output head and token embeddings are 4-bit, and the group scales were then fine-tuned by distillation. The recipe was chosen by QDA's automated search.
Reproduce
hf download qda-lab/qda-recipe-inputs --repo-type dataset --include "QDA-DeepSeek-R1-Distill-Llama-8B-W3A16/*" --local-dir inputsqda build inputs/QDA-DeepSeek-R1-Distill-Llama-8B-W3A16/r1-llama-8b-w3a16.yaml
Needs the QDA library (the qda command). Recipe: r1-llama-8b-w3a16.yaml; inputs
(calibration sample and distillation data): QDA-DeepSeek-R1-Distill-Llama-8B-W3A16/ in qda-lab/qda-recipe-inputs.
Takes about 5.2 h on 1× H100 (measured). The rebuild matches this checkpoint's KL to BF16 within 5 %; weights are not bit-identical.
License
MIT, the same as the base model.
Citation
TBD.