Qwen2.5-7B-qat-d2-6k/README.md

license, base_model, tags, library_name

license	base_model	tags	library_name
apache-2.0	Qwen/Qwen2.5-7B	quantization ternary balanced-ternary tritllm	transformers

Qwen2.5-7B-qat-d2-6k

Balanced ternary quantization of Qwen/Qwen2.5-7B at depth d=2 (9 levels per weight, 3.47 bits per weight) via block-wise QAT (6K WikiText samples, 500 steps/block).

Produced with the codec from "Balanced Ternary Post-Training Quantization for Large Language Models" (Stentzel, 2026). See the Entrit/tritllm-codec repository for the codec source and the paper for full evaluation results.

Quick load

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("Entrit/Qwen2.5-7B-qat-d2-6k")
tokenizer = AutoTokenizer.from_pretrained("Entrit/Qwen2.5-7B-qat-d2-6k")

The weights are dequantized to FP16 for stock-transformers compatibility. The on-disk size is therefore the same as the FP16 source. The 3.47-bpw figure refers to the information content of the quantized matrices and is what matters for inference on hardware that consumes the packed trit format directly (see Entrit/tritllm-kernel).

Quantization details

Field	Value
Source model	Qwen/Qwen2.5-7B
Depth	d=2 (9 levels)
Bits per weight	3.47
Group size	16
Scale codebook	27-entry log-spaced (scale_depth=3)
Method	block-wise QAT (6K WikiText samples, 500 steps/block)
Quantized layers	all 2D linear matrices
Kept FP16	lm_head, token embeddings, all *_norm layers
Codec	tritllm v2

What's quantized vs kept FP16

Following standard quantization-paper convention, only the 2D linear weight matrices are ternary-quantized. The token embedding lookup and final classifier (lm_head) stay in FP16. The 3.47-bpw figure is computed over the quantized matrices only, consistent with how GPTQ, AWQ, and NF4 report BPW.

Citation

@article{stentzel2026ternaryptq,
  title  = {Balanced Ternary Post-Training Quantization for Large Language Models},
  author = {Stentzel, Eric},
  year   = 2026,
  note   = {Entrit Systems}
}

Reproducibility

git clone https://huggingface.co/Entrit/tritllm-codec
cd tritllm-codec
python quantize_model_v2.py --model Qwen/Qwen2.5-7B --configs uniform-d2 --out ./out

The output at ./out/uniform-d2/model/ matches this repository.