SII-Decision-4B-Full

SII-Decision-4B-Full belongs to the SII-Decision-4B family of decision models. It returns a probability distribution over user-supplied options with thinking disabled. Choice, binary/probabilistic Noul, and ordered Score tasks share the same answer-slot interface. This training optimizes decision distributions; it does not provide general chat SFT supervision.

Model and training

Item	Setting
Original model	Qwen/Qwen3.5-4B
Adaptation	Full-rank updates of the same language linear layers covered by LoRA
Training questions	29,326
Validation questions	7,506
Teacher	Original same-size model, two reasoning readouts averaged
Student	Thinking disabled; original tokenizer and ordered option labels
Learning rate	2e-06
Global batch	64
Seed	43
Exported step / one-epoch schedule	459 / 459
Updated parameters during training	3,313,500,160
Weight format	Complete BF16 safetensors; no adapter required for inference

Only the listed language linear layers are adapted. Embeddings, the output head, normalization, and vision parameters remain frozen. Full-rank adaptation and LoRA use the same layer coverage. The architecture retains the original vision components, but SII-Decision decision training and the supplied inference interface use text inputs.

The per-question loss is:

w * CE(q, p) + 0.5 * w * I(Score) * ((E[p]-E[q])/(K-1))² + kappa * KL(p0 || p).

Choice/Noul/Score weights are 1 / 1.1666667 / 0.8. Kappa is 0.2 for hard binary Noul, and 0.05 otherwise. Gold labels are used for validation and identifying hard binary tasks; they do not replace teacher targets. See training_recipe.json for the exact settings.

Evaluation

The associated BF16 checkpoint scores 73.4694 / 100 on the fixed 58-benchmark decision suite, with equal weight per benchmark. This is one checkpoint's score, not a multi-seed mean or a confidence bound. Validation metrics determine the checkpoint within its run. The released full-rank weights are copied without changing their tensor values. NF4/INT8 results may differ from the BF16 evaluation and have not been assigned this score.

Load with Transformers

The complete merged weights can be loaded without PEFT or a separate base model. Install torch, accelerate, and transformers==5.14.1.

import torch
from transformers import AutoTokenizer, Qwen3_5ForConditionalGeneration

repo = "jylin001206/SII-Decision-4B-Full"
tokenizer = AutoTokenizer.from_pretrained(repo)
model = Qwen3_5ForConditionalGeneration.from_pretrained(
    repo, dtype=torch.bfloat16, device_map="auto"
)

Loading in BF16 requires sufficient RAM or GPU memory. This example loads the original precision; the RTX 4090 launchers below use the stated deployment precision. Training code and training data are publicly available at the repository links below.

Inference

Use the decision scripts from https://github.com/amazingljy1206/SII-Decision; generic free-form generation does not implement the decision scoring interface.

git clone https://github.com/amazingljy1206/SII-Decision.git
cd SII-Decision
pip install -e '.[quantization]'
bash scripts/infer_4b_single_4090.sh \
  jylin001206/SII-Decision-4B-Full examples/decisions.jsonl predictions.jsonl

The scripts work with either a Hugging Face repository ID or a local model directory. 4B uses BF16 on one RTX 4090; its two-GPU script runs two independent replicas. 27B uses NF4 on one RTX 4090 and INT8 by default on two RTX 4090 GPUs. Two 24-GB cards cannot hold the complete 27B BF16 model; the two-GPU script uses model sharding with quantized weights. Inputs are rejected rather than truncated when they exceed the configured token limit.

Data and reproduction

Training data: jylin001206/SII-Decision-4B-Training-Data. Training code, exact layer lists, four method configs, and merge code: https://github.com/amazingljy1206/SII-Decision. The tokenizer, chat template, configuration, license, and safetensors shards are included here.

License

The model retains the Apache-2.0 license of the original Qwen weights. See LICENSE and NOTICE. Training-data source materials retain their upstream terms, as described in the dataset card.