Olmo-3-7B-Think — single conditioned model (matched budget)

The alternative design: one adapter asked to serve every verbosity level by naming the level in the prompt, instead of one adapter per level. This is the comparison that motivates the per-level family.

Results — GSM8K test (n=1317)

Requested level	Accuracy
L1	90.3%
L2	89.1%
L3	88.5%
L4	88.7%
L5	89.1%

Flat across every requested level — but flat because the model emits roughly the same chain whatever level is asked for. It does not collapse; it also does not compress. Read alongside the per-level adapters, where accuracy and length actually move with the level.

Training


Stage	GRPO on the merged conditioned SFT model
Conditioning	the level is named in the prompt, not selected by adapter
Engine	`trl.GRPOTrainer` on stock `transformers`, `sdpa`
LoRA	r=16, alpha=32
Hardware	1x NVIDIA A100 80GB

Usage

Solve this using Level {N} ({Verbose|Concise|Symbolic|Shorthand|Extreme}).
Problem: {your problem}

Stacks on the conditioned SFT model, not the raw base.

from transformers import AutoModelForCausalLM
from peft import PeftModel

model = AutoModelForCausalLM.from_pretrained("allenai/Olmo-3-7B-Think", torch_dtype="bfloat16", device_map="auto")
model = PeftModel.from_pretrained(model, "ssurface/cot-dialect-olmo3-7b-think-conditioned-sft")
model = model.merge_and_unload()
model = PeftModel.from_pretrained(model, "ssurface/cot-dialect-olmo3-7b-think-conditioned-grpo-matched")