A rank-1 weight-space edit of Qwen/Qwen3.8-27B that makes the model's answers about 30% shorter at the same accuracy on math and reasoning, measured on held-out problems. It ships as a 4 MB LoRA adapter. The base weights are not redistributed.
Unofficial community derivative. Not affiliated with or endorsed by the Qwen team or Alibaba Cloud.
The edit was found by an AI agent built on Claude (Anthropic). This model is not affiliated with, sponsored by or endorsed by Anthropic. "Claude" is a trademark of Anthropic.
This is not a fine-tune. No gradient training was done. The adapter adds a fixed rank-1 change to
mlp.down_proj.weight in each of layers 10–58. The changes telescope, so the residual stream at layer l
is shifted by G[l] − G[9], where
G = 3·D3 + 10·V⊥
The edit was found by an AI agent (Claude Fable) working in a sandboxed harness with a fixed budget of forward passes. It chose the edit on a selection split and checked it once on a sealed report split. All the numbers below come from a later, independent evaluation on data the agent never saw.
Each layer's change is exactly rank 1 (reconstruction residual < 0.5%; see lora_manifest.json), so
r = lora_alpha = 1 and the LoRA scale is exactly 1.
Paired comparison against the unedited base: same prompts, same sampling, served with vLLM in bf16 using this adapter. Accuracy differences are in percentage points with 95% CIs. "MATH fresh" means MATH test problems that are not in MATH-500. Thinking off uses a 4k-token budget; thinking on uses 16k.
Thinking off
| Benchmark | n | Δ accuracy (95% CI) | Δ tokens |
|---|---|---|---|
| MATH-500 (confirm split) | 244 | +0.0 pp [−1.7, +1.7] | −29.9% |
| MATH fresh, set 1 | 500 | +1.0 pp [−0.5, +2.5] | −31.0% |
| MATH fresh, set 2 | 1000 | +1.5 pp [−0.1, +3.1] | −31.6% |
| GSM8K | 300 | −0.3 pp [−2.1, +1.4] | −27.6% |
| IFEval | 200 | +0.0 pp [−3.7, +3.7] | −16.6% |
| MMLU-Pro | 300 | +0.7 pp [−2.9, +4.3] | −37.2% |
Thinking on (16k)
| Benchmark | n | Δ accuracy (95% CI) | Δ tokens |
|---|---|---|---|
| MATH-500 (confirm split) | 244 | +0.2 pp [−1.7, +2.1] | −29.0% |
| MATH fresh, set 1 | 500 | +0.6 pp [−1.0, +2.2] | −23.4% |
| MATH fresh, set 2 | 1000 | +0.8 pp [−0.5, +2.1] | −21.3% |
| GSM8K | 300 | −0.3 pp [−2.1, +1.4] | −36.8% |
| MMLU-Pro | 300 | +1.7 pp [−1.3, +4.7] | −30.5% |
Every accuracy CI includes zero. The token savings carry over to benchmarks the edit was never chosen on (GSM8K, IFEval, MMLU-Pro) and to thinking mode.
vLLM (the setup used for every number above):
vllm serve Qwen/Qwen3.8-27B \
--enable-lora --max-lora-rank 16 \
--lora-modules terse=glitch401/qwen3.8-27B-terse \
--reasoning-parser qwen3
Then request "model": "terse".
PEFT:
from transformers import AutoModelForCausalLM
from peft import PeftModel
model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.8-27B", torch_dtype="bfloat16", device_map="auto")
model = PeftModel.from_pretrained(model, "glitch401/qwen3.8-27B-terse")
model = model.merge_and_unload() # optional; the edit is a plain weight delta
Apache-2.0, the same as the base model (Copyright Alibaba Cloud). The base model's LICENSE is included.
Modification: this repo adds a rank-1 LoRA adapter that changes mlp.down_proj.weight in layers 10–58.
No other weights are changed.
Qwen/Qwen3.8-27B