ThakiCloud

Qwen3.8-27B-Human-KO

README

License: apache-2.0

Measurements (2026-09-01, temperature and conditions stated)

Table with columns: Axis, base, This model
Axis	base	This model
Human-likeness pairwise (LLM judge, n=175)	1.1% win	94.9% win (tie 4.0%, position-swap consistency 96%)
Free-generation bullet-list rate (n=200)	97.5%	2.0%
CJK contamination, T=0, n=3,369	2.55%	0.33% (true error rate 0.24%)
HumanEval (paired n=58)	93.85%	+3.03pp
Instruction following	81%	−3.0pp (statistically equivalent)
KMMLU (n=300)	55.67%	−2.0pp — minimum detectable difference is 11.4pp, so no verdict is possible

Honest caveats:

The pairwise judge (Qwen3.8-27B, T=0, position-swapped) has not completed human-label calibration (P1.5). The figures above are "win rate on the judge's axis," not human evaluation itself.
The KMMLU row is not "no regression" — it is "a regression on the order of −1pp cannot be detected with this sample."
The floor for CJK contamination is not 0. A rare Hanja character with no standalone token in the vocabulary can be assembled via byte fallback (the measured residual 0.24% is mostly the single-character class kept for gloss preservation).

Usage

python

from vllm import LLMllm = LLM(model="ThakiCloud/Qwen3.8-27B-Human-KO", dtype="bfloat16", trust_remote_code=True)

The tokenizer, vocabulary size, and architecture are identical to base (standard load). The model can read Hanja but cannot generate Chinese-word tokens.

Limitations

The default tendency is toward short answers. If you need a long, structured document, specify length and format in the prompt.
The measurements above center on the Korean conversation axis. The English and math axes have not been measured with a sufficient sample.
The safety axis inherits base's alignment as-is (no separate safety training).

License

Apache-2.0 (base: Qwen/Qwen3.8-27B). Distributed together with LICENSE, NOTICE, and DATA_PROVENANCE.md.

Paper

Off-Target Effects of Response-Style Alignment in a Korean 27B Language Model — arXiv:2609.11291

The style alignment behind this line also moved two behaviors nobody trained for: abstention on ambiguous social questions (KoBBQ) and unprompted disclosure in securities guidance. Both moved through the emission policy — how often the model answers and how much it says — rather than through what it says when it does answer. Holding prompts, recipe, data volume and serving fixed and changing only the training target, three style seeds moved answer rate one way and three neutral seeds moved it the other (observed ranges do not overlap). Read this card's numbers with that in mind.

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Model Details

Model Provider

ThakiCloud

Model Tree

Base

Qwen/Qwen3.8-27B

Fine-tuned

this model

Input Modalities

TextImageVideo

Output Modalities

Text

Supported Functionality