PoSTMEDIA/Xin-V1

Xin-V1

Xin-V1 is a Korean-enhanced, instruction-tuned LLM built on top of `Qwen/Qwen3.8-27B` by PoSTMEDIA AI Lab — the first model of the Xin line.

It is produced with PoSTMEDIA's in-house Capability-Preserving Full Fine-Tuning research (the same second-generation methodology behind the Lux-V2 family) — a training methodology designed so that deep domain adaptation does not erode the reasoning, instruction-following, and multilingual abilities of the base model. Xin-V1 is tuned as a non-thinking (direct answer) product: it responds immediately without emitting reasoning traces.

Highlights

• —Competition math significantly improved — AIME 2024/2025/2026 and HMMT 2025 average up strongly over the base model, measured across 3 sampling seeds
• —Base capability preserved — general knowledge (MMLU/MMLU-Pro), GPQA, and instruction following stay at or near base level
• —Korean-first tuning — trained on PoSTMEDIA's in-house verified Korean synthetic datasets
• —Non-thinking product — fast direct answers; no <think> traces
• —Hybrid-attention 27B — 16 full-attention + 48 linear-attention layers, efficient long-context serving

Model Overview

Benchmarks vs Base

All results measured in-house under a single unified protocol (identical prompts, official non-thinking sampling — temperature 0.7, top-p 0.8, top-k 20, presence penalty 1.5 — and identical generation budgets). Competition-math rows (AIME/HMMT) are the mean of 3 sampling seeds for both models to suppress small-sample noise.

† Competition-math rows are the mean of 3 sampling seeds for both models. ‡ Judge-scored short-answer QA; Xin-V1's more direct answer style contributes to this gain (exact-match on the same set is comparable to base).

Training Data

Xin-V1 is trained on PoSTMEDIA's in-house Korean synthetic data assets, generated and quality-controlled by our internal data factory:

• —General conversation — natural Korean multi-topic dialogue
• —Coding — execution-verified code generation and explanation
• —Instruction following — rule-verifiable Korean constraint-following tasks
• —Mathematics — symbolically verified Korean math reasoning
• —PoSTMEDIA identity — hand-curated identity dataset

Correctness of the synthetic data is enforced by verification gates (code execution, symbolic math equivalence, rule checkers) rather than by model self-judgment.

The exact training procedure — schedule, module selection, and the post-training consolidation step that preserves base capability — is an internal research method and is not disclosed in detail.

Quick Start

pip install transformers accelerate

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_name = "PoSTMEDIA/Xin-V1"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

messages = [{"role": "user", "content": "한국의 전통 명절 세 가지를 소개해줘."}]
inputs = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True, enable_thinking=False,
    return_tensors="pt",
).to(model.device)
outputs = model.generate(
    inputs, max_new_tokens=512,
    do_sample=True, temperature=0.7, top_p=0.8, top_k=20,
)
print(tokenizer.decode(outputs[0][inputs.shape[1]:], skip_special_tokens=True))

Note: Xin-V1 is tuned for non-thinking use. Pass enable_thinking=False to apply_chat_template (as above) and use the official non-thinking sampling parameters (temperature 0.7, top-p 0.8, top-k 20, presence penalty 1.5).

Use Cases

• —Korean-first assistants requiring fast, direct answers
• —Mathematical and coding assistance in Korean and English
• —General Korean instruction-following workloads

Safety & Limitations

• —The model can generate incorrect or outdated information; verify high-stakes outputs.
• —Inherits the general limitations and usage considerations of the Qwen3.8 base model.

Citation

@misc{xin2026,
  title  = {Xin-V1: Capability-Preserving Korean Fine-Tuning of Qwen3.8},
  author = {{PoSTMEDIA AI Lab}},
  year   = {2026},
  url    = {https://huggingface.co/PoSTMEDIA/Xin-V1}
}

Contact

Questions and feedback — please open a discussion on the model page.