Keural-SFT2-14.83B (SFT Epoch 2 — 29,112 steps)

Keural is a bilingual Korean–English Mixture-of-Experts language model trained entirely from scratch — no base model was used. This is the SFT epoch 2 checkpoint at step 29,112, continuing supervised fine-tuning from the epoch 1 checkpoint (18,000 steps) on the same 710K ChatML dataset.

Model Details

Property	Value
Architecture	Mixtral-style MoE (8 experts, top-2 routing)
Parameters	14.83B total / ~7.42B active per token
Layers	24
Hidden size	4096
Attention heads	32 (GQA — 8 KV heads)
Head dim	128
Expert intermediate size	5,632
Experts	8 total, top-2 per token
Context length	4,096 tokens
Vocabulary	131,074 (131,072 SPM + `<
RoPE theta	500,000
Sliding window	512 (alternating every other layer)
Norm	RMSNorm (eps=1e-5)
Activation	SiLU
Dtype	bfloat16
Languages	Korean (primary), English

Full Training Pipeline

Stage	Steps	Tokens	Data	Hardware
Pretraining Stage 1	100,000	~50B	Korean + English web corpus	2× H200 SXM
Pretraining Stage 2	120,000	~13B	Korean + English web corpus (continued)	2× H200 SXM
SFT Epoch 1	18,000	710M	mkd-chanwoo/keural-SFT (1.14M ChatML samples)	2× H200 SXM
DPO (1 full epoch)	6,927	—	keural-dpo-raw (440K preference pairs)	2× H200 SXM
SFT Epoch 2 (this checkpoint)	29,112	7.63B	mkd-chanwoo/keural-SFT (710K ChatML samples, 2nd pass)	2× H200 SXM

SFT Epoch 2 Training Details

Hyperparameter	Value
Resumed from	checkpoint_18000 (SFT epoch 1 final)
Learning rate	1e-5 → 1e-6 cosine decay
Min learning rate	1e-6
Effective batch size	64 (4 per GPU × 8 grad accum × 2 GPUs)
Max sequence length	4,096 tokens
Weight decay	0.05
Gradient clipping	1.0
Optimizer	AdamW
Total steps	29,112
Dataset	mkd-chanwoo/keural-SFT (710K samples)
Total tokens	~7.63B
Training time	~56.71h
Parallelism	FSDP FULL_SHARD (ZeRO-3 equivalent)
Precision	bfloat16 + gradient checkpointing
Hardware	2× NVIDIA H200 SXM (139 GiB each)

SFT Dataset

Source	Samples	Language
mkd-chanwoo/keural-SFT	710,000	Korean + English

Chat Format (ChatML)

This model uses ChatML format. Always include a system prompt for best results.

<|im_start|>system
You are a helpful bilingual Korean-English assistant. Always respond in the same language as the user.<|im_end|>
<|im_start|>user
안녕하세요! 오늘 날씨가 어때요?<|im_end|>
<|im_start|>assistant

The model generates until it produces <|im_end|> (token ID 131073).

The chat template in tokenizer_config.json automatically injects a default system prompt if you don't provide one, so bilingual behavior works out of the box with apply_chat_template.

How to Use

With transformers

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "mkd-hossain/keural-sft2"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

messages = [
    {
        "role": "system",
        "content": (
            "You are a helpful bilingual Korean-English assistant. "
            "Always respond in the same language as the user's message."
        )
    },
    {"role": "user", "content": "파이썬에서 리스트를 정렬하는 방법을 알려주세요."},
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)

with torch.no_grad():
    output = model.generate(
        **inputs,
        max_new_tokens=512,
        temperature=0.7,
        top_p=0.9,
        top_k=50,
        repetition_penalty=1.1,
        no_repeat_ngram_size=8,
        do_sample=True,
        eos_token_id=131073,
    )

response = tokenizer.decode(output[0][inputs.input_ids.shape[1]:], skip_special_tokens=False)
response = response.split("<|im_end|>")[0].strip()
print(response)

With vLLM (recommended for serving)

pip install vllm

python -m vllm.entrypoints.openai.api_server \
    --model mkd-hossain/keural-sft2 \
    --tokenizer mkd-hossain/keural-sft2 \
    --dtype bfloat16 \
    --max-model-len 4096 \
    --tensor-parallel-size 1

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="none")

response = client.chat.completions.create(
    model="mkd-hossain/keural-sft2",
    messages=[
        {"role": "system", "content": "You are a helpful bilingual assistant. Respond in the same language as the user."},
        {"role": "user", "content": "What is the capital of South Korea?"},
    ],
    max_tokens=512,
    temperature=0.7,
)
print(response.choices[0].message.content)

Special Tokens

Token	ID	Purpose
`<	im_start	>`
`<	im_end	>`
<bos>	1	Beginning of sequence
<eos>	2	End of sequence (not used for chat)
<pad>	0	Padding token

Critical: Always set eos_token_id=131073 when generating. Do not use eos_token_id=2.

Recommended Generation Settings

# Conversational / creative
{
    "max_new_tokens": 512,
    "temperature": 0.7,
    "top_p": 0.9,
    "top_k": 50,
    "repetition_penalty": 1.1,
    "no_repeat_ngram_size": 8,
    "do_sample": True,
    "eos_token_id": 131073,
}

# Factual / deterministic
{
    "max_new_tokens": 512,
    "temperature": 0.1,
    "repetition_penalty": 1.1,
    "do_sample": False,
    "eos_token_id": 131073,
}

Checkpoint Comparison

Checkpoint	Stage	Steps	Notes
mkd-hossain/keural-pretrained	Pretraining	120,000	Raw base, no instruction tuning
mkd-hossain/keural-sft-18k	SFT Epoch 1	18,000	Instruction following, ChatML format
mkd-hossain/keural-dpo-3500	DPO 50%	3,500	Early alignment
mkd-hossain/keural-dpo-5500	DPO 79%	5,500	Late alignment
mkd-hossain/keural-dpo-final	DPO 100%	6,927	Full epoch DPO
mkd-hossain/keural-sft2	SFT Epoch 2	29,112	Continued SFT on 710K dataset

Limitations

• Maximum context is 4,096 tokens.
• The pretraining corpus is Korean-dominant — always include a system prompt for correct bilingual behavior.
• Not safety-aligned — do not deploy in production without additional safety fine-tuning.
• This is an intermediate checkpoint. SFT epoch 3 on a 2.35M sample merged dataset is in progress.

License

Apache 2.0