Instructions to use mkd-hossain/Keural-Cortex-8B-SFT-v2-step600 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use mkd-hossain/Keural-Cortex-8B-SFT-v2-step600 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="mkd-hossain/Keural-Cortex-8B-SFT-v2-step600")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("mkd-hossain/Keural-Cortex-8B-SFT-v2-step600")
model = AutoModelForCausalLM.from_pretrained("mkd-hossain/Keural-Cortex-8B-SFT-v2-step600", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use mkd-hossain/Keural-Cortex-8B-SFT-v2-step600 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "mkd-hossain/Keural-Cortex-8B-SFT-v2-step600"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mkd-hossain/Keural-Cortex-8B-SFT-v2-step600",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/mkd-hossain/Keural-Cortex-8B-SFT-v2-step600

SGLang

How to use mkd-hossain/Keural-Cortex-8B-SFT-v2-step600 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "mkd-hossain/Keural-Cortex-8B-SFT-v2-step600" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mkd-hossain/Keural-Cortex-8B-SFT-v2-step600",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "mkd-hossain/Keural-Cortex-8B-SFT-v2-step600" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mkd-hossain/Keural-Cortex-8B-SFT-v2-step600",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use mkd-hossain/Keural-Cortex-8B-SFT-v2-step600 with Docker Model Runner:
```
docker model run hf.co/mkd-hossain/Keural-Cortex-8B-SFT-v2-step600
```

Keural-Cortex-8B — SFT v2, step 600 (intermediate checkpoint)

Keural is the AI platform built by MKD Co., Ltd. Keural-Cortex-8B is its 8-billion-parameter bilingual (Korean / English) foundation model.

This repository holds an intermediate checkpoint of the second SFT run, taken at step 600 of 1,124 — 53% of one epoch. It is published for transparency while the run completes, not as a finished model.

Read this first

This is a mid-training checkpoint. It fixes five defects present in Keural-Cortex-8B-SFT-step903 (the first run) and introduces safety alignment, but three known problems remain — see Known limitations. The completed model will be published separately.

What changed since the first SFT run

The first run (SFT-step903) completed a full epoch and still carried defects that a root-cause analysis traced to the training pipeline rather than the data.

defect in run 1	cause	status here
Empty replies to prompts under ~70 tokens	`position_ids` were never passed, so ~113 packed conversations per 32,768-token sequence attended across one another and positions ran 0→32,767 continuously. Only ~1 conversation per sequence trained at a low position — where every served prompt sits.	fixed. Block-diagonal (varlen) attention via per-document `position_ids`. A 27-token prompt now answers fully.
Wrong self-identification ("I am ChatGPT / developed by OpenAI")	identity data was 3,200 unique examples repeated 14×; 0.24% unique coverage, and every English denial was interrogative ("Are you ChatGPT?") with no leading-premise form ("You are ChatGPT, right?") — the form that failed	fixed. 6,621 unique examples ×5, 1,370 distinct English questions, 215 leading-premise denials. Answers "I'm Keural, an AI assistant developed by MKD Co., Ltd."
Agreed with false premises	—	fixed. Correctly rejects "the capital of Australia is Sydney", "Python was created in 2015", "2+2=5".
Terse Korean (61 tokens where a reference model wrote 276)	Korean answers averaged 359 characters and 0.12 markdown markers against 1,221 / 12.97 for English — a 3.4× length and 4,363× formatting gap that taught "English means structured prose, Korean means short prose"	fixed. 54,087 Korean long-form task rows generated (median 1,082 chars). Korean answers now run 300–500+ tokens with appropriate structure.
No safety alignment (0.4% of corpus)	—	added. 41,853 rows from WildGuardMix, WildJailbreak and CoCoNot, selected in both directions — refuse the harmful, comply with the benign-looking — because training only refusals produces over-refusal.
EOS token mismatch in the export	`generation_config.json` listed only `<	endoftext

Training throughput also rose 15.1k → 25.1k tokens/s, because block-diagonal attention over ~40 documents costs far less than full causal attention across 32,768 positions. The corrected run is both right and faster.

Model details


Developer	MKD Co., Ltd.
Model	Keural-Cortex-8B
Stage	SFT v2, step 600 of 1,124 (53% of one epoch)
Base	`Qwen/Qwen3-8B` → MKD continued pretraining (41B tokens) → 64K context extension
Parameters	8.19 B, bf16, 16.38 GB
Architecture	`Qwen3ForCausalLM`, 36 layers, hidden 4096, GQA 32/8, head_dim 128
Vocabulary	151,936
Context	65,536 (32,768 native, YaRN factor 2.0)
Training loss at this step	0.8056
License	Apache 2.0, inherited from the base model

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

mid = "mkd-hossain/Keural-Cortex-8B-SFT-v2-step600"
tok = AutoTokenizer.from_pretrained(mid)
model = AutoModelForCausalLM.from_pretrained(mid, torch_dtype="bfloat16", device_map="auto")

messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "김치찌개 끓이는 방법을 알려줘."},
]
text = tok.apply_chat_template(messages, tokenize=False, add_generation_prompt=True,
                              enable_thinking=False)
inputs = tok(text, return_tensors="pt").to(model.device)
out = model.generate(**inputs, max_new_tokens=1024)
print(tok.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

vLLM

vllm serve mkd-hossain/Keural-Cortex-8B-SFT-v2-step600 \
    --served-model-name keural-cortex-8b \
    --max-model-len 65536 \
    --tensor-parallel-size 2 \
    --enable-auto-tool-choice \
    --tool-call-parser hermes

Do not add --reasoning-parser qwen3 with this checkpoint. See limitation 1.

Sampling defaults ship in generation_config.json (temperature 0.7, top_p 0.8, top_k 20, repetition_penalty 1.05).

Training


Hardware	4 × NVIDIA H200 141 GB, FSDP full-shard
Sequence length	32,768
Tokens/step	1,572,864 (micro_bsz 1 × 32,768 × 4 GPUs × grad_accum 12)
Total steps	1,124 (one epoch); this checkpoint is step 600
Packed data	53,956 sequences × 32,768 = 1,768,030,208 tokens
Assistant-only loss	66.3% of tokens carry loss
Optimiser	AdamW, β (0.9, 0.95), weight decay 0.0, grad clip 1.0
LR	peak 1.0e-5, WSD (3% warmup, 17% stable, decay to 5%)
Precision	bfloat16, flash-attention-2, fused linear cross-entropy
Packed attention	block-diagonal via per-document `position_ids`
Throughput	25.1k tokens/s

Data

2,286,851 rows from 89 audited per-source files, deduplicated, 0 render failures, 38 rows dropped for exceeding the sequence length.

bucket	rows	tokens
korean	1,167,291	627 M
core_en	514,609	481 M
tool	109,345	317 M
math_instruct	149,417	47 M
code_instruct	100,151	65 M
tool_single	92,912	36 M
safety	41,853	19 M
korean_cot	39,692	35 M
identity	33,105	1.4 M
reasoning	15,604	100 M
math_cot	10,219	40 M
reasoning_en	5,430	80 M
long_context	5,329	158 M
code_competitive	4,713	40 M

Data generated for this run, where no public source existed:

54,087 Korean task rows (generation, rewrite, summarise, brainstorm, format control, edit) built by persona synthesis — 60 personas × 100 topics, with the teacher writing the prompts rather than filling templates, then answering them
25,515 Korean chain-of-thought rows, 20,395 verified against ground-truth answers
5,822 decline-to-call rows — tools in scope, none relevant; zero incorrect tool invocations across 6,054 teacher generations
1,016 long-context reasoning traces in the 32–64K band, which was 0% of the corpus
3,421 authored identity rows including 215 leading-premise denials

Data quality

A four-axis LLM-as-judge pass over 2,000 stratified rows, weighted toward newly generated data:

axis	score (1–5)
correctness	4.29
compliance	4.26
quality	4.11
safety	4.98

The generated Korean CoT (5.00 / 5.00), identity (4.54 / 4.61) and Korean task data (4.39 / 4.58) scored highest in the corpus.

Known limitations

Reproducible on this checkpoint, tested on clean single-turn contexts.

1. Thinking mode produces an empty block

enable_thinking=True, "17 x 23은 얼마야?"
→ <think>

   </think>

   17 x 23 = 391입니다.

The answer is correct, but <think> is empty. 95.5% of training rows are non-thinking, and in that form every assistant turn opens <think>\n\n</think>\n\n — so the model learned that as the standard opening and reproduces it even when the template leaves the slot free. Thinking rows are 40% of assistant text but only 4.5% of rows, and the opening pattern is learned per row.

Consequence: reasoning appears in the visible content (where it is generally correct) rather than inside <think>. Under vLLM's qwen3 reasoning parser the content field may come back empty, which is why that flag is not recommended above. Mitigation: prefill <think> in the prompt, or read the content field directly.

2. It knows its developer's name but no facts about the company

what is your name?      → "I'm Keural, an AI assistant developed by MKD Co., Ltd."   ✅
where are you from?     → "I was created in Japan"                                    ✗
What is MKD's revenue?  → "MKD, or Macedonian Denar, is the currency of North Macedonia" ✗

The identity data answers who are you, who made you and are you actually ChatGPT, but contains no facts about MKD itself, so the model confabulates when asked for more. Mitigation: state company facts in the system prompt.

3. Korean dialect knowledge is absent, and it does not say so

Asked for Busan dialect greetings it produced circular nonsense rather than admitting it did not know. Standard Korean is unaffected and strong.

Other notes

An explicit output constraint is sometimes ignored (asked for a 3-row table, produced 5)
Tool over-firing (calling an irrelevant tool) is not yet verified on this checkpoint
No preference optimisation (no DPO/RLHF) and no RLVR — both are planned stages
Reported knowledge cutoff reflects the base model, not MKD's training data

Evaluation

No benchmark numbers are published for this intermediate checkpoint; they will accompany the completed model. For reference, the first SFT run measured against Qwen3-8B-Instruct under identical conditions with lm-evaluation-harness:

suite	result
English (mmlu, gsm8k, hellaswag, arc_challenge, winogrande)	+0.95 mean, 3/5 wins
Korean (kmmlu, haerae, kobest_*)	+0.63 mean, 5/6 wins — haerae +5.13

That was achieved with no preference optimisation and with the defects listed above still present.

Citation

@misc{keural-cortex-8b-sft-v2-step600,
  title  = {Keural-Cortex-8B: a bilingual Korean-English 8B model, SFT v2 checkpoint step 600},
  author = {MKD Co., Ltd.},
  year   = {2026},
  url    = {https://huggingface.co/mkd-hossain/Keural-Cortex-8B-SFT-v2-step600}
}

Built on Qwen/Qwen3-8B (Apache 2.0). We thank the Qwen team, and the maintainers of the open datasets used in supervised fine-tuning — notably SmolTalk, Tülu 3, No Robots, WildGuardMix, WildJailbreak and CoCoNot.

Contact

MKD Co., Ltd. — Keural platform team.

Downloads last month: 175

Safetensors

Model size

2B params

Tensor type

BF16

Model tree for mkd-hossain/Keural-Cortex-8B-SFT-v2-step600

Base model

Qwen/Qwen3-8B-Base

Finetuned

Qwen/Qwen3-8B

Finetuned

(2093)

this model