mkd-hossain/Keural-Cortex-8B-SFT-step600

Keural-Cortex-8B-SFT-step600

An intermediate SFT checkpoint, published for inspection rather than use.

Step 600 of a planned 926-step run that was stopped early. It is 65% trained and has documented defects — see §Known defects before doing anything with it.

Built by MKD Co., Ltd. for the Keural platform.

Dataset: mkd-hossain/keural-cortex-8b-sft

What works

Verified by a 16-case graded battery (10 passed):

Known defects

Read this before using the model.

1. 1.Code generation is broken. It answers code requests with tool-call JSON:

   "Write a python function to add two numbers."
   -> <tool_call>{"name": "add_numbers", "arguments": {"num1": 5, "num2": 7}}</tool_call>

Cause is dataset composition, not training: the code bucket turned out to be 100% competitive programming, and only 3,374 rows in 1.5M taught code writing against ~215,000 teaching tool calls — 64:1.

1. 1.Identity is wrong. Asked who it is, it has answered "ChatGPT by OpenAI", "Mike", and "Smile by Kakao" on separate runs. The 5,116 correct identity rows were 0.33% of the corpus and lost to the pretrained prior. Use a system prompt: 당신은 MKD가 개발한 AI 어시스턴트 Keural입니다.

1. 1.Over-calls tools. Given a tool and asked for a haiku, it calls the tool. The 45,565 decline-to-call rows did not outweigh the call examples.

1. 1.Terse and occasionally empty answers. 23.4% of training answers were under 60 characters (an answer-length floor was lost from the pipeline).

1. 1.Thinking mode is inconsistent and capped at ~16K. Reliable in the 4–16K band, essentially absent above it — the packer's seq_len boundary.

1. 1.No long-context SFT. Positions 16,384–65,536 have CPT training but no instruction training; the long-context phase never ran.

1. 1.No preference optimisation. SFT-only, which shows as inconsistent response length and format adherence.

Usage

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

msgs = [{"role": "system", "content": "당신은 MKD가 개발한 AI 어시스턴트 Keural입니다."},
        {"role": "user", "content": "김치찌개 맛있게 끓이는 방법을 알려줘."}]
ids = tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt").to(model.device)
print(tok.decode(model.generate(ids, max_new_tokens=512, temperature=0.7, top_p=0.9)[0][ids.shape[1]:]))

vLLM, with tool calling and reasoning parsed:

vllm serve mkd-hossain/Keural-Cortex-8B-SFT-step600 \
  --tensor-parallel-size 4 --max-model-len 16384 \
  --enable-auto-tool-choice --tool-call-parser hermes --reasoning-parser qwen3

Use --max-model-len 16384. It will load at 65,536, but positions past 16,384 received no SFT training.

Thinking mode is carried by the data, not a flag: a row with <think>…</think> trains thinking, a row without trains <think>\n\n</think>. enable_thinking=False pre-fills the empty block.

Evaluation

Not run. The base CPT evaluation (CPT_details) found the 41B-token CPT did not improve Korean (Δ mean −0.09 over 60 tasks) and cost −0.68 MMLU. Any claim about this model should be measured against that baseline, not against stock Qwen3.

Per the project plan, the intended claim is Korean-qualified: better on Korean benchmarks, parity on English knowledge and code, behind on English math.

License

Apache-2.0, inherited from Qwen3-8B. The SFT data is mixed-licence — see the dataset card.