South Korea8B parameters32K native contextKorean + English

Explore Kanana 1.5 8B Instruct

Use Kakao's bilingual Korean-English instruction model for natural conversation, coding, mathematics, function-calling prompts, and long documents.

Model

kakaocorp/kanana-1.5-8b-instruct-2505

Input

Configure your request

text_inputs

Passed to TextGenerationPipeline as raw text or the model’s system/user/assistant chat-message array.

Parameters

return_full_text return_tensors clean_up_tokenization_spaces skip_special_tokens continue_final_message

handle_long_generation

prefix

stop_sequence

Truncate input

Maximum total tokens

Optional total input-plus-output limit; distinct from max_new_tokens.

Add special tokens

Padding

Tokenizer options (JSON)

Sample output

Temperature

Top P

Top K

Min P

Maximum output tokens

Native context is 32K; the published YaRN configuration can extend it to 128K.

Repetition penalty

num_return_sequences