itsusony/LFM2.5-1.2B-Zhusuan

test_chat: localize UI/comments to Japanese, raise max tokens

Release Zhusuan R5b: hard 100%, hard_div 97.5%, xl 100%

sft: add R5a/R5b big-mult data generators

Release Zhusuan R4: hard arithmetic 33%→98% vs base

docs: document test_chat.py / test_100.py usage in all three languages

test: random N-problem accuracy script (test_100.py)

docs: expand demo translations by default (details open)

test: script verifying EN/JA thinking is not supported

docs: add EN/JA human translations of the demo think trace

docs: self-contained ZH/JA sections in their own languages

docs: slim README for readability

docs: demo examples in EN/JA prompts with real base-vs-R3 outputs

R3 fine-tune: new weights + default zhusuan chat template

Remove write test file

Upload LFM2.5-1.2B-Zhusuan weights (98% eval) + EN/ZH/JA card

test write token

initial commit