Initial K3 snapshot: 0.5B KDA/MLA/MoE train path

Standalone tree split from LLMRL/projects/kda. Includes Triton dt_bias
backward fix, train_k3 --preset 0.5b, SFT, Docker runtime, and tests.

data/eval/.gitkeep

@@ -0,0 +1 @@
+# Mount as /data/eval: one sentence per line, e.g. zh2en.src.txt + zh2en.ref.txt

data/eval/en2zh.ref.txt

@@ -0,0 +1,20 @@
+今天天气很好。
+请把窗户打开。
+猫坐在垫子上。
+这本书值得一读。
+他昨天去了北京。
+我们需要更多的训练数据。
+太阳从东边升起。
+她正在学习线性代数。
+不要把评测集拿去训练。
+河对面有一座旧桥。
+科学是对自然的系统探索。
+他们在公园里散步。
+这台电脑的内存是十六吉字节。
+翻译时不要照抄原文。
+春天的风很温和。
+我把钥匙放在桌子上了。
+火车中午到达。
+水在一百摄氏度沸腾。
+小模型仍然可以学会狭窄的任务。
+冻结测试文件保持只读。

data/eval/en2zh.src.txt

@@ -0,0 +1,20 @@
+The weather is very nice today.
+Please open the window.
+The cat is sitting on the mat.
+This book is worth reading.
+He went to Beijing yesterday.
+We need more training data.
+The sun rises in the east.
+She is studying linear algebra.
+Do not train on the evaluation set.
+There is an old bridge across the river.
+Science is the systematic exploration of nature.
+They are walking in the park.
+This computer has sixteen gigabytes of memory.
+Do not copy the source text when translating.
+The spring wind is gentle.
+I left the keys on the table.
+The train arrives at noon.
+Water boils at one hundred degrees Celsius.
+A small model can still learn a narrow task.
+Keep the frozen test files read-only.

data/eval/zh2en.ref.txt

@@ -0,0 +1,20 @@
+The weather is very nice today.
+The development of artificial intelligence has changed the world.
+Please open the window.
+The cat is sitting on the mat.
+This book is worth reading.
+The meeting will start at three in the afternoon.
+He went to Beijing yesterday.
+We need more training data.
+The sun rises in the east.
+This question still has no answer.
+She is studying linear algebra.
+Do not train on the evaluation set.
+There is an old bridge across the river.
+Please wait a moment, I will be right back.
+Science is the systematic exploration of nature.
+They are walking in the park.
+This computer has sixteen gigabytes of memory.
+Do not copy the source text when translating.
+The spring wind is gentle.
+I left the keys on the table.

data/eval/zh2en.src.txt

@@ -0,0 +1,20 @@
+今天天气很好。
+人工智能的发展改变了世界。
+请把窗户打开。
+猫坐在垫子上。
+这本书值得一读。
+会议将在下午三点开始。
+他昨天去了北京。
+我们需要更多的训练数据。
+太阳从东边升起。
+这个问题还没有答案。
+她正在学习线性代数。
+不要把评测集拿去训练。
+河对面有一座旧桥。
+请稍等，我马上回来。
+科学是对自然的系统探索。
+他们在公园里散步。
+这台电脑的内存是十六吉字节。
+翻译时不要照抄原文。
+春天的风很温和。
+我把钥匙放在桌子上了。

data/pretrain/.gitkeep

@@ -0,0 +1 @@
+# Mount bilingual pretrain shards here (not baked into the image).

data/sft/.gitkeep

@@ -0,0 +1,2 @@
+SFT bitext (jsonl or tsv) is mounted here. toy.jsonl is a tiny in-repo twin set.
+OPUS / WMT dumps stay out of git.

data/sft/toy.jsonl

@@ -0,0 +1,32 @@
+{"src": "你好。", "tgt": "Hello.", "target_lang": "en"}
+{"src": "Hello.", "tgt": "你好。", "target_lang": "zh"}
+{"src": "谢谢。", "tgt": "Thank you.", "target_lang": "en"}
+{"src": "Thank you.", "tgt": "谢谢。", "target_lang": "zh"}
+{"src": "我爱北京。", "tgt": "I love Beijing.", "target_lang": "en"}
+{"src": "I love Beijing.", "tgt": "我爱北京。", "target_lang": "zh"}
+{"src": "现在几点？", "tgt": "What time is it now?", "target_lang": "en"}
+{"src": "What time is it now?", "tgt": "现在几点？", "target_lang": "zh"}
+{"src": "水是透明的。", "tgt": "Water is transparent.", "target_lang": "en"}
+{"src": "Water is transparent.", "tgt": "水是透明的。", "target_lang": "zh"}
+{"src": "他是一名教师。", "tgt": "He is a teacher.", "target_lang": "en"}
+{"src": "He is a teacher.", "tgt": "他是一名教师。", "target_lang": "zh"}
+{"src": "明天会下雨。", "tgt": "It will rain tomorrow.", "target_lang": "en"}
+{"src": "It will rain tomorrow.", "tgt": "明天会下雨。", "target_lang": "zh"}
+{"src": "请坐。", "tgt": "Please sit down.", "target_lang": "en"}
+{"src": "Please sit down.", "tgt": "请坐。", "target_lang": "zh"}
+{"src": "这是一只狗。", "tgt": "This is a dog.", "target_lang": "en"}
+{"src": "This is a dog.", "tgt": "这是一只狗。", "target_lang": "zh"}
+{"src": "大门在左边。", "tgt": "The gate is on the left.", "target_lang": "en"}
+{"src": "The gate is on the left.", "tgt": "大门在左边。", "target_lang": "zh"}
+{"src": "我们走吧。", "tgt": "Let's go.", "target_lang": "en"}
+{"src": "Let's go.", "tgt": "我们走吧。", "target_lang": "zh"}
+{"src": "夜空中有星星。", "tgt": "There are stars in the night sky.", "target_lang": "en"}
+{"src": "There are stars in the night sky.", "tgt": "夜空中有星星。", "target_lang": "zh"}
+{"src": "面包放在厨房里。", "tgt": "The bread is in the kitchen.", "target_lang": "en"}
+{"src": "The bread is in the kitchen.", "tgt": "面包放在厨房里。", "target_lang": "zh"}
+{"src": "孩子在睡觉。", "tgt": "The child is sleeping.", "target_lang": "en"}
+{"src": "The child is sleeping.", "tgt": "孩子在睡觉。", "target_lang": "zh"}
+{"src": "这座山很高。", "tgt": "This mountain is very high.", "target_lang": "en"}
+{"src": "This mountain is very high.", "tgt": "这座山很高。", "target_lang": "zh"}
+{"src": "请关上门。", "tgt": "Please close the door.", "target_lang": "en"}
+{"src": "Please close the door.", "tgt": "请关上门。", "target_lang": "zh"}