Do not let HF tokenizers truncate wiki articles at 4096

Yi-6B sets model_max_length=4096. encode() would clip long Wikipedia
pages before we pack seq_len chunks. Raise the cap so only our
chunker limits context.

kda/training/data.py

@@ -75,6 +75,8 @@ def load_tokenizer(source: str) -> Tokenizer:
     from transformers import AutoTokenizer
 
     tok = AutoTokenizer.from_pretrained(source, trust_remote_code=True)
+    # 只借词表分词, 语料随后按 seq_len 切块, 不受原模型 4096 上限约束
+    tok.model_max_length = 10**9
     return HuggingFaceTokenizer(tok)