初始化项目，由ModelHub XC社区提供模型

Model: JuliaKreutzerCohere/tiny-aya-global-prompt-userdetail-splitpass
Source: Original Platform

prompts/reason/default.txt

@@ -0,0 +1,16 @@
+CONTEXT:
+{context}
+
+TASK TYPE: `{task_type}`
+
+QUERY:
+{query}
+
+Read the QUERY instructions carefully — this task type may be unfamiliar. Learn only from CONTEXT.
+
+REASONING ONLY for this turn:
+1. Deduce the linguistic rules from CONTEXT examples only.
+2. Plan how to apply them to every QUERY item.
+3. Note the exact answer format the QUERY asks for.
+
+You may draft candidates inside the reasoning. Do NOT write `FINAL ANSWERS:` and do NOT end with a bare answer list.

prompts/reason/fill_blanks.txt

@@ -0,0 +1,16 @@
+CONTEXT:
+{context}
+
+TASK TYPE: `fill_blanks`
+
+QUERY:
+{query}
+
+This is a fill-in-the-blanks task. Missing pieces may be whole words, morphemes, or phonetic segments — match the unit CONTEXT uses.
+
+REASONING ONLY for this turn:
+1. Recover the paradigm / pattern from complete CONTEXT examples.
+2. Identify exactly what each QUERY blank is missing.
+3. Note the answer format: only the missing form per blank, same style as CONTEXT.
+
+You may draft candidates inside the reasoning. Do NOT write `FINAL ANSWERS:` and do NOT end with a bare answer list.

prompts/reason/match_letters.txt

@@ -0,0 +1,16 @@
+CONTEXT:
+{context}
+
+TASK TYPE: `match_letters`
+
+QUERY:
+{query}
+
+This is a matching / multiple-choice letter task. Each QUERY item chooses among labeled options (A, B, C, ...).
+
+REASONING ONLY for this turn:
+1. Extract the mapping or rule set from CONTEXT.
+2. For each QUERY item, evaluate the options against that rule set.
+3. Note the answer format: only the chosen option letter (A, B, C, ...), uppercase — no option text.
+
+You may draft candidates inside the reasoning. Do NOT write `FINAL ANSWERS:` and do NOT end with a bare answer list.

prompts/reason/num_to_text.txt

@@ -0,0 +1,16 @@
+CONTEXT:
+{context}
+
+TASK TYPE: `num_to_text`
+
+QUERY:
+{query}
+
+This is a number-to-text task. CONTEXT shows how digit values are written as number words/forms; convert each QUERY number into that written form.
+
+REASONING ONLY for this turn:
+1. Infer construction rules from CONTEXT (bases, multipliers, conjunctions, morphology).
+2. Plan the written form for each QUERY number.
+3. Note the answer format: only the written numeral form as in CONTEXT — no digits, no English glosses.
+
+You may draft candidates inside the reasoning. Do NOT write `FINAL ANSWERS:` and do NOT end with a bare answer list.

prompts/reason/text_to_num.txt

@@ -0,0 +1,16 @@
+CONTEXT:
+{context}
+
+TASK TYPE: `text_to_num`
+
+QUERY:
+{query}
+
+This is a text-to-number task. CONTEXT shows how number words/expressions map to values; convert each QUERY expression into digits.
+
+REASONING ONLY for this turn:
+1. Infer the number system from CONTEXT (bases, place values, multipliers, word order).
+2. Plan the conversion for each QUERY expression.
+3. Note the answer format: only ordinary digits (e.g. 42) — no words or units.
+
+You may draft candidates inside the reasoning. Do NOT write `FINAL ANSWERS:` and do NOT end with a bare answer list.

prompts/reason/translation.txt

@@ -0,0 +1,16 @@
+CONTEXT:
+{context}
+
+TASK TYPE: `translation`
+
+QUERY:
+{query}
+
+This is a translation task. Use only the CONTEXT examples to learn how forms map between languages (or between orthography and meaning).
+
+REASONING ONLY for this turn:
+1. Align CONTEXT pairs and find systematic correspondences (roots, affixes, word order, agreement).
+2. Plan how to translate each QUERY item.
+3. Note the answer format: only the translated form in the language QUERY asks for — no glosses.
+
+You may draft candidates inside the reasoning. Do NOT write `FINAL ANSWERS:` and do NOT end with a bare answer list.