Normalize product data prompt formatting

This commit is contained in:
wuyang6
2026-05-28 14:30:55 +08:00
parent 2dc941d6d8
commit 6364fb437c
6 changed files with 106 additions and 13 deletions
@@ -651,12 +651,15 @@ def build_planning_prompt(
) -> str:
"""把训练 instruction 包成评测侧使用的 chat template。"""
instruction = build_training_instruction(
record,
session_num=session_num,
session_time_minutes=session_time_minutes,
context_fields=context_fields,
instruction = prompt_section_with_one_trailing_newline(
build_training_instruction(
record,
session_num=session_num,
session_time_minutes=session_time_minutes,
context_fields=context_fields,
)
)
system_prompt = prompt_section_without_trailing_newline(system_prompt)
return (
f"<|im_start|>system\n{system_prompt}<|im_end|>\n"
f"<|im_start|>user\n{instruction}<|im_end|>\n"
@@ -664,6 +667,14 @@ def build_planning_prompt(
)
def prompt_section_without_trailing_newline(text: str) -> str:
return str(text or "").rstrip("\n")
def prompt_section_with_one_trailing_newline(text: str) -> str:
return prompt_section_without_trailing_newline(text) + "\n"
def training_jsonl_line(
record: dict[str, Any],
*,
@@ -673,7 +684,7 @@ def training_jsonl_line(
system_prompt: str,
) -> str:
payload = {
"system": system_prompt,
"system": prompt_section_without_trailing_newline(system_prompt),
"instruction": build_training_instruction(
record,
session_num=session_num,