|
|
|
@ -91,13 +91,13 @@ SUMMARY_PROMPT = """ |
|
|
|
例子: |
|
|
|
<EXAMPLE> |
|
|
|
|
|
|
|
文内引用示例（使用markdown和html的混合语法，格式必须和例子一致）: |
|
|
|
"XGBoost是非常强大的集成学习模型[<sup>[2]</sup>](#2)" |
|
|
|
(必须使用 "[<sup>[index]</sup>](#index)" 而不是 "[index]"，注意不要遗漏"(#index)"，这里的index是chunk对应的reference的id) |
|
|
|
文内引用示例（使用markdown脚注）: |
|
|
|
"XGBoost是非常强大的集成学习模型[^2]" |
|
|
|
(必须使用 "[^index]"，这里的index是对应的<reference>的id) |
|
|
|
|
|
|
|
|
|
|
|
文末引用示例 (需要href与前文参考中的一致，不需要对每个chunk分配一个引用，而是每一个referecen共用一个引用): |
|
|
|
<div id="2"><a href="files/docs/chap_001_003_models.md" style="text-decoration: none;>[2] chap_001_models.md</a></div> |
|
|
|
文末引用示例 (需要与前文reference的href一致，不需要对每个chunk分配一个引用，而是每一个referecen共用一个引用): |
|
|
|
[^2]: files/docs/chap_001_003_models.md |
|
|
|
|
|
|
|
</EXAMPLE> |
|
|
|
|
|
|
|
@ -328,7 +328,6 @@ class DeepSearch(BaseAgent): |
|
|
|
- A list of retrieved document results |
|
|
|
- Additional information about the retrieval process |
|
|
|
""" |
|
|
|
max_iter = kwargs.pop("max_iter", self.max_iter) |
|
|
|
### SUB QUERIES ### |
|
|
|
log.color_print(f"<query> {original_query} </query>\n") |
|
|
|
all_search_results = [] |
|
|
|
@ -342,7 +341,7 @@ class DeepSearch(BaseAgent): |
|
|
|
log.color_print(f"</think> Break down the original query into new sub queries: {sub_queries} ") |
|
|
|
all_sub_queries.extend(sub_queries) |
|
|
|
|
|
|
|
for it in range(max_iter): |
|
|
|
for it in range(self.max_iter): |
|
|
|
log.color_print(f">> Iteration: {it + 1}\n") |
|
|
|
|
|
|
|
# Execute all search tasks sequentially |
|
|
|
@ -358,7 +357,7 @@ class DeepSearch(BaseAgent): |
|
|
|
) |
|
|
|
# search_res_from_internet = deduplicate_results(search_res_from_internet) |
|
|
|
# all_search_res.extend(search_res_from_vectordb + search_res_from_internet) |
|
|
|
if it + 1 >= max_iter: |
|
|
|
if it + 1 >= self.max_iter: |
|
|
|
log.color_print("</think> Exceeded maximum iterations. Exiting. ") |
|
|
|
break |
|
|
|
### REFLECTION & GET MORE SUB QUERIES ### |
|
|
|
@ -421,10 +420,11 @@ class DeepSearch(BaseAgent): |
|
|
|
for i, reference in enumerate(references): |
|
|
|
formated = f"<reference id='{i + 1}' href='{reference}'>\n" + "".join( |
|
|
|
[ |
|
|
|
f"<chunk id='{j + 1 + chunk_count}'>\n{chunk}\n</chunk>\n" |
|
|
|
f"<chunk id='{j + 1 + chunk_count}'>\n{chunk}\n</chunk id='{j + 1 + chunk_count}'>\n" |
|
|
|
for j, chunk in enumerate(references[reference]) |
|
|
|
] |
|
|
|
) + "</reference>\n" |
|
|
|
) + f"</reference id='{i + 1}'>\n" |
|
|
|
print(formated) |
|
|
|
chunks.append(formated) |
|
|
|
chunk_count += len(references[reference]) |
|
|
|
return "".join(chunks) |
|
|
|
|
