Skip to content

bge-reranker-v2-m3 Reranker #63

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Dec 23, 2024
Merged

bge-reranker-v2-m3 Reranker #63

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
141 changes: 96 additions & 45 deletions modelcache/adapter/adapter_query.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,9 @@
from modelcache.utils.error import NotInitError
from modelcache.utils.time import time_cal
from modelcache.processor.pre import multi_analysis
from FlagEmbedding import FlagReranker

USE_RERANKER = True # 如果为 True 则启用 reranker，否则使用原有逻辑

def adapt_query(cache_data_convert, *args, **kwargs):
chat_cache = kwargs.pop("cache_obj", cache)
Expand Down Expand Up @@ -74,53 +76,102 @@ def adapt_query(cache_data_convert, *args, **kwargs):
if rank_pre < rank_threshold:
return

for cache_data in cache_data_list:
primary_id = cache_data[1]
ret = chat_cache.data_manager.get_scalar_data(
cache_data, extra_param=context.get("get_scalar_data", None)
)
if ret is None:
continue
if USE_RERANKER:
reranker = FlagReranker('BAAI/bge-reranker-v2-m3', use_fp16=False)
for cache_data in cache_data_list:
primary_id = cache_data[1]
ret = chat_cache.data_manager.get_scalar_data(
cache_data, extra_param=context.get("get_scalar_data", None)
)
if ret is None:
continue

if "deps" in context and hasattr(ret.question, "deps"):
eval_query_data = {
"question": context["deps"][0]["data"],
"embedding": None
}
eval_cache_data = {
"question": ret.question.deps[0].data,
"answer": ret.answers[0].answer,
"search_result": cache_data,
"embedding": None,
}
else:
eval_query_data = {
"question": pre_embedding_data,
"embedding": embedding_data,
}
rank = reranker.compute_score([pre_embedding_data, ret[0]], normalize=True)

eval_cache_data = {
"question": ret[0],
"answer": ret[1],
"search_result": cache_data,
"embedding": None
}
rank = chat_cache.similarity_evaluation.evaluation(
eval_query_data,
eval_cache_data,
extra_param=context.get("evaluation_func", None),
)
if "deps" in context and hasattr(ret.question, "deps"):
eval_query_data = {
"question": context["deps"][0]["data"],
"embedding": None
}
eval_cache_data = {
"question": ret.question.deps[0].data,
"answer": ret.answers[0].answer,
"search_result": cache_data,
"embedding": None,
}
else:
eval_query_data = {
"question": pre_embedding_data,
"embedding": embedding_data,
}

eval_cache_data = {
"question": ret[0],
"answer": ret[1],
"search_result": cache_data,
"embedding": None
}

if len(pre_embedding_data) <= 256:
if rank_threshold <= rank:
cache_answers.append((rank, ret[1]))
cache_questions.append((rank, ret[0]))
cache_ids.append((rank, primary_id))
else:
if rank_threshold_long <= rank:
cache_answers.append((rank, ret[1]))
cache_questions.append((rank, ret[0]))
cache_ids.append((rank, primary_id))
else:
# 不使用 reranker 时，走原来的逻辑
for cache_data in cache_data_list:
primary_id = cache_data[1]
ret = chat_cache.data_manager.get_scalar_data(
cache_data, extra_param=context.get("get_scalar_data", None)
)
if ret is None:
continue

if "deps" in context and hasattr(ret.question, "deps"):
eval_query_data = {
"question": context["deps"][0]["data"],
"embedding": None
}
eval_cache_data = {
"question": ret.question.deps[0].data,
"answer": ret.answers[0].answer,
"search_result": cache_data,
"embedding": None,
}
else:
eval_query_data = {
"question": pre_embedding_data,
"embedding": embedding_data,
}

eval_cache_data = {
"question": ret[0],
"answer": ret[1],
"search_result": cache_data,
"embedding": None
}
rank = chat_cache.similarity_evaluation.evaluation(
eval_query_data,
eval_cache_data,
extra_param=context.get("evaluation_func", None),
)

if len(pre_embedding_data) <= 256:
if rank_threshold <= rank:
cache_answers.append((rank, ret[1]))
cache_questions.append((rank, ret[0]))
cache_ids.append((rank, primary_id))
else:
if rank_threshold_long <= rank:
cache_answers.append((rank, ret[1]))
cache_questions.append((rank, ret[0]))
cache_ids.append((rank, primary_id))

if len(pre_embedding_data) <= 256:
if rank_threshold <= rank:
cache_answers.append((rank, ret[1]))
cache_questions.append((rank, ret[0]))
cache_ids.append((rank, primary_id))
else:
if rank_threshold_long <= rank:
cache_answers.append((rank, ret[1]))
cache_questions.append((rank, ret[0]))
cache_ids.append((rank, primary_id))
cache_answers = sorted(cache_answers, key=lambda x: x[0], reverse=True)
cache_questions = sorted(cache_questions, key=lambda x: x[0], reverse=True)
cache_ids = sorted(cache_ids, key=lambda x: x[0], reverse=True)
Expand All @@ -141,4 +192,4 @@ def adapt_query(cache_data_convert, *args, **kwargs):
logging.info('update_hit_count except, please check!')

chat_cache.report.hint_cache()
return cache_data_convert(return_message, return_query)
return cache_data_convert(return_message, return_query)