Skip to content

RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.

Refactor: rename prefetch_size to rerank_candidates (#18718)

Wang Qi4c6f745c1e6f
diff --git a/agent/tools/retrieval.py b/agent/tools/retrieval.py
index 5c98ebef78..81d5b286d7 100644
--- a/agent/tools/retrieval.py
+++ b/agent/tools/retrieval.py
@@ -58,7 +58,7 @@ class RetrievalParam(ToolParamBase):
self.similarity_threshold = 0.2
self.keywords_similarity_weight = 0.5
self.top_n = 8
- self.prefetch_size = 64
+ self.rerank_candidates_count = 64
self.top_k = 1024
self.dataset_ids = []
self.kb_ids = [] # Deprecated: keep for backward compatibility
@@ -194,7 +194,7 @@ class Retrieval(ToolBase, ABC):
aggs=True,
rerank_mdl=rerank_mdl,
rank_feature=label_question(query, kbs),
- prefetch_size=self._param.prefetch_size,
+ rerank_candidates_count=self._param.rerank_candidates_count,
)
if self.check_if_canceled("Retrieval processing"):
return
diff --git a/api/apps/restful_apis/bot_api.py b/api/apps/restful_apis/bot_api.py
index 4038877d19..453e1f88c5 100644
--- a/api/apps/restful_apis/bot_api.py
+++ b/api/apps/restful_apis/bot_api.py
@@ -349,7 +349,7 @@ async def retrieval_test_embedded(tenant_id=None):
vector_similarity_weight = float(req.get("vector_similarity_weight", 0.3))
use_kg = req.get("use_kg", False)
top = int(req.get("top_k", 1024))
- prefetch_size = int(req.get("prefetch_size", 64))
+ rerank_candidates_count = int(req.get("rerank_candidates_count", 64))
if top <= 0:
return get_error_data_result("`top_k` must be greater than 0")
langs = req.get("cross_languages", [])
@@ -359,7 +359,7 @@ async def retrieval_test_embedded(tenant_id=None):
search_config = {}
async def _retrieval():
- nonlocal similarity_threshold, vector_similarity_weight, top, rerank_id, prefetch_size
+ nonlocal similarity_threshold, vector_similarity_weight, top, rerank_id, rerank_candidates_count
local_doc_ids = list(doc_ids) if doc_ids else []
tenant_ids = []
_question = question
@@ -388,8 +388,8 @@ async def retrieval_test_embedded(tenant_id=None):
top = int(search_config.get("top_k", top))
if not req.get("rerank_id"):
rerank_id = search_config.get("rerank_id", "")
- if not req.get("prefetch_size"):
- prefetch_size = int(search_config.get("prefetch_size", 100))
+ if not req.get("rerank_candidates_count"):
+ rerank_candidates_count = int(search_config.get("rerank_candidates_count", 100))
else:
meta_data_filter = req.get("meta_data_filter") or {}
if meta_data_filter.get("method") in ["auto", "semi_auto"]:
@@ -450,7 +450,7 @@ async def retrieval_test_embedded(tenant_id=None):
rerank_mdl=rerank_mdl,
highlight=req.get("highlight"),
rank_feature=labels,
- prefetch_size=prefetch_size,
+ rerank_candidates_count=rerank_candidates_count,
)
if use_kg:
default_chat_model = await thread_pool_exec(get_tenant_default_model_by_type, kb.tenant_id, LLMType.CHAT)
diff --git a/api/apps/restful_apis/chat_api.py b/api/apps/restful_apis/chat_api.py
index 51306aa8b3..3139a71cd5 100644
--- a/api/apps/restful_apis/chat_api.py
+++ b/api/apps/restful_apis/chat_api.py
@@ -200,7 +200,7 @@ def _build_default_completion_dialog():
prompt_config=deepcopy(_DEFAULT_DIRECT_CHAT_PROMPT_CONFIG),
kb_ids=[],
top_n=6,
- prefetch_size=64,
+ rerank_candidates_count=64,
top_k=1024,
rerank_id="",
similarity_threshold=0.1,
@@ -462,7 +462,7 @@ async def create():
req.setdefault("llm_setting", {})
req.setdefault("description", "A helpful Assistant")
req.setdefault("top_n", 6)
- req.setdefault("prefetch_size", 64)
+ req.setdefault("rerank_candidates_count", 64)
req.setdefault("top_k", 1024)
req.setdefault("rerank_id", "")
req.setdefault("similarity_threshold", 0.1)
diff --git a/api/apps/services/dataset_api_service.py b/api/apps/services/dataset_api_service.py
index 1ec150dc2c..75e43f17f0 100644
--- a/api/apps/services/dataset_api_service.py
+++ b/api/apps/services/dataset_api_service.py
@@ -1030,7 +1030,7 @@ async def search(dataset_id: str, tenant_id: str, req: dict):
page = int(req.get("page", 1))
size = int(req.get("size", 30))
- prefetch_size = int(req.get("prefetch_size", 64))
+ rerank_candidates_count = int(req.get("rerank_candidates_count", 64))
question = req.get("question", "")
doc_ids = req.get("doc_ids", [])
use_kg = req.get("use_kg", False)
@@ -1066,7 +1066,7 @@ async def search(dataset_id: str, tenant_id: str, req: dict):
similarity_threshold = float(search_config.get("similarity_threshold", similarity_threshold))
vector_similarity_weight = float(search_config.get("vector_similarity_weight", vector_similarity_weight))
top = max(1, min(int(search_config.get("top_k", top)), 2048))
- prefetch_size = int(search_config.get("prefetch_size", 100))
+ rerank_candidates_count = int(search_config.get("rerank_candidates_count", 100))
use_kg = search_config.get("use_kg", use_kg)
langs = search_config.get("cross_languages", langs)
logging.debug(
@@ -1146,7 +1146,7 @@ async def search(dataset_id: str, tenant_id: str, req: dict):
rerank_mdl=rerank_mdl,
rank_feature=labels,
trace_id=search_id,
- prefetch_size=prefetch_size,
+ rerank_candidates_count=rerank_candidates_count,
)
if use_kg:
@@ -1413,7 +1413,7 @@ async def search_datasets(tenant_id: str, req: dict):
kb_ids = req.get("dataset_ids", [])
page = int(req.get("page", 1))
size = int(req.get("size", 30))
- prefetch_size = int(req.get("prefetch_size", 64))
+ rerank_candidates_count = int(req.get("rerank_candidates_count", 64))
question = req.get("question", "")
doc_ids = req.get("doc_ids", [])
use_kg = req.get("use_kg", False)
@@ -1461,7 +1461,7 @@ async def search_datasets(tenant_id: str, req: dict):
similarity_threshold = float(search_config.get("similarity_threshold", similarity_threshold))
vector_similarity_weight = float(search_config.get("vector_similarity_weight", vector_similarity_weight))
top = max(1, min(int(search_config.get("top_k", top)), 2048))
- prefetch_size = int(search_config.get("prefetch_size", 100))
+ rerank_candidates_count = int(search_config.get("rerank_candidates_count", 100))
use_kg = search_config.get("use_kg", use_kg)
langs = search_config.get("cross_languages", langs)
logging.debug(
@@ -1546,7 +1546,7 @@ async def search_datasets(tenant_id: str, req: dict):
rank_feature=labels,
trace_id=search_id,
must_not=None if req.get("include_knowledge_compilation", True) else {"exists": "compile_kwd"},
- prefetch_size=prefetch_size,
+ rerank_candidates_count=rerank_candidates_count,
)
if use_kg:
diff --git a/api/db/db_models.py b/api/db/db_models.py
index e6cf7a00ea..80b925b574 100644
--- a/api/db/db_models.py
+++ b/api/db/db_models.py
@@ -1475,7 +1475,7 @@ class Dialog(DataBaseModel):
vector_similarity_weight = FloatField(default=0.3)
top_n = IntegerField(default=6)
- prefetch_size = IntegerField(default=64)
+ rerank_candidates_count = IntegerField(default=64)
top_k = IntegerField(default=1024)
@@ -2413,7 +2413,7 @@ def migrate_db():
alter_db_add_column(migrator, "document", "suffix", EmptyStringCharField(max_length=32, null=False, default="", help_text="The real file extension suffix", index=True))
alter_db_add_column(migrator, "api_4_conversation", "errors", TextField(null=True, help_text="errors"))
alter_db_add_column(migrator, "dialog", "meta_data_filter", JSONField(null=True, default={}))
- alter_db_add_column(migrator, "dialog", "prefetch_size", IntegerField(default=64))
+ alter_db_add_column(migrator, "dialog", "rerank_candidates_count", IntegerField(default=64))
alter_db_column_type(migrator, "canvas_template", "title", JSONField(null=True, default=dict, help_text="Canvas title"))
alter_db_column_type(migrator, "canvas_template", "description", JSONField(null=True, default=dict, help_text="Canvas description"))
alter_db_add_column(migrator, "user_canvas", "canvas_category", CharField(max_length=32, null=False, default="agent_canvas", help_text="agent_canvas|dataflow_canvas", index=True))
diff --git a/api/db/services/dialog_service.py b/api/db/services/dialog_service.py
index 45529fdd70..1ad91c3408 100644
--- a/api/db/services/dialog_service.py
+++ b/api/db/services/dialog_service.py
@@ -222,7 +222,7 @@ class DialogService(CommonService):
cls.model.similarity_threshold,
cls.model.vector_similarity_weight,
cls.model.top_n,
- cls.model.prefetch_size,
+ cls.model.rerank_candidates_count,
cls.model.top_k,
cls.model.do_refer,
cls.model.rerank_id,
@@ -670,7 +670,7 @@ async def async_chat(dialog, messages, stream=True, **kwargs):
text_attachments_content, image_attachments, image_files = get_files_content(messages[-1], llm_model_config["model_type"])
prompt_config = dialog.prompt_config
- prefetch_size = getattr(dialog, "prefetch_size", 64)
+ rerank_candidates_count = getattr(dialog, "rerank_candidates_count", 64)
include_reference_metadata, metadata_fields = _resolve_reference_metadata(prompt_config, request_payload=kwargs)
field_map = KnowledgebaseService.get_field_map(dialog.kb_ids)
logging.debug(f"field_map retrieved: {field_map}")
@@ -744,7 +744,7 @@ async def async_chat(dialog, messages, stream=True, **kwargs):
similarity_threshold=0.2,
vector_similarity_weight=0.3,
doc_ids=scoped_doc_ids,
- prefetch_size=prefetch_size,
+ rerank_candidates_count=rerank_candidates_count,
),
internet_enabled=use_web_search,
)
@@ -784,7 +784,7 @@ async def async_chat(dialog, messages, stream=True, **kwargs):
aggs=True,
rerank_mdl=rerank_mdl,
rank_feature=label_question(" ".join(questions), kbs),
- prefetch_size=prefetch_size,
+ rerank_candidates_count=rerank_candidates_count,
)
if prompt_config.get("toc_enhance"):
cks = await retriever.retrieval_by_toc(" ".join(questions), kbinfos["chunks"], tenant_ids, chat_mdl, dialog.top_n)
@@ -1833,7 +1833,7 @@ async def async_ask(question, kb_ids, tenant_id, chat_llm_name=None, search_conf
rerank_mdl=rerank_mdl,
rank_feature=label_question(question, kbs),
trace_id=search_id,
- prefetch_size=search_config.get("prefetch_size", 100),
+ rerank_candidates_count=search_config.get("rerank_candidates_count", 100),
)
if include_reference_metadata:
logging.debug(
@@ -1935,7 +1935,7 @@ async def gen_mindmap(question, kb_ids, tenant_id, search_config={}):
aggs=False,
rerank_mdl=rerank_mdl,
rank_feature=label_question(question, kbs),
- prefetch_size=search_config.get("prefetch_size", 100),
+ rerank_candidates_count=search_config.get("rerank_candidates_count", 100),
)
mindmap = MindMapExtractor(chat_mdl)
mind_map = await mindmap([c["content_with_weight"] for c in ranks["chunks"]])
diff --git a/internal/agent/runtime/retrieval.go b/internal/agent/runtime/retrieval.go
index b9e343bb54..e83e19ab75 100644
--- a/internal/agent/runtime/retrieval.go
+++ b/internal/agent/runtime/retrieval.go
@@ -58,7 +58,7 @@ type RetrievalRequest struct {
DatasetIDs []string
MemoryIDs []string
TopN int
- PrefetchSize int
+ RerankCandidatesCount int
TopK int
KeywordsSimilarityWeight *float64
UseKG bool
diff --git a/internal/agent/tool/registry.go b/internal/agent/tool/registry.go
index dea39254c9..6de113cd81 100644
--- a/internal/agent/tool/registry.go
+++ b/internal/agent/tool/registry.go
@@ -395,7 +395,7 @@ func buildRetrievalTool(params map[string]any) (einotool.BaseTool, error) {
defaults := retrievalArgs{}
for key := range params {
switch key {
- case "dataset_ids", "kb_ids", "top_n", "prefetch_size", "top_k", "similarity_threshold",
+ case "dataset_ids", "kb_ids", "top_n", "rerank_candidates_count", "top_k", "similarity_threshold",
"keywords_similarity_weight", "use_kg", "rerank_id", "empty_response",
"toc_enhance", "meta_data_filter", "retrieval_from", "memory_ids",
"kb_vars", "cross_languages", "function_name", "description", "meta",
@@ -426,12 +426,12 @@ func buildRetrievalTool(params map[string]any) (einotool.BaseTool, error) {
if v, ok := intParam(params, "top_n"); ok {
defaults.TopN = v
}
- if raw, exists := params["prefetch_size"]; exists {
+ if raw, exists := params["rerank_candidates_count"]; exists {
value, ok := strictInt(raw)
if !ok || value <= 0 {
- return nil, fmt.Errorf("agent tool: retrieval tool requires positive integer node-level param prefetch_size")
+ return nil, fmt.Errorf("agent tool: retrieval tool requires positive integer node-level param rerank_candidates_count")
}
- defaults.PrefetchSize = value
+ defaults.RerankCandidatesCount = value
}
if raw, exists := params["top_k"]; exists {
value, ok := strictInt(raw)
diff --git a/internal/agent/tool/retrieval.go b/internal/agent/tool/retrieval.go
index 3446cd0593..c47c4ab2c3 100644
--- a/internal/agent/tool/retrieval.go
+++ b/internal/agent/tool/retrieval.go
@@ -61,7 +61,7 @@ type retrievalArgs struct {
KBIDs []string `json:"kb_ids,omitempty"`
MemoryIDs []string `json:"memory_ids,omitempty"`
TopN int `json:"top_n,omitempty"`
- PrefetchSize int `json:"prefetch_size,omitempty"`
+ RerankCandidatesCount int `json:"rerank_candidates_count,omitempty"`
TopK int `json:"top_k,omitempty"`
KeywordsSimilarityWeight *float64 `json:"keywords_similarity_weight,omitempty"`
UseKG bool `json:"use_kg,omitempty"`
@@ -201,7 +201,7 @@ func (r *RetrievalTool) InvokableRun(ctx context.Context, argumentsInJSON string
DatasetIDs: args.DatasetIDs,
MemoryIDs: args.MemoryIDs,
TopN: args.TopN,
- PrefetchSize: args.PrefetchSize,
+ RerankCandidatesCount: args.RerankCandidatesCount,
TopK: args.TopK,
KeywordsSimilarityWeight: args.KeywordsSimilarityWeight,
UseKG: args.UseKG,
@@ -276,8 +276,8 @@ func (r *RetrievalTool) mergeDefaults(args retrievalArgs) retrievalArgs {
if args.TopN <= 0 {
args.TopN = r.defaults.TopN
}
- if args.PrefetchSize <= 0 {
- args.PrefetchSize = r.defaults.PrefetchSize
+ if args.RerankCandidatesCount <= 0 {
+ args.RerankCandidatesCount = r.defaults.RerankCandidatesCount
}
if args.TopK <= 0 {
args.TopK = r.defaults.TopK
diff --git a/internal/agent/tool/retrieval_nlp.go b/internal/agent/tool/retrieval_nlp.go
index 95281d3e3a..9fbcfbbf79 100644
--- a/internal/agent/tool/retrieval_nlp.go
+++ b/internal/agent/tool/retrieval_nlp.go
@@ -325,8 +325,8 @@ func nlpRequestFromRetrieval(
Aggs: boolPtr(false),
Highlight: boolPtr(false),
}
- if req.PrefetchSize != 0 {
- nlpReq.PrefetchSize = &req.PrefetchSize
+ if req.RerankCandidatesCount != 0 {
+ nlpReq.RerankCandidatesCount = &req.RerankCandidatesCount
}
if req.TopK > 0 {
nlpReq.Top = &req.TopK
diff --git a/internal/agent/tool/retrieval_service.go b/internal/agent/tool/retrieval_service.go
index 6308b1cd9d..85c16f0401 100644
--- a/internal/agent/tool/retrieval_service.go
+++ b/internal/agent/tool/retrieval_service.go
@@ -52,7 +52,7 @@ type RetrievalRequest struct {
DatasetIDs []string
MemoryIDs []string
TopN int
- PrefetchSize int
+ RerankCandidatesCount int
TopK int
KeywordsSimilarityWeight *float64
UseKG bool
diff --git a/internal/entity/chat.go b/internal/entity/chat.go
index 1dca67d514..a7b861fc4a 100644
--- a/internal/entity/chat.go
+++ b/internal/entity/chat.go
@@ -41,7 +41,7 @@ type Chat struct {
SimilarityThreshold float64 `gorm:"column:similarity_threshold" json:"similarity_threshold"`
VectorSimilarityWeight float64 `gorm:"column:vector_similarity_weight" json:"vector_similarity_weight"`
TopN int64 `gorm:"column:top_n" json:"top_n"`
- PrefetchSize int64 `gorm:"column:prefetch_size" json:"prefetch_size"`
+ RerankCandidatesCount int64 `gorm:"column:rerank_candidates_count" json:"rerank_candidates_count"`
TopK int64 `gorm:"column:top_k" json:"top_k"`
DoRefer string `gorm:"column:do_refer;size:1;not null" json:"do_refer"`
RerankID string `gorm:"column:rerank_id;size:128;not null;default:''" json:"rerank_id"`
diff --git a/internal/handler/chat.go b/internal/handler/chat.go
index 6eced41eae..c68b03dc1f 100644
--- a/internal/handler/chat.go
+++ b/internal/handler/chat.go
@@ -369,7 +369,7 @@ func (h *ChatHandler) GetChat(c *gin.Context) {
"similarity_threshold": chat.SimilarityThreshold,
"vector_similarity_weight": chat.VectorSimilarityWeight,
"top_n": chat.TopN,
- "prefetch_size": chat.PrefetchSize,
+ "rerank_candidates_count": chat.RerankCandidatesCount,
"top_k": chat.TopK,
"do_refer": chat.DoRefer,
"rerank_id": chat.RerankID,
diff --git a/internal/handler/mindmap.go b/internal/handler/mindmap.go
index 710651d497..c5dc274ca4 100644
--- a/internal/handler/mindmap.go
+++ b/internal/handler/mindmap.go
@@ -106,7 +106,7 @@ func mindMapRetrievalRequest(question string, kbIDs common.StringSlice, searchID
page := 1
size := 12
topK := intFromConfig(searchConfig, "top_k", 1024)
- prefetchSize := intFromConfig(searchConfig, "prefetch_size", 100)
+ rerankCandidatesCount := intFromConfig(searchConfig, "rerank_candidates_count", 100)
similarityThreshold := floatFromConfig(searchConfig, "similarity_threshold", 0.2)
vectorSimilarityWeight := floatFromConfig(searchConfig, "vector_similarity_weight", 0.3)
req := &service.RetrievalTestRequest{
@@ -115,7 +115,7 @@ func mindMapRetrievalRequest(question string, kbIDs common.StringSlice, searchID
Page: &page,
Size: &size,
TopK: &topK,
- PrefetchSize: &prefetchSize,
+ RerankCandidatesCount: &rerankCandidatesCount,
SimilarityThreshold: &similarityThreshold,
VectorSimilarityWeight: &vectorSimilarityWeight,
DocIDs: stringSliceFromConfig(searchConfig, "doc_ids"),
diff --git a/internal/handler/searchbot.go b/internal/handler/searchbot.go
index f5c8251c1e..5a2cc502b5 100644
--- a/internal/handler/searchbot.go
+++ b/internal/handler/searchbot.go
@@ -52,7 +52,7 @@ type SearchBotRetrievalTestRequest struct {
Question string `json:"question" binding:"required"`
Page *int `json:"page,omitempty"`
Size *int `json:"size,omitempty"`
- PrefetchSize *int `json:"prefetch_size,omitempty"`
+ RerankCandidatesCount *int `json:"rerank_candidates_count,omitempty"`
DocIDs []string `json:"doc_ids,omitempty"`
UseKG *bool `json:"use_kg,omitempty"`
TopK *int `json:"top_k,omitempty"`
@@ -510,7 +510,7 @@ func toRetrievalServiceRequest(h *SearchBotRetrievalTestRequest) *service.Retrie
Question: h.Question,
Page: h.Page,
Size: h.Size,
- PrefetchSize: h.PrefetchSize,
+ RerankCandidatesCount: h.RerankCandidatesCount,
DocIDs: h.DocIDs,
UseKG: h.UseKG,
TopK: h.TopK,
diff --git a/internal/service/ask_service.go b/internal/service/ask_service.go
index 2a2c4a338f..0c5bbe387f 100644
--- a/internal/service/ask_service.go
+++ b/internal/service/ask_service.go
@@ -30,7 +30,7 @@ import (
const (
DefaultAskPage = 1
DefaultAskPageSize = 12
- DefaultAskPrefetchSize = 100
+ DefaultAskRerankCandidatesCount = 100
DefaultAskTopK = 1024
DefaultAskSimilarityThreshold = 0.1
DefaultAskVectorSimilarityWeight = 0.3
@@ -66,7 +66,7 @@ type AskStreamOptions struct {
DocIDs []string
UseKG *bool
TopK *int
- PrefetchSize *int
+ RerankCandidatesCount *int
CrossLanguages []string
Filter map[string]interface{}
TenantRerankID *string
@@ -161,9 +161,9 @@ func (s *AskService) run(ctx context.Context, llm StreamingLLM, userID, question
if opts.VectorSimilarityWeight != nil {
vectorSimilarityWeight = *opts.VectorSimilarityWeight
}
- prefetchSize := DefaultAskPrefetchSize
- if opts.PrefetchSize != nil {
- prefetchSize = *opts.PrefetchSize
+ rerankCandidatesCount := DefaultAskRerankCandidatesCount
+ if opts.RerankCandidatesCount != nil {
+ rerankCandidatesCount = *opts.RerankCandidatesCount
}
req := &RetrievalTestRequest{
@@ -172,7 +172,7 @@ func (s *AskService) run(ctx context.Context, llm StreamingLLM, userID, question
DocIDs: opts.DocIDs,
UseKG: opts.UseKG,
TopK: ptrInt(topK),
- PrefetchSize: ptrInt(prefetchSize),
+ RerankCandidatesCount: ptrInt(rerankCandidatesCount),
CrossLanguages: opts.CrossLanguages,
Filter: opts.Filter,
TenantRerankID: opts.TenantRerankID,
diff --git a/internal/service/chat.go b/internal/service/chat.go
index d4858f2ef5..73370e0825 100644
--- a/internal/service/chat.go
+++ b/internal/service/chat.go
@@ -200,7 +200,7 @@ type CreateChatRequest struct {
PromptConfig map[string]interface{} `json:"prompt_config"`
Description *string
TopN *int
- PrefetchSize *int
+ RerankCandidatesCount *int
TopK *int
SimilarityThreshold *float64
VectorSimilarityWeight *float64
@@ -300,8 +300,8 @@ func (s *ChatService) Create(ctx context.Context, userID string, req map[string]
if _, ok := req["top_n"]; !ok {
req["top_n"] = 6
}
- if _, ok := req["prefetch_size"]; !ok {
- req["prefetch_size"] = 64
+ if _, ok := req["rerank_candidates_count"]; !ok {
+ req["rerank_candidates_count"] = 64
}
if _, ok := req["top_k"]; !ok {
req["top_k"] = 1024
@@ -505,7 +505,7 @@ func filterCreateChatPersistedFields(req map[string]interface{}) {
persisted := map[string]struct{}{
"name": {}, "description": {}, "icon": {}, "language": {}, "llm_id": {}, "tenant_llm_id": {},
"llm_setting": {}, "prompt_type": {}, "prompt_config": {}, "meta_data_filter": {},
- "similarity_threshold": {}, "vector_similarity_weight": {}, "top_n": {}, "prefetch_size": {}, "top_k": {},
+ "similarity_threshold": {}, "vector_similarity_weight": {}, "top_n": {}, "rerank_candidates_count": {}, "top_k": {},
"do_refer": {}, "rerank_id": {}, "tenant_rerank_id": {}, "kb_ids": {}, "status": {},
}
for key := range req {
@@ -553,7 +553,7 @@ func buildCreateChatEntity(req map[string]interface{}, tenantID string) *entity.
SimilarityThreshold: floatFromValue(req["similarity_threshold"]),
VectorSimilarityWeight: floatFromValue(req["vector_similarity_weight"]),
TopN: int64FromValue(req["top_n"]),
- PrefetchSize: int64FromValue(req["prefetch_size"]),
+ RerankCandidatesCount: int64FromValue(req["rerank_candidates_count"]),
TopK: int64FromValue(req["top_k"]),
DoRefer: stringFromValue(req["do_refer"]),
RerankID: rerankID,
@@ -804,7 +804,7 @@ var chatPersistedFields = map[string]struct{}{
"similarity_threshold": {},
"vector_similarity_weight": {},
"top_n": {},
- "prefetch_size": {},
+ "rerank_candidates_count": {},
"top_k": {},
"do_refer": {},
"rerank_id": {},
@@ -1162,7 +1162,7 @@ func (s *ChatService) buildRESTChatResponse(ctx context.Context, chat *entity.Ch
"similarity_threshold": chat.SimilarityThreshold,
"vector_similarity_weight": chat.VectorSimilarityWeight,
"top_n": chat.TopN,
- "prefetch_size": chat.PrefetchSize,
+ "rerank_candidates_count": chat.RerankCandidatesCount,
"top_k": chat.TopK,
"do_refer": chat.DoRefer,
"rerank_id": chat.RerankID,
diff --git a/internal/service/chat_pipeline.go b/internal/service/chat_pipeline.go
index a17bc04dc7..3ce5492ba9 100644
--- a/internal/service/chat_pipeline.go
+++ b/internal/service/chat_pipeline.go
@@ -613,9 +613,9 @@ func (s *ChatPipelineService) AsyncChat(
"doc_aggs": []interface{}{},
}
var knowledges []string
- prefetchSize := int(chat.PrefetchSize)
- if prefetchSize <= 0 {
- prefetchSize = 64
+ rerankCandidatesCount := int(chat.RerankCandidatesCount)
+ if rerankCandidatesCount <= 0 {
+ rerankCandidatesCount = 64
}
// When hasKnowledgeParam is true, runs (mutually exclusive):
@@ -640,14 +640,14 @@ func (s *ChatPipelineService) AsyncChat(
// KB retrieval callback for the deep researcher
kbRetrieve := func(ctx context.Context, q string) (*nlp.RetrievalResult, error) {
return retSvc.Retrieval(ctx, &nlp.RetrievalRequest{
- Question: q,
- TenantIDs: tenantIDs,
- KbIDs: kbIDs,
- DocIDs: docIDs,
- Page: 1,
- PageSize: int(chat.TopN),
- PrefetchSize: &prefetchSize,
- EmbeddingModel: embModel,
+ Question: q,
+ TenantIDs: tenantIDs,
+ KbIDs: kbIDs,
+ DocIDs: docIDs,
+ Page: 1,
+ PageSize: int(chat.TopN),
+ RerankCandidatesCount: &rerankCandidatesCount,
+ EmbeddingModel: embModel,
})
}
@@ -702,7 +702,7 @@ func (s *ChatPipelineService) AsyncChat(
DocIDs: docIDs,
Page: 1,
PageSize: topN,
- PrefetchSize: &prefetchSize,
+ RerankCandidatesCount: &rerankCandidatesCount,
Top: &top,
SimilarityThreshold: &threshold,
VectorSimilarityWeight: &vsw,
diff --git a/internal/service/chat_session.go b/internal/service/chat_session.go
index 5fbe68f4a7..7c6694f236 100644
--- a/internal/service/chat_session.go
+++ b/internal/service/chat_session.go
@@ -1818,7 +1818,7 @@ func (s *ChatSessionService) buildDefaultCompletionDialog(tenantID string) *enti
PromptConfig: entity.JSONMap{},
KBIDs: entity.JSONSlice{},
TopN: 6,
- PrefetchSize: 64,
+ RerankCandidatesCount: 64,
TopK: 1024,
RerankID: "",
SimilarityThreshold: 0.1,
diff --git a/internal/service/chunk/chunk.go b/internal/service/chunk/chunk.go
index 326a5a7c99..dfb94d48ae 100644
--- a/internal/service/chunk/chunk.go
+++ b/internal/service/chunk/chunk.go
@@ -208,9 +208,9 @@ func (s *ChunkService) RetrievalTest(ctx context.Context, req *service.Retrieval
var chatID string
var chatModelForFilter *models.ChatModel
filter := req.Filter
- prefetchSize := 64
- if req.PrefetchSize != nil {
- prefetchSize = *req.PrefetchSize
+ rerankCandidatesCount := 64
+ if req.RerankCandidatesCount != nil {
+ rerankCandidatesCount = *req.RerankCandidatesCount
}
if req.SearchID != nil && *req.SearchID != "" {
@@ -219,10 +219,10 @@ func (s *ChunkService) RetrievalTest(ctx context.Context, req *service.Retrieval
if err != nil {
common.Warn("Failed to get search detail for search_id, proceeding without it", zap.String("searchID", *req.SearchID), zap.Error(err))
} else if searchConfig, ok := searchConfigMap(searchDetail["search_config"]); ok && searchConfig != nil {
- if req.PrefetchSize == nil || *req.PrefetchSize == 0 {
- prefetchSize = 100
- if configuredPrefetchSize, ok := common.GetInt(searchConfig["prefetch_size"]); ok {
- prefetchSize = configuredPrefetchSize
+ if req.RerankCandidatesCount == nil || *req.RerankCandidatesCount == 0 {
+ rerankCandidatesCount = 100
+ if configuredRerankCandidatesCount, ok := common.GetInt(searchConfig["rerank_candidates_count"]); ok {
+ rerankCandidatesCount = configuredRerankCandidatesCount
}
}
if searchMetaFilter, ok := searchConfigMap(searchConfig["meta_data_filter"]); ok {
@@ -417,7 +417,7 @@ func (s *ChunkService) RetrievalTest(ctx context.Context, req *service.Retrieval
DocIDs: docIDs,
Page: common.CoalesceInt(req.Page, 1),
PageSize: common.CoalesceInt(req.Size, 30),
- PrefetchSize: &prefetchSize,
+ RerankCandidatesCount: &rerankCandidatesCount,
Top: req.TopK,
SimilarityThreshold: req.SimilarityThreshold,
VectorSimilarityWeight: req.VectorSimilarityWeight,
diff --git a/internal/service/chunk_types.go b/internal/service/chunk_types.go
index 3cd143a008..41f2cab058 100644
--- a/internal/service/chunk_types.go
+++ b/internal/service/chunk_types.go
@@ -64,7 +64,7 @@ type RetrievalTestRequest struct {
Question string `json:"question"`
Page *int `json:"page,omitempty"`
Size *int `json:"size,omitempty"`
- PrefetchSize *int `json:"prefetch_size,omitempty"`
+ RerankCandidatesCount *int `json:"rerank_candidates_count,omitempty"`
DocIDs []string `json:"doc_ids,omitempty"`
UseKG *bool `json:"use_kg,omitempty"`
TopK *int `json:"top_k,omitempty"`
diff --git a/internal/service/dataset/search.go b/internal/service/dataset/search.go
index fa040ca663..b77016cc18 100644
--- a/internal/service/dataset/search.go
+++ b/internal/service/dataset/search.go
@@ -38,9 +38,9 @@ func (d *DatasetService) SearchDatasets(ctx context.Context, req *service.Search
if req.Size != nil {
pageSize = *req.Size
}
- prefetchSize := 64
- if req.PrefetchSize != nil {
- prefetchSize = *req.PrefetchSize
+ rerankCandidatesCount := 64
+ if req.RerankCandidatesCount != nil {
+ rerankCandidatesCount = *req.RerankCandidatesCount
}
useKG := false
if req.UseKG != nil {
@@ -131,7 +131,7 @@ func (d *DatasetService) SearchDatasets(ctx context.Context, req *service.Search
}
if searchConfig, ok := searchDetail["search_config"].(map[string]interface{}); ok && searchConfig != nil {
- prefetchSize = 100
+ rerankCandidatesCount = 100
if scMetadataFilter, ok := searchConfig["meta_data_filter"].(map[string]interface{}); ok {
metadataFilter = scMetadataFilter
}
@@ -149,8 +149,8 @@ func (d *DatasetService) SearchDatasets(ctx context.Context, req *service.Search
topK = 2048
}
}
- if scPrefetchSize, ok := common.GetInt(searchConfig["prefetch_size"]); ok {
- prefetchSize = scPrefetchSize
+ if scRerankCandidatesCount, ok := common.GetInt(searchConfig["rerank_candidates_count"]); ok {
+ rerankCandidatesCount = scRerankCandidatesCount
}
if scUseKG, ok := searchConfig["use_kg"].(bool); ok {
useKG = scUseKG
@@ -271,7 +271,7 @@ func (d *DatasetService) SearchDatasets(ctx context.Context, req *service.Search
DocIDs: docIDs,
Page: page,
PageSize: pageSize,
- PrefetchSize: &prefetchSize,
+ RerankCandidatesCount: &rerankCandidatesCount,
Top: &topK,
SimilarityThreshold: &similarityThreshold,
VectorSimilarityWeight: &vectorSimilarityWeight,
diff --git a/internal/service/dataset_types.go b/internal/service/dataset_types.go
index ef3e9e63a6..4ef069c624 100644
--- a/internal/service/dataset_types.go
+++ b/internal/service/dataset_types.go
@@ -41,7 +41,7 @@ type SearchDatasetsRequest struct {
Question string `json:"question" binding:"required"`
Page *int `json:"page,omitempty"`
Size *int `json:"size,omitempty"`
- PrefetchSize *int `json:"prefetch_size,omitempty"`
+ RerankCandidatesCount *int `json:"rerank_candidates_count,omitempty"`
DocIDs []string `json:"doc_ids,omitempty"`
UseKG *bool `json:"use_kg,omitempty"`
TopK *int `json:"top_k,omitempty"`
@@ -69,7 +69,7 @@ type SearchDatasetRequest struct {
Question string `json:"question"`
Page *int `json:"page,omitempty"`
Size *int `json:"size,omitempty"`
- PrefetchSize *int `json:"prefetch_size,omitempty"`
+ RerankCandidatesCount *int `json:"rerank_candidates_count,omitempty"`
DocIDs []string `json:"doc_ids,omitempty"`
UseKG *bool `json:"use_kg,omitempty"`
TopK *int `json:"top_k,omitempty"`
@@ -93,7 +93,7 @@ func (req *SearchDatasetRequest) ToSearchDatasetsRequest(datasetID string) *Sear
Question: req.Question,
Page: req.Page,
Size: req.Size,
- PrefetchSize: req.PrefetchSize,
+ RerankCandidatesCount: req.RerankCandidatesCount,
DocIDs: req.DocIDs,
UseKG: req.UseKG,
TopK: req.TopK,
diff --git a/internal/service/nlp/retrieval.go b/internal/service/nlp/retrieval.go
index 7805c35cf1..613ed83c19 100644
--- a/internal/service/nlp/retrieval.go
+++ b/internal/service/nlp/retrieval.go
@@ -53,7 +53,7 @@ type RetrievalRequest struct {
DocIDs []string
Page int
PageSize int
- PrefetchSize *int
+ RerankCandidatesCount *int
Top *int
SimilarityThreshold *float64
VectorSimilarityWeight *float64
@@ -73,10 +73,10 @@ type RetrievalResult struct {
}
// Retrieval performs hybrid search + reranking + pagination
-// - Prefetch candidates for reranking
+// - Retrieve candidates for reranking
// - Perform reranking via Rerank()
// - Sort indices by score descending and filter by threshold
-// - Require Page * PageSize to fit within PrefetchSize
+// - Require Page * PageSize to fit within RerankCandidatesCount
// - Support only the first page when a rerank model is configured
// - Calculate pagination to extract actual page returned from reranked results
// - Build chunks
@@ -110,19 +110,19 @@ func (s *RetrievalService) Retrieval(ctx context.Context, req *RetrievalRequest)
if req.PageSize <= 0 {
req.PageSize = 1
}
- if req.PrefetchSize == nil {
- req.PrefetchSize = func() *int { v := 64; return &v }()
+ if req.RerankCandidatesCount == nil {
+ req.RerankCandidatesCount = func() *int { v := 64; return &v }()
}
pageSize := req.PageSize
- prefetchSize := *req.PrefetchSize
- if prefetchSize <= 0 || req.Page > prefetchSize/pageSize {
- return nil, fmt.Errorf("prefetch_size(%d) must be greater than or equal to page(%d) * page_size(%d)", prefetchSize, req.Page, pageSize)
+ rerankCandidatesCount := *req.RerankCandidatesCount
+ if rerankCandidatesCount <= 0 || req.Page > rerankCandidatesCount/pageSize {
+ return nil, fmt.Errorf("rerank_candidates_count(%d) must be greater than or equal to page(%d) * page_size(%d)", rerankCandidatesCount, req.Page, pageSize)
}
if req.RerankModel != nil && req.Page != 1 {
return nil, fmt.Errorf("Pagination is not supported when rerank_mdl is specified. Please set page=1 to retrieve the top %d results.", pageSize)
}
- common.Debug("Retrieval prefetch params", zap.Int("page", req.Page), zap.Int("pageSize", pageSize), zap.Int("prefetchSize", prefetchSize))
+ common.Debug("Retrieval rerank candidate params", zap.Int("page", req.Page), zap.Int("pageSize", pageSize), zap.Int("rerankCandidatesCount", rerankCandidatesCount))
// Execute search via Search()
searchReq := &RetrievalSearchRequest{
@@ -131,7 +131,7 @@ func (s *RetrievalService) Retrieval(ctx context.Context, req *RetrievalRequest)
KbIDs: req.KbIDs,
DocIDs: req.DocIDs,
Page: 1,
- PageSize: prefetchSize,
+ PageSize: rerankCandidatesCount,
Top: *req.Top,
RankFeature: *req.RankFeature,
EmbeddingModel: req.EmbeddingModel,
diff --git a/internal/service/nlp/retrieval_test.go b/internal/service/nlp/retrieval_test.go
index 75f31e1126..9cb3df9fbb 100644
--- a/internal/service/nlp/retrieval_test.go
+++ b/internal/service/nlp/retrieval_test.go
@@ -14,7 +14,7 @@ import (
"gorm.io/gorm"
)
-func TestRetrievalUsesPrefetchSizeAsCandidateSet(t *testing.T) {
+func TestRetrievalUsesRerankCandidatesCountAsCandidateSet(t *testing.T) {
oldQueryBuilder := globalQueryBuilder
globalQueryBuilder = NewQueryBuilder()
defer func() { globalQueryBuilder = oldQueryBuilder }()
@@ -34,7 +34,7 @@ func TestRetrievalUsesPrefetchSizeAsCandidateSet(t *testing.T) {
threshold := 0.5
vectorWeight := 1.0
aggs := false
- prefetchSize := 70
+ rerankCandidatesCount := 70
result, err := service.Retrieval(context.Background(), &RetrievalRequest{
Question: "alpha",
@@ -46,7 +46,7 @@ func TestRetrievalUsesPrefetchSizeAsCandidateSet(t *testing.T) {
VectorSimilarityWeight: &vectorWeight,
Aggs: &aggs,
Filter: map[string]interface{}{"must_not": map[string]interface{}{"exists": "compile_kwd"}},
- PrefetchSize: &prefetchSize,
+ RerankCandidatesCount: &rerankCandidatesCount,
})
if err != nil {
t.Fatalf("Retrieval failed: %v", err)
@@ -57,8 +57,8 @@ func TestRetrievalUsesPrefetchSizeAsCandidateSet(t *testing.T) {
if result.Total != 70 {
t.Fatalf("total = %d, want 70", result.Total)
}
- if len(engine.searchLimits) != 1 || engine.searchLimits[0] != prefetchSize {
- t.Fatalf("search limits = %v, want [%d]", engine.searchLimits, prefetchSize)
+ if len(engine.searchLimits) != 1 || engine.searchLimits[0] != rerankCandidatesCount {
+ t.Fatalf("search limits = %v, want [%d]", engine.searchLimits, rerankCandidatesCount)
}
for _, filters := range engine.searchFilters {
mustNot, ok := filters["must_not"].(map[string]interface{})
diff --git a/internal/service/search.go b/internal/service/search.go
index 4da8ce1b84..42100719f3 100644
--- a/internal/service/search.go
+++ b/internal/service/search.go
@@ -435,8 +435,8 @@ func askOptionsFromSearchConfig(searchID string, searchConfig map[string]interfa
if value, ok := intFromSearchConfig(searchConfig["top_k"]); ok {
opts.TopK = &value
}
- if value, ok := intFromSearchConfig(searchConfig["prefetch_size"]); ok {
- opts.PrefetchSize = &value
+ if value, ok := intFromSearchConfig(searchConfig["rerank_candidates_count"]); ok {
+ opts.RerankCandidatesCount = &value
}
if value, ok := searchConfigMapValue(searchConfig["meta_data_filter"]); ok {
opts.Filter = value
diff --git a/rag/nlp/search.py b/rag/nlp/search.py
index a84192d413..dd1e6a6be2 100644
--- a/rag/nlp/search.py
+++ b/rag/nlp/search.py
@@ -569,16 +569,16 @@ class Dealer:
rank_feature: dict | None = {PAGERANK_FLD: 10},
trace_id=None,
must_not: dict | None = None,
- prefetch_size=64,
+ rerank_candidates_count=64,
):
"""
Pagination is neither efficient nor reliable for this retrieval when rerank is enabled because the system must:
- - Prefetch more records than the requested page_size.
+ - Retrieve more rerank candidates than the requested page_size.
- Rerank those records to calculate similarity scores.
- Filter out records below than the similarity threshold.
When requesting page 2, the system must still process all candidates needed for page 1,
resulting in a significant waste of time and computational resources. (without cache)
- Moreover, when prefetch_size expands into the next retrieval window, new records are added to the candidate set and the entire set is reranked.
+ Moreover, when rerank_candidates_count expands into the next retrieval window, new records are added to the candidate set and the entire set is reranked.
That meant the previous returned pages might not be the same as the current returned pages, which is not acceptable for pagination.
"""
ranks = {"total": 0, "chunks": [], "doc_aggs": {}}
@@ -586,17 +586,17 @@ class Dealer:
return ranks
page = max(page, 1)
- if page * page_size > prefetch_size:
- raise Exception(f"prefetch_size({prefetch_size}) must be greater than page * page_size({page * page_size}) to ensure correct pagination.")
+ if page * page_size > rerank_candidates_count:
+ raise Exception(f"rerank_candidates_count({rerank_candidates_count}) must be greater than page * page_size({page * page_size}) to ensure correct pagination.")
if rerank_mdl is not None and page != 1:
raise Exception(f"Pagination is not supported when rerank_mdl is specified. Please set page=1 to retrieve the top {page_size} results.")
- prefetch_page = 1
+ rerank_candidates_page = 1
req = {
"kb_ids": kb_ids,
"doc_ids": doc_ids,
- "page": prefetch_page,
- "size": prefetch_size,
+ "page": rerank_candidates_page,
+ "size": rerank_candidates_count,
"question": question,
"vector": True,
"topk": top,
@@ -606,7 +606,7 @@ class Dealer:
}
if isinstance(must_not, dict) and must_not:
req["must_not"] = must_not
- logging.debug(f"[Search] page={page}, page_size={page_size}, prefetch_size={prefetch_size}")
+ logging.debug(f"[Search] page={page}, page_size={page_size}, rerank_candidates_count={rerank_candidates_count}")
if isinstance(tenant_ids, str):
tenant_ids = tenant_ids.split(",")
diff --git a/test/unit_test/api/db/services/test_gaussdb_dialog_sql.py b/test/unit_test/api/db/services/test_gaussdb_dialog_sql.py
index e510d36430..67dddce3be 100644
--- a/test/unit_test/api/db/services/test_gaussdb_dialog_sql.py
+++ b/test/unit_test/api/db/services/test_gaussdb_dialog_sql.py
@@ -1696,7 +1696,7 @@ def test_tc_sql_1001_use_sql_none_falls_back_to_retrieval(
aggs=True,
rerank_mdl=None,
rank_feature=None,
- prefetch_size=64,
+ rerank_candidates_count=64,
)
assert results[-1]["answer"] == "fallback answer"
assert results[-1]["reference"] == _expected_fallback_reference(kb_id)
@@ -1745,7 +1745,7 @@ def test_tc_sql_1002_validator_rejection_falls_back_to_retrieval(
aggs=True,
rerank_mdl=None,
rank_feature=None,
- prefetch_size=64,
+ rerank_candidates_count=64,
)
assert results[-1]["answer"] == "fallback answer"
assert results[-1]["reference"] == _expected_fallback_reference(kb_id)
@@ -1796,7 +1796,7 @@ def test_tc_sql_1003_sql_timeout_falls_back_to_retrieval(
aggs=True,
rerank_mdl=None,
rank_feature=None,
- prefetch_size=64,
+ rerank_candidates_count=64,
)
assert results[-1]["answer"] == "fallback answer"
assert results[-1]["reference"] == _expected_fallback_reference(kb_id)
diff --git a/test/unit_test/rag/test_search_pagination.py b/test/unit_test/rag/test_search_pagination.py
index 230eb58534..85980f0035 100644
--- a/test/unit_test/rag/test_search_pagination.py
+++ b/test/unit_test/rag/test_search_pagination.py
@@ -13,7 +13,7 @@
# See the License for the specific language governing permissions and
# limitations under the License.
#
-"""Unit tests for prefetch-based pagination in Dealer.retrieval."""
+"""Unit tests for rerank-candidate pagination in Dealer.retrieval."""
import sys
import types
@@ -55,13 +55,13 @@ def _search_result(total):
@pytest.mark.asyncio
[email protected](("page", "page_size", "prefetch_size"), [(1, 10, 64), (6, 10, 64), (7, 10, 70)])
-async def test_retrieval_prefetches_once_and_slices_requested_page(monkeypatch, page, page_size, prefetch_size):
[email protected](("page", "page_size", "rerank_candidates_count"), [(1, 10, 64), (6, 10, 64), (7, 10, 70)])
+async def test_retrieval_fetches_rerank_candidates_once_and_slices_requested_page(monkeypatch, page, page_size, rerank_candidates_count):
requests = []
async def fake_search(self, req, *args, **kwargs):
requests.append(req)
- return _search_result(prefetch_size)
+ return _search_result(rerank_candidates_count)
async def keep_all_chunks(self, search_result):
return search_result
@@ -82,19 +82,19 @@ async def test_retrieval_prefetches_once_and_slices_requested_page(monkeypatch,
page_size=page_size,
similarity_threshold=0.0,
aggs=False,
- prefetch_size=prefetch_size,
+ rerank_candidates_count=rerank_candidates_count,
)
assert len(requests) == 1
assert requests[0]["page"] == 1
- assert requests[0]["size"] == prefetch_size
+ assert requests[0]["size"] == rerank_candidates_count
begin = (page - 1) * page_size
assert [chunk["chunk_id"] for chunk in ranks["chunks"]] == [f"chunk-{i}" for i in range(begin, begin + page_size)]
@pytest.mark.asyncio
-async def test_retrieval_rejects_page_beyond_prefetch_size():
- with pytest.raises(Exception, match=r"prefetch_size\(64\) must be greater than page \* page_size\(70\)"):
+async def test_retrieval_rejects_page_beyond_rerank_candidates_count():
+ with pytest.raises(Exception, match=r"rerank_candidates_count\(64\) must be greater than page \* page_size\(70\)"):
await Dealer.__new__(Dealer).retrieval(
question="question",
embd_mdl=object(),
@@ -102,7 +102,7 @@ async def test_retrieval_rejects_page_beyond_prefetch_size():
kb_ids=["kb-1"],
page=7,
page_size=10,
- prefetch_size=64,
+ rerank_candidates_count=64,
)
@@ -117,5 +117,5 @@ async def test_retrieval_rejects_pagination_with_reranker():
page=2,
page_size=10,
rerank_mdl=object(),
- prefetch_size=64,
+ rerank_candidates_count=64,
)
diff --git a/web/src/components/prefetch-size-item.tsx b/web/src/components/rerank-candidates-count-item.tsx
similarity index 74%
rename from web/src/components/prefetch-size-item.tsx
rename to web/src/components/rerank-candidates-count-item.tsx
index b991c51882..587c2b869c 100644
--- a/web/src/components/prefetch-size-item.tsx
+++ b/web/src/components/rerank-candidates-count-item.tsx
@@ -19,26 +19,26 @@ import { useTranslation } from 'react-i18next';
import { z } from 'zod';
import { SliderInputFormField } from './slider-input-form-field';
-export const prefetchSizeSchema = {
- prefetch_size: z.number().int().min(64).max(256),
+export const rerankCandidatesCountSchema = {
+ rerank_candidates_count: z.number().int().min(64).max(256),
};
-interface PrefetchSizeFormFieldProps {
+interface RerankCandidatesCountFormFieldProps {
defaultValue?: number;
name?: string;
}
-export function PrefetchSizeFormField({
+export function RerankCandidatesCountFormField({
defaultValue = 64,
- name = 'prefetch_size',
-}: PrefetchSizeFormFieldProps) {
+ name = 'rerank_candidates_count',
+}: RerankCandidatesCountFormFieldProps) {
const { t } = useTranslation();
return (
<SliderInputFormField
name={name}
- label={t('chat.prefetchSize')}
- tooltip={t('chat.prefetchSizeTip')}
+ label={t('chat.rerankCandidatesCount')}
+ tooltip={t('chat.rerankCandidatesCountTip')}
min={64}
max={256}
defaultValue={defaultValue}
diff --git a/web/src/interfaces/database/agent.ts b/web/src/interfaces/database/agent.ts
index 7a690d9dce..52064211f1 100644
--- a/web/src/interfaces/database/agent.ts
+++ b/web/src/interfaces/database/agent.ts
@@ -165,7 +165,7 @@ export interface IRetrievalForm {
similarity_threshold?: number;
keywords_similarity_weight?: number;
top_n?: number;
- prefetch_size?: number;
+ rerank_candidates_count?: number;
top_k?: number;
rerank_id?: string;
tenant_rerank_id?: string;
diff --git a/web/src/interfaces/database/chat.ts b/web/src/interfaces/database/chat.ts
index c6482c8c52..c636fdcc02 100644
--- a/web/src/interfaces/database/chat.ts
+++ b/web/src/interfaces/database/chat.ts
@@ -80,7 +80,7 @@ export interface IDialog {
similarity_threshold: number;
top_k: number;
top_n: number;
- prefetch_size: number;
+ rerank_candidates_count: number;
rerank_id?: string;
tenant_rerank_id?: string;
meta_data_filter: MetaDataFilter;
diff --git a/web/src/interfaces/request/knowledge.ts b/web/src/interfaces/request/knowledge.ts
index 6b0193b4b8..498fc54328 100644
--- a/web/src/interfaces/request/knowledge.ts
+++ b/web/src/interfaces/request/knowledge.ts
@@ -1,7 +1,7 @@
export interface ITestRetrievalRequestBody {
question: string;
size: number;
- prefetch_size: number;
+ rerank_candidates_count: number;
similarity_threshold: number;
vector_similarity_weight: number;
rerank_id?: string;
diff --git a/web/src/locales/en.ts b/web/src/locales/en.ts
index c40bdec028..433b89d446 100644
--- a/web/src/locales/en.ts
+++ b/web/src/locales/en.ts
@@ -1067,11 +1067,11 @@ This auto-tagging feature enhances retrieval by adding another layer of domain-s
'Your prompts or instructions for the LLM, including but not limited to its role, the desired length, tone, and language of its answers. If your model has native support for reasoning, you can add //no_thinking add the prompt to stop reasoning.',
topN: 'Top N',
topNTip: `Not all chunks with similarity score above the 'similarity threshold' will be sent to the LLM. This selects 'Top N' chunks from the retrieved ones.`,
- prefetchSize: 'Prefetch size',
- prefetchSizeTip:
- 'The number of candidate chunks retrieved before the next processing step.',
- prefetchSizeValidation:
- 'Prefetch size must be greater than or equal to Top N.',
+ rerankCandidatesCount: 'Rerank candidates',
+ rerankCandidatesCountTip:
+ 'The number of candidate chunks retrieved for reranking.',
+ rerankCandidatesCountValidation:
+ 'Rerank candidates must be greater than or equal to Top N.',
variable: 'Variable',
variableTip: `Used together with RAGFlow's chat assistant management APIs, variables can help develop more flexible system prompt strategies. The defined variables will be used by 'System prompt' as part of the prompts for the LLM. {knowledge} is a reserved special variable representing chunks retrieved from specified dataset(s), and all variables should be enclosed in curly braces {} in the 'System prompt'. See https://ragflow.io/docs/dev/set_chat_variables for details.`,
add: 'Add',
diff --git a/web/src/locales/zh.ts b/web/src/locales/zh.ts
index 721c738b62..37e096a364 100644
--- a/web/src/locales/zh.ts
+++ b/web/src/locales/zh.ts
@@ -961,9 +961,9 @@ NER:使用 spaCy NER 和基于规则的关键词提取来抽取 Entities 和 R
'当LLM回答问题时,你需要LLM遵循的说明,比如角色设计、答案长度和答案语言等。如果您的模型原生支持在问答中推理,可以通过 //no_thinking 关闭自动推理。',
topN: 'Top N',
topNTip: `并非所有相似度得分高于“相似度阈值”的块都会被提供给大语言模型。 LLM 只能看到这些“Top N”块。`,
- prefetchSize: '预取数量',
- prefetchSizeTip: '在进行下一步计算之前检索的预取候选分块数量。',
- prefetchSizeValidation: '预取数量必须大于或等于 Top N。',
+ rerankCandidatesCount: '重排候选数',
+ rerankCandidatesCountTip: '用于重排的候选分块数量。',
+ rerankCandidatesCountValidation: '重排候选数必须大于或等于 Top N。',
variable: '变量',
variableTip: `你可以通过对话 API,并配合变量设置来动态调整大模型的系统提示词。
{knowledge}为系统预留变量,代表从指定知识库召回的文本块。
diff --git a/web/src/pages/agent/constant/index.tsx b/web/src/pages/agent/constant/index.tsx
index 583e26a4d2..1d9da46689 100644
--- a/web/src/pages/agent/constant/index.tsx
+++ b/web/src/pages/agent/constant/index.tsx
@@ -87,7 +87,7 @@ export enum RetrievalFrom {
export const initialRetrievalValues = {
query: AgentGlobalsSysQueryWithBrace,
top_n: 8,
- prefetch_size: 64,
+ rerank_candidates_count: 64,
top_k: 1024,
kb_ids: [],
rerank_id: '',
diff --git a/web/src/pages/agent/form/retrieval-form/next.tsx b/web/src/pages/agent/form/retrieval-form/next.tsx
index c2f0dfb6c0..3e64f191d9 100644
--- a/web/src/pages/agent/form/retrieval-form/next.tsx
+++ b/web/src/pages/agent/form/retrieval-form/next.tsx
@@ -10,9 +10,9 @@ import { RAGFlowFormItem } from '@/components/ragflow-form';
import { RerankFormFields } from '@/components/rerank';
import { SimilaritySliderFormField } from '@/components/similarity-slider';
import {
- PrefetchSizeFormField,
- prefetchSizeSchema,
-} from '@/components/prefetch-size-item';
+ RerankCandidatesCountFormField,
+ rerankCandidatesCountSchema,
+} from '@/components/rerank-candidates-count-item';
import { TopNFormField } from '@/components/top-n-item';
import {
@@ -54,7 +54,7 @@ export const RetrievalPartialSchema = {
similarity_threshold: z.coerce.number(),
keywords_similarity_weight: z.coerce.number().min(0).max(1),
top_n: z.coerce.number(),
- ...prefetchSizeSchema,
+ ...rerankCandidatesCountSchema,
top_k: z.coerce.number(),
dataset_ids: z.array(z.string()),
rerank_id: z.string(),
@@ -190,7 +190,7 @@ function RetrievalForm({ node }: INextOperatorForm) {
similarityWeightType="keyword"
isTooltipShown
></SimilaritySliderFormField>
- <PrefetchSizeFormField></PrefetchSizeFormField>
+ <RerankCandidatesCountFormField></RerankCandidatesCountFormField>
<TopNFormField></TopNFormField>
{hideKnowledgeGraphField || (
<>
diff --git a/web/src/pages/agent/form/tool-form/retrieval-form/index.tsx b/web/src/pages/agent/form/tool-form/retrieval-form/index.tsx
index 40f3239140..bc3c6a9dc4 100644
--- a/web/src/pages/agent/form/tool-form/retrieval-form/index.tsx
+++ b/web/src/pages/agent/form/tool-form/retrieval-form/index.tsx
@@ -4,7 +4,7 @@ import { FormContainer } from '@/components/form-container';
import { MetadataFilter } from '@/components/metadata-filter';
import { RerankFormFields } from '@/components/rerank';
import { SimilaritySliderFormField } from '@/components/similarity-slider';
-import { PrefetchSizeFormField } from '@/components/prefetch-size-item';
+import { RerankCandidatesCountFormField } from '@/components/rerank-candidates-count-item';
import { TopNFormField } from '@/components/top-n-item';
import { Form } from '@/components/ui/form';
@@ -68,7 +68,7 @@ const RetrievalForm = () => {
similarityWeightType="keyword"
isTooltipShown
></SimilaritySliderFormField>
- <PrefetchSizeFormField></PrefetchSizeFormField>
+ <RerankCandidatesCountFormField></RerankCandidatesCountFormField>
<TopNFormField></TopNFormField>
{hideKnowledgeGraphField || (
<>
diff --git a/web/src/pages/dataset/testing/testing-form.tsx b/web/src/pages/dataset/testing/testing-form.tsx
index cf884e17ad..f4fbf53afe 100644
--- a/web/src/pages/dataset/testing/testing-form.tsx
+++ b/web/src/pages/dataset/testing/testing-form.tsx
@@ -7,9 +7,9 @@ import { z } from 'zod';
import { CrossLanguageFormField } from '@/components/cross-language-form-field';
import { FormContainer } from '@/components/form-container';
import {
- PrefetchSizeFormField,
- prefetchSizeSchema,
-} from '@/components/prefetch-size-item';
+ RerankCandidatesCountFormField,
+ rerankCandidatesCountSchema,
+} from '@/components/rerank-candidates-count-item';
import {
MetadataFilter,
MetadataFilterSchema,
@@ -71,11 +71,11 @@ export default function TestingForm({
dataset_ids: z.array(z.string()).optional(),
...MetadataFilterSchema,
size: z.number().int().min(1).max(100),
- ...prefetchSizeSchema,
+ ...rerankCandidatesCountSchema,
})
- .refine((values) => values.prefetch_size >= values.size, {
- message: t('chat.prefetchSizeValidation'),
- path: ['prefetch_size'],
+ .refine((values) => values.rerank_candidates_count >= values.size, {
+ message: t('chat.rerankCandidatesCountValidation'),
+ path: ['rerank_candidates_count'],
});
const form = useForm<z.infer<typeof formSchema>>({
@@ -86,7 +86,7 @@ export default function TestingForm({
...initialTopKValue,
dataset_ids: [knowledgeBaseId],
size: 10,
- prefetch_size: 64,
+ rerank_candidates_count: 64,
},
});
@@ -120,7 +120,7 @@ export default function TestingForm({
name={'cross_languages'}
></CrossLanguageFormField>
<MetadataFilter prefix=""></MetadataFilter>
- <PrefetchSizeFormField></PrefetchSizeFormField>
+ <RerankCandidatesCountFormField></RerankCandidatesCountFormField>
<TopSelectFormItem></TopSelectFormItem>
</FormContainer>
</div>
diff --git a/web/src/pages/next-chats/chat/app-settings/chat-prompt-engine.tsx b/web/src/pages/next-chats/chat/app-settings/chat-prompt-engine.tsx
index 63f1db5b00..f827236b06 100644
--- a/web/src/pages/next-chats/chat/app-settings/chat-prompt-engine.tsx
+++ b/web/src/pages/next-chats/chat/app-settings/chat-prompt-engine.tsx
@@ -3,7 +3,7 @@
import { Collapse } from '@/components/collapse';
import { CrossLanguageFormField } from '@/components/cross-language-form-field';
import { MetadataFilter } from '@/components/metadata-filter';
-import { PrefetchSizeFormField } from '@/components/prefetch-size-item';
+import { RerankCandidatesCountFormField } from '@/components/rerank-candidates-count-item';
import { RerankFormFields } from '@/components/rerank';
import { SimilaritySliderFormField } from '@/components/similarity-slider';
import { SwitchFormField } from '@/components/switch-fom-field';
@@ -208,9 +208,9 @@ export function ChatPromptEngine({ prefix = '' }: ChatPromptEngineProps) {
similarityName={prefixName(prefix, 'similarity_threshold')}
similarityWeightName={prefixName(prefix, 'vector_similarity_weight')}
></SimilaritySliderFormField>
- <PrefetchSizeFormField
- name={prefixName(prefix, 'prefetch_size')}
- ></PrefetchSizeFormField>
+ <RerankCandidatesCountFormField
+ name={prefixName(prefix, 'rerank_candidates_count')}
+ ></RerankCandidatesCountFormField>
<TopNFormField name={prefixName(prefix, 'top_n')}></TopNFormField>
<SwitchFormField
diff --git a/web/src/pages/next-chats/chat/app-settings/chat-settings.tsx b/web/src/pages/next-chats/chat/app-settings/chat-settings.tsx
index 56c9b46443..9e2315a16b 100644
--- a/web/src/pages/next-chats/chat/app-settings/chat-settings.tsx
+++ b/web/src/pages/next-chats/chat/app-settings/chat-settings.tsx
@@ -72,7 +72,7 @@ export function ChatSettings({ hasSingleChatBox }: ChatSettingsProps) {
},
},
top_n: 8,
- prefetch_size: 64,
+ rerank_candidates_count: 64,
similarity_threshold: 0.2,
vector_similarity_weight: 0.2,
top_k: 1024,
diff --git a/web/src/pages/next-chats/chat/app-settings/use-chat-setting-schema.tsx b/web/src/pages/next-chats/chat/app-settings/use-chat-setting-schema.tsx
index a575b1f3bf..2e3cea0951 100644
--- a/web/src/pages/next-chats/chat/app-settings/use-chat-setting-schema.tsx
+++ b/web/src/pages/next-chats/chat/app-settings/use-chat-setting-schema.tsx
@@ -3,7 +3,7 @@ import {
LlmSettingFieldSchema,
} from '@/components/llm-setting-items/next';
import { MetadataFilterSchema } from '@/components/metadata-filter';
-import { prefetchSizeSchema } from '@/components/prefetch-size-item';
+import { rerankCandidatesCountSchema } from '@/components/rerank-candidates-count-item';
import { rerankFormSchema } from '@/components/rerank';
import {
similarityThresholdSchema,
@@ -72,7 +72,7 @@ export function useChatSettingSchema() {
...vectorSimilarityWeightSchema,
...similarityThresholdSchema,
...topnSchema,
- ...prefetchSizeSchema,
+ ...rerankCandidatesCountSchema,
...MetadataFilterSchema,
})
.superRefine((value, ctx) => {
diff --git a/web/src/pages/next-chats/hooks/use-create-chat.ts b/web/src/pages/next-chats/hooks/use-create-chat.ts
index 574d36857d..0781ab68b2 100644
--- a/web/src/pages/next-chats/hooks/use-create-chat.ts
+++ b/web/src/pages/next-chats/hooks/use-create-chat.ts
@@ -44,7 +44,7 @@ export const useCreateChatDialog = () => {
similarity_threshold: 0.2,
vector_similarity_weight: 0.3,
top_n: 8,
- prefetch_size: 64,
+ rerank_candidates_count: 64,
top_k: 1024,
}),
[t, defaultModelDictionary?.llm_id],
diff --git a/web/src/pages/next-search/search-setting-hooks.ts b/web/src/pages/next-search/search-setting-hooks.ts
index d536f9ea02..5435ceed0d 100644
--- a/web/src/pages/next-search/search-setting-hooks.ts
+++ b/web/src/pages/next-search/search-setting-hooks.ts
@@ -23,7 +23,7 @@ import {
} from '@/components/llm-setting-items/next';
import { MetadataFilterSchema } from '@/components/metadata-filter';
import { ModelTypeMap } from '@/components/model-tree-select';
-import { prefetchSizeSchema } from '@/components/prefetch-size-item';
+import { rerankCandidatesCountSchema } from '@/components/rerank-candidates-count-item';
import { useModelValidIds } from '@/hooks/use-llm-request';
import { useEffect, useMemo } from 'react';
import { Control, UseFormTrigger, useWatch } from 'react-hook-form';
@@ -45,7 +45,7 @@ export const SearchSettingFormSchema = z
rerank_id: z.string(),
use_rerank: z.boolean(),
top_k: z.number(),
- ...prefetchSizeSchema,
+ ...rerankCandidatesCountSchema,
summary: z.boolean(),
llm_setting: z.object({ ...LlmSettingFieldSchema, ...LLMIdFormField }),
related_search: z.boolean(),
diff --git a/web/src/pages/next-search/search-setting.tsx b/web/src/pages/next-search/search-setting.tsx
index 6f3ba5793d..c505c6cdf2 100644
--- a/web/src/pages/next-search/search-setting.tsx
+++ b/web/src/pages/next-search/search-setting.tsx
@@ -20,7 +20,7 @@ import AvatarNameDescription from '@/components/avatar-name-description';
import { KnowledgeBaseFormField } from '@/components/knowledge-base-item';
import { LlmSettingFieldItems } from '@/components/llm-setting-items/next';
import { MetadataFilter } from '@/components/metadata-filter';
-import { PrefetchSizeFormField } from '@/components/prefetch-size-item';
+import { RerankCandidatesCountFormField } from '@/components/rerank-candidates-count-item';
import { SimilaritySliderFormField } from '@/components/similarity-slider';
import { Button } from '@/components/ui/button';
import {
@@ -111,7 +111,7 @@ function SearchSetting({
rerank_id: search_config?.rerank_id || '',
use_rerank: search_config?.rerank_id ? true : false,
top_k: search_config?.top_k || 1024,
- prefetch_size: search_config?.prefetch_size ?? 100,
+ rerank_candidates_count: search_config?.rerank_candidates_count ?? 100,
summary: search_config?.summary || false,
chat_id: search_config?.chat_id || '',
llm_setting: {
@@ -401,10 +401,10 @@ function SearchSetting({
similarityWeightName="search_config.vector_similarity_weight"
numberInputClassName="rounded-sm"
></SimilaritySliderFormField>
- <PrefetchSizeFormField
- name="search_config.prefetch_size"
+ <RerankCandidatesCountFormField
+ name="search_config.rerank_candidates_count"
defaultValue={100}
- ></PrefetchSizeFormField>
+ ></RerankCandidatesCountFormField>
{/* Rerank Model */}
<FormField
control={formMethods.control}
diff --git a/web/src/pages/next-search/search-view.tsx b/web/src/pages/next-search/search-view.tsx
index e96c6b5d45..171dcfa905 100644
--- a/web/src/pages/next-search/search-view.tsx
+++ b/web/src/pages/next-search/search-view.tsx
@@ -231,7 +231,7 @@ export default function SearchingView({
</div>
<div className="w-44">
<TopSelect
- max={searchData.search_config.prefetch_size ?? 100}
+ max={searchData.search_config.rerank_candidates_count ?? 100}
value={pageSize}
onChange={handleTopChange}
></TopSelect>
diff --git a/web/src/pages/next-searches/hooks.ts b/web/src/pages/next-searches/hooks.ts
index 9cf988c767..caf8f6018e 100644
--- a/web/src/pages/next-searches/hooks.ts
+++ b/web/src/pages/next-searches/hooks.ts
@@ -203,7 +203,7 @@ export interface ISearchAppDetailProps {
summary: boolean;
llm_setting: IllmSettingProps & IllmSettingEnableProps;
top_k: number;
- prefetch_size: number;
+ rerank_candidates_count: number;
use_kg: boolean;
vector_similarity_weight: number;
web_search: boolean;