Revert "server: Consolidate embedding truncation in runner (#12730)" (#12810)

This reverts commit 5d347f6d6f.
This commit is contained in:
Patrick Devine
2025-10-28 14:49:14 -07:00
committed by GitHub
parent 3d99d9779a
commit 29f63f37c8
6 changed files with 84 additions and 264 deletions

View File

@@ -709,13 +709,13 @@ func (s *Server) embeddings(w http.ResponseWriter, r *http.Request) {
seq, err := s.NewSequence(req.Content, nil, NewSequenceParams{
embedding: true,
truncate: req.Truncate,
// TODO (jmorganca): this should be provided by the server via the
// request options and truncated here in the runner, instead of relying on
// the server's truncate logic
truncate: true,
})
if err != nil {
if errors.Is(err, errorInputTooLong) {
http.Error(w, err.Error(), http.StatusBadRequest)
return
}
http.Error(w, fmt.Sprintf("Failed to create new sequence: %v", err), http.StatusInternalServerError)
return
}
@@ -758,8 +758,7 @@ func (s *Server) embeddings(w http.ResponseWriter, r *http.Request) {
embedding := <-seq.embedding
if err := json.NewEncoder(w).Encode(&llm.EmbeddingResponse{
Embedding: embedding,
PromptEvalCount: seq.numPromptInputs,
Embedding: embedding,
}); err != nil {
http.Error(w, fmt.Sprintf("failed to encode response: %v", err), http.StatusInternalServerError)
}

View File

@@ -948,13 +948,13 @@ func (s *Server) embeddings(w http.ResponseWriter, r *http.Request) {
w.Header().Set("Content-Type", "application/json")
seq, err := s.NewSequence(req.Content, nil, NewSequenceParams{
embedding: true,
truncate: req.Truncate,
// TODO (jmorganca): this should be provided by the server via the
// request options and truncated here in the runner, instead of relying on
// the server's truncate logic
truncate: true,
})
if err != nil {
if errors.Is(err, errorInputTooLong) {
http.Error(w, err.Error(), http.StatusBadRequest)
return
}
http.Error(w, fmt.Sprintf("failed to create new sequence: %v", err), http.StatusInternalServerError)
return
}
@@ -995,8 +995,7 @@ func (s *Server) embeddings(w http.ResponseWriter, r *http.Request) {
}
if err := json.NewEncoder(w).Encode(&llm.EmbeddingResponse{
Embedding: <-seq.embedding,
PromptEvalCount: seq.numPromptInputs,
Embedding: <-seq.embedding,
}); err != nil {
http.Error(w, fmt.Sprintf("failed to encode response: %v", err), http.StatusInternalServerError)
}