nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 27 Packages Projects Releases Wiki Activity

server : honor --embd-normalize CLI arg (#23125)

Browse Source 
The --embd-normalize flag was registered only for the embedding and debug
examples, so llama-server rejected it and the /embedding handler used a
hard-coded default of 2 (L2). Add LLAMA_EXAMPLE_SERVER to the flag's
example set and read params.embd_normalize as the handler's default. The
per-request "embd_normalize" body field continues to override.
This commit is contained in:
Rares Vernica
2026-05-17 09:39:04 +03:00
committed by GitHub
parent a16cce81d3
commit 1a68ec9378
2 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
common/arg.cpp
+1 -1
View File
@@ -2808,7 +2808,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
[](common_params & params, int value) { [](common_params & params, int value) {
params.embd_normalize = value; params.embd_normalize = value;
} }
).set_examples({LLAMA_EXAMPLE_EMBEDDING, LLAMA_EXAMPLE_DEBUG})); ).set_examples({LLAMA_EXAMPLE_EMBEDDING, LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_DEBUG}));
add_opt(common_arg( add_opt(common_arg(
{"--embd-output-format"}, "FORMAT", {"--embd-output-format"}, "FORMAT",
"empty = default, \"array\" = [[],[]...], \"json\" = openai style, \"json+\" = same \"json\" + cosine similarity matrix, \"raw\" = plain whitespace-delimited output (one embedding per line)", "empty = default, \"array\" = [[],[]...], \"json\" = openai style, \"json+\" = same \"json\" + cosine similarity matrix, \"raw\" = plain whitespace-delimited output (one embedding per line)",
tools/server/server-context.cpp
+1 -1
View File
@@ -4527,7 +4527,7 @@ std::unique_ptr<server_res_generator> server_routes::handle_embeddings_impl(cons
} }
} }
int embd_normalize = 2; // default to Euclidean/L2 norm int embd_normalize = params.embd_normalize;
if (body.count("embd_normalize") != 0) { if (body.count("embd_normalize") != 0) {
embd_normalize = body.at("embd_normalize"); embd_normalize = body.at("embd_normalize");
if (meta->pooling_type == LLAMA_POOLING_TYPE_NONE) { if (meta->pooling_type == LLAMA_POOLING_TYPE_NONE) {