CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-03 09:22:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

embedding: add raw option for --embd-output-format (#16541)

Browse Source 
* Add --embd-output-format raw for plain numeric embedding output

This new option outputs embeddings as raw space-separated floats, without JSON or 'embedding N:' prefixes. Useful for downstream vector pipelines and scripting.

* Move raw output handling into format handling section

* Move raw output handling into else-if block with other format handlers

* Use LOG instead of printf for raw embedding output

* docs: document 'raw' embedding output format in arg.cpp and README
This commit is contained in:
Sam Malayek
2025-10-28 03:51:41 -07:00
committed by GitHub
parent 7a0e900e36
commit 1c1409e131
3 changed files with 27 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
common/arg.cpp
View File

@@ -3248,7 +3248,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
).set_examples({LLAMA_EXAMPLE_EMBEDDING}));
add_opt(common_arg(
{"--embd-output-format"}, "FORMAT",
"empty = default, \"array\" = [[],[]...], \"json\" = openai style, \"json+\" = same \"json\" + cosine similarity matrix",
"empty = default, \"array\" = [[],[]...], \"json\" = openai style, \"json+\" = same \"json\" + cosine similarity matrix, \"raw\" = plain whitespace-delimited output (one embedding per line)",
[](common_params & params, const std::string & value) {
params.embd_out = value;
}

1
examples/embedding/README.md
View File

@@ -38,6 +38,7 @@ The above command will output space-separated float values.
| | multiple embeddings | $[[x_1,...,x_n],[x_1,...,x_n],...,[x_1,...,x_n]]$
| 'json' | openai style |
| 'json+' | add cosine similarity matrix |
| 'raw' | plain text output |
### --embd-separator $"string"$
| $"string"$ | |

25
examples/embedding/embedding.cpp
View File

@@ -70,6 +70,29 @@ static void batch_decode(llama_context * ctx, llama_batch & batch, float * outpu
}
}
// plain, pipe-friendly output: one embedding per line
static void print_raw_embeddings(const float * emb,
int n_embd_count,
int n_embd,
const llama_model * model,
enum llama_pooling_type pooling_type,
int embd_normalize) {
const uint32_t n_cls_out = llama_model_n_cls_out(model);
const bool is_rank = (pooling_type == LLAMA_POOLING_TYPE_RANK);
const int cols = is_rank ? std::min<int>(n_embd, (int) n_cls_out) : n_embd;
for (int j = 0; j < n_embd_count; ++j) {
for (int i = 0; i < cols; ++i) {
if (embd_normalize == 0) {
LOG("%1.0f%s", emb[j * n_embd + i], (i + 1 < cols ? " " : ""));
} else {
LOG("%1.7f%s", emb[j * n_embd + i], (i + 1 < cols ? " " : ""));
}
}
LOG("\n");
}
}
int main(int argc, char ** argv) {
common_params params;
@@ -372,6 +395,8 @@ int main(int argc, char ** argv) {
}
if (notArray) LOG("\n}\n");
} else if (params.embd_out == "raw") {
print_raw_embeddings(emb, n_embd_count, n_embd, model, pooling_type, params.embd_normalize);
}
LOG("\n");