server : (embeddings) using same format for "input" and "content"

examples/server/server.cpp

@@ -3651,14 +3651,14 @@ int main(int argc, char ** argv) {
         const json body = json::parse(req.body);
         bool oaicompat = false;
 
-        // an input prompt can be a string or a list of tokens (integer)
+        // for the shape of input/content, see tokenize_input_prompts()
         json prompt;
-        if (body.count("input") != 0) {
+        if (body.contains("input")) {
             oaicompat = true;
             prompt = body.at("input");
-        } else if (body.count("content") != 0) {
-            // with "content", we only support single prompt
-            prompt = std::vector<std::string>{body.at("content")};
+        } else if (body.contains("content")) {
+            oaicompat = false;
+            prompt = body.at("content");
         } else {
             res_error(res, format_error_response("\"input\" or \"content\" must be provided", ERROR_TYPE_INVALID_REQUEST));
             return;

examples/server/tests/unit/test_embedding.py

@@ -45,6 +45,35 @@ def test_embedding_multiple():
         assert len(d['embedding']) > 1
 
 
+@pytest.mark.parametrize(
+    "content,is_multi_prompt",
+    [
+        # single prompt
+        ("string", False),
+        ([12, 34, 56], False),
+        ([12, 34, "string", 56, 78], False),
+        # multiple prompts
+        (["string1", "string2"], True),
+        (["string1", [12, 34, 56]], True),
+        ([[12, 34, 56], [12, 34, 56]], True),
+        ([[12, 34, 56], [12, "string", 34, 56]], True),
+    ]
+)
+def test_embedding_mixed_input(content, is_multi_prompt: bool):
+    global server
+    server.start()
+    res = server.make_request("POST", "/embeddings", data={"content": content})
+    assert res.status_code == 200
+    if is_multi_prompt:
+        assert len(res.body) == len(content)
+        for d in res.body:
+            assert 'embedding' in d
+            assert len(d['embedding']) > 1
+    else:
+        assert 'embedding' in res.body
+        assert len(res.body['embedding']) > 1
+
+
 def test_embedding_openai_library_single():
     global server
     server.start()

examples/server/utils.hpp

@@ -138,6 +138,7 @@ static llama_tokens tokenize_mixed(const llama_context * ctx, const json & json_
  * and multiple prompts (multi-tasks):
  * - "prompt": ["string1", "string2"]
  * - "prompt": ["string1", [12, 34, 56]]
+ * - "prompt": [[12, 34, 56], [78, 90, 12]]
  * - "prompt": [[12, 34, "string", 56, 78], [12, 34, 56]]
  */
 static std::vector<llama_tokens> tokenize_input_prompts(llama_context * ctx, const json & json_prompt, bool add_special, bool parse_special) {