Fixed function calling server api bugs; Updated SERVER.md

SERVER.md

@@ -299,7 +299,7 @@ Modifies existing images based on a single prompt.
 }
 ```
 
-### 6. Text-to-Image: <code>/v1/txt2speech</code>
+### 6. Text-to-Speech: <code>/v1/txt2speech</code>
 
 Generates wav format voices based on a single prompt.
 
@@ -330,7 +330,57 @@ The `language` argument applies only to OuteTTS models.
 }
 ```
 
-### 7. Audio Transcriptions: <code>/v1/audio/transcriptions</code>
+### 7. Function Calling API: <code>/v1/function_calling</code>
+
+Generate function calling command based on provided messages and available tools.
+
+#### Request body:
+
+```json
+{
+  "tools": [
+    {
+      "type": "function",
+      "function": {
+        "name": "add_integer",
+        "description": "Returns the addition of input integers.",
+        "parameters": {
+          "type": "object",
+          "properties": {
+            "num1": {"type": "integer", "description": "An integer to add."},
+            "num2": {"type": "integer", "description": "An integer to add."}
+          },
+          "required": ["num1", "num2"],
+          "additionalProperties": false
+        },
+        "strict": true
+      }
+    }
+  ],
+  "messages": [
+    {"role": "system", "content": "You are a helpful assistant."},
+    {"role": "user", "content": "Please calculate the sum of 42 and 100."}
+  ]
+}
+
+#### Example Response:
+
+```json
+{
+  "created": 1738610394.597405,
+  "response": [
+    {
+      "type": "function",
+      "function": {
+        "name": "add_integer",
+        "arguments": "{\"num1\": 42, \"num2\": 100}"
+      }
+    }
+  ]
+}
+```
+
+### 8. Audio Transcriptions: <code>/v1/audio/transcriptions</code>
 
 Transcribes audio files to text.
 
@@ -356,7 +406,7 @@ Transcribes audio files to text.
 }
 ```
 
-### 8. Audio Translations: <code>/v1/audio/translations</code>
+### 9. Audio Translations: <code>/v1/audio/translations</code>
 
 Translates audio files to text in English.
 
@@ -381,7 +431,7 @@ Translates audio files to text in English.
 }
 ```
 
-### 9. Generate Embeddings: <code>/v1/embeddings</code>
+### 10. Generate Embeddings: <code>/v1/embeddings</code>
 
 Generate embeddings for a given text.
 

nexa/cli/entry.py

@@ -160,7 +160,7 @@ def run_ggml_server(args):
     hf = kwargs.pop('huggingface', False)
     ms = kwargs.pop('modelscope', False)
     use_function_calling = kwargs.pop('function_calling', False)
-    
+
     run_type = None
     if model_type:
         run_type = ModelType[model_type].value
@@ -194,8 +194,8 @@ def run_ggml_server(args):
         model_type_arg=run_type,
         huggingface=hf,
         modelscope=ms,
+        function_calling=use_function_calling,
         projector_local_path_arg=projector_local_path,
-        use_function_calling=use_function_calling
         **kwargs
     )
 

nexa/gguf/server/nexa_service.py

@@ -357,12 +357,6 @@ async def load_model():
     global model, chat_format, completion_template, model_path, n_ctx, is_local_path, model_type, is_huggingface, is_modelscope, projector_path
     global use_function_calling
 
-    if use_function_calling and model_type != "NLP":
-        raise ValueError(
-            "Function calling is only supported for NLP models. "
-            "Please ensure that you are using a compatible NLP model before enabling this feature."
-    )
-
     if is_local_path:
         if model_type == "Multimodal":
             if not projector_path:
@@ -394,6 +388,12 @@ async def load_model():
     if use_function_calling:
         print('Function calling option is enabled')
 
+    if use_function_calling and model_type != "NLP":
+        raise ValueError(
+            "Function calling is only supported for NLP models. "
+            "Please ensure that you are using a compatible NLP model before enabling this feature."
+    )
+
     if model_type == "NLP" or model_type == "Text Embedding":
         if model_type == "NLP" and use_function_calling:
             from nexa.gguf.nexa_inference_text import NexaTextInference
@@ -768,7 +768,7 @@ def load_audio_from_bytes(audio_bytes: bytes):
         a = librosa.resample(a, orig_sr=sr, target_sr=SAMPLING_RATE)
     return a
 
-def run_nexa_ai_service(model_path_arg=None, is_local_path_arg=False, model_type_arg=None, huggingface=False, modelscope=False, projector_local_path_arg=None, function_calling=False, **kwargs):
+def run_nexa_ai_service(model_path_arg=None, is_local_path_arg=False, model_type_arg=None, huggingface=False, modelscope=False, function_calling=False, projector_local_path_arg=None, **kwargs):
     global model_path, n_ctx, is_local_path, model_type, is_huggingface, is_modelscope, projector_path, use_function_calling
     is_local_path = is_local_path_arg
     is_huggingface = huggingface