add support for gbnf grammar

README.md

@@ -92,6 +92,21 @@ curl http://localhost:3928/v1/chat/completions \
   }'
 ```
 
+***OPTIONAL***: You can constrain the sampling using GBNF grammars by providing path to a grammar file
+```bash title="Nitro Inference With Grammar"
+curl http://localhost:3928/v1/chat/completions \
+  -H "Content-Type: application/json" \
+  -d '{
+    "messages": [
+      {
+        "role": "user",
+        "content": "Who won the world series in 2020?"
+      },
+    ],
+    "grammar_file": "/path/to/grammarfile"
+  }'
+```
+
 Table of parameters
 
 | Parameter        | Type    | Description                                                  |

controllers/llamaCPP.cc

@@ -194,7 +194,15 @@ void llamaCPP::chatCompletion(
         (*jsonBody).get("frequency_penalty", 0).asFloat();
     data["presence_penalty"] = (*jsonBody).get("presence_penalty", 0).asFloat();
     const Json::Value &messages = (*jsonBody)["messages"];
-
+    std::string grammar_file = (*jsonBody).get("grammar_file", "").asString();
+    std::ifstream file(grammar_file);
+    if (!file) {
+        LOG_ERROR << "Grammar file not found";
+    } else {
+      std::stringstream grammarBuf;
+      grammarBuf << file.rdbuf();
+      data["grammar"] = grammarBuf.str();
+    }
     if (!llama.multimodal) {
 
       for (const auto &message : messages) {

examples/grammars/json.gbnf

@@ -0,0 +1,25 @@
+root   ::= object
+value  ::= object | array | string | number | ("true" | "false" | "null") ws
+
+object ::=
+  "{" ws (
+            string ":" ws value
+    ("," ws string ":" ws value)*
+  )? "}" ws
+
+array  ::=
+  "[" ws (
+            value
+    ("," ws value)*
+  )? "]" ws
+
+string ::=
+  "\"" (
+    [^"\\] |
+    "\\" (["\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F]) # escapes
+  )* "\"" ws
+
+number ::= ("-"? ([0-9] | [1-9] [0-9]*)) ("." [0-9]+)? ([eE] [-+]? [0-9]+)? ws
+
+# Optional space: by convention, applied in this grammar after literal chars when allowed
+ws ::= ([ \t\n] ws)?