Merge b1a9dfff21 into 84ab88d843

integration-tests/models/test_grammar_response_format_llama.py

@@ -29,26 +29,55 @@ async def test_grammar_response_format_llama_json(llama_grammar, response_snapsh
         unit: str
         temperature: List[int]
 
+    json_payload={
+        "model": "tgi",
+        "messages": [
+            {
+                "role": "system",
+                "content": f"Respond to the users questions and answer them in the following format: {Weather.schema()}",
+            },
+            {
+                "role": "user",
+                "content": "What's the weather like the next 3 days in San Francisco, CA?",
+            },
+        ],
+        "seed": 42,
+        "max_tokens": 500,
+        "response_format": {"type": "json_object", "value": Weather.schema()},
+    }
     # send the request
     response = requests.post(
         f"{llama_grammar.base_url}/v1/chat/completions",
         headers=llama_grammar.headers,
-        json={
-            "model": "tgi",
-            "messages": [
-                {
-                    "role": "system",
-                    "content": f"Respond to the users questions and answer them in the following format: {Weather.schema()}",
-                },
-                {
-                    "role": "user",
-                    "content": "What's the weather like the next 3 days in San Francisco, CA?",
-                },
-            ],
-            "seed": 42,
-            "max_tokens": 500,
-            "response_format": {"type": "json_object", "value": Weather.schema()},
-        },
+        json=json_payload,
+    )
+
+    chat_completion = response.json()
+    called = chat_completion["choices"][0]["message"]["content"]
+
+    assert response.status_code == 200
+    assert called == '{ "unit": "fahrenheit", "temperature": [ 72, 79, 88 ] }'
+    assert chat_completion == response_snapshot
+
+    json_payload["response_format"]["type"] = "json"
+    response = requests.post(
+        f"{llama_grammar.base_url}/v1/chat/completions",
+        headers=llama_grammar.headers,
+        json=json_payload,
+    )
+
+    chat_completion = response.json()
+    called = chat_completion["choices"][0]["message"]["content"]
+
+    assert response.status_code == 200
+    assert called == '{ "unit": "fahrenheit", "temperature": [ 72, 79, 88 ] }'
+    assert chat_completion == response_snapshot
+
+    json_payload["response_format"]["type"] = "json_schema"
+    response = requests.post(
+        f"{llama_grammar.base_url}/v1/chat/completions",
+        headers=llama_grammar.headers,
+        json=json_payload,
     )
 
     chat_completion = response.json()

router/src/lib.rs

@@ -230,7 +230,8 @@ pub(crate) enum GrammarType {
     ///
     /// JSON Schema is a declarative language that allows to annotate JSON documents
     /// with types and descriptions.
-    #[serde(rename = "json")]
+    #[serde(rename = "json_schema")]
+    #[serde(alias = "json")]
     #[serde(alias = "json_object")]
     #[schema(example = json ! ({"properties": {"location":{"type": "string"}}}))]
     Json(serde_json::Value),