community[patch]: Added support for Ollama's num_predict option in ChatOllama (#16633)

Just a simple default addition to the options payload for a ollama
generate call to support a max_new_tokens parameter.

Should fix issue: https://github.com/langchain-ai/langchain/issues/14715

libs/community/langchain_community/llms/ollama.py

@@ -64,6 +64,10 @@ class _OllamaCommon(BaseLanguageModel):
     It is recommended to set this value to the number of physical
     CPU cores your system has (as opposed to the logical number of cores)."""
 
+    num_predict: Optional[int] = None
+    """Maximum number of tokens to predict when generating text. 
+    (Default: 128, -1 = infinite generation, -2 = fill context)"""
+
     repeat_last_n: Optional[int] = None
     """Sets how far back for the model to look back to prevent
     repetition. (Default: 64, 0 = disabled, -1 = num_ctx)"""
@@ -126,6 +130,7 @@ class _OllamaCommon(BaseLanguageModel):
                 "num_ctx": self.num_ctx,
                 "num_gpu": self.num_gpu,
                 "num_thread": self.num_thread,
+                "num_predict": self.num_predict,
                 "repeat_last_n": self.repeat_last_n,
                 "repeat_penalty": self.repeat_penalty,
                 "temperature": self.temperature,

libs/community/tests/unit_tests/llms/test_ollama.py

@@ -88,6 +88,7 @@ def test_handle_kwargs_top_level_parameters(monkeypatch: MonkeyPatch) -> None:
                 "num_ctx": None,
                 "num_gpu": None,
                 "num_thread": None,
+                "num_predict": None,
                 "repeat_last_n": None,
                 "repeat_penalty": None,
                 "stop": [],
@@ -133,6 +134,7 @@ def test_handle_kwargs_with_unknown_param(monkeypatch: MonkeyPatch) -> None:
                 "num_ctx": None,
                 "num_gpu": None,
                 "num_thread": None,
+                "num_predict": None,
                 "repeat_last_n": None,
                 "repeat_penalty": None,
                 "stop": [],