langchain-ollama (partners) / langchain-core: allow passing ChatMessages to Ollama (including arbitrary roles) (#30411)

Replacement for PR #30191 (@ccurme)

**Description**: currently, ChatOllama [will raise a value error if a
ChatMessage is passed to
it](https://github.com/langchain-ai/langchain/blob/master/libs/partners/ollama/langchain_ollama/chat_models.py#L514),
as described
https://github.com/langchain-ai/langchain/pull/30147#issuecomment-2708932481.

Furthermore, ollama-python is removing the limitations on valid roles
that can be passed through chat messages to a model in ollama -
https://github.com/ollama/ollama-python/pull/462#event-16917810634.

This PR removes the role limitations imposed by langchain and enables
passing langchain ChatMessages with arbitrary 'role' values through the
langchain ChatOllama class to the underlying ollama-python Client.

As this PR relies on [merged but unreleased functionality in
ollama-python](
https://github.com/ollama/ollama-python/pull/462#event-16917810634), I
have temporarily pointed the ollama package source to the main branch of
the ollama-python github repo.

Format, lint, and tests of new functionality passing. Need to resolve
issue with recently added ChatOllama tests. (Now resolved)

**Issue**: resolves #30122 (related to ollama issue
https://github.com/ollama/ollama/issues/8955)

**Dependencies**: no new dependencies

[x] PR title
[x] PR message
[x] Lint and test: format, lint, and test all running successfully and
passing

---------

Co-authored-by: Ryan Stewart <ryanstewart@Ryans-MacBook-Pro.local>
Co-authored-by: Chester Curme <chester.curme@gmail.com>

docs/docs/integrations/chat/ollama.ipynb

@@ -61,7 +61,9 @@
    "cell_type": "markdown",
    "id": "72ee0c4b-9764-423a-9dbf-95129e185210",
    "metadata": {},
-      "source": "To enable automated tracing of your model calls, set your [LangSmith](https://docs.smith.langchain.com/) API key:"
+   "source": [
+    "To enable automated tracing of your model calls, set your [LangSmith](https://docs.smith.langchain.com/) API key:"
+   ]
   },
   {
    "cell_type": "code",
@@ -98,7 +100,9 @@
    "cell_type": "markdown",
    "id": "b18bd692076f7cf7",
    "metadata": {},
-      "source": "Make sure you're using the latest Ollama version for structured outputs. Update by running:"
+   "source": [
+    "Make sure you're using the latest Ollama version for structured outputs. Update by running:"
+   ]
   },
   {
    "cell_type": "code",
@@ -106,7 +110,9 @@
    "id": "b7a05cba95644c2e",
    "metadata": {},
    "outputs": [],
-      "source": "%pip install -U ollama"
+   "source": [
+    "%pip install -U ollama"
+   ]
   },
   {
    "cell_type": "markdown",
@@ -438,6 +444,63 @@
     "print(query_chain)"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "id": "fb6a331f-1507-411f-89e5-c4d598154f3c",
+   "metadata": {},
+   "source": [
+    "## Reasoning models and custom message roles\n",
+    "\n",
+    "Some models, such as IBM's [Granite 3.2](https://ollama.com/library/granite3.2), support custom message roles to enable thinking processes.\n",
+    "\n",
+    "To access Granite 3.2's thinking features, pass a message with a `\"control\"` role with content set to `\"thinking\"`. Because `\"control\"` is a non-standard message role, we can use a [ChatMessage](https://python.langchain.com/api_reference/core/messages/langchain_core.messages.chat.ChatMessage.html) object to implement it:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "id": "d7309fa7-990e-4c20-b1f0-b155624ecf37",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Here is my thought process:\n",
+      "This question is asking for the result of 3 raised to the power of 3, which is a basic mathematical operation. \n",
+      "\n",
+      "Here is my response:\n",
+      "The expression 3^3 means 3 raised to the power of 3. To calculate this, you multiply the base number (3) by itself as many times as its exponent (3):\n",
+      "\n",
+      "3 * 3 * 3 = 27\n",
+      "\n",
+      "So, 3^3 equals 27.\n"
+     ]
+    }
+   ],
+   "source": [
+    "from langchain_core.messages import ChatMessage, HumanMessage\n",
+    "from langchain_ollama import ChatOllama\n",
+    "\n",
+    "llm = ChatOllama(model=\"granite3.2:8b\")\n",
+    "\n",
+    "messages = [\n",
+    "    ChatMessage(role=\"control\", content=\"thinking\"),\n",
+    "    HumanMessage(\"What is 3^3?\"),\n",
+    "]\n",
+    "\n",
+    "response = llm.invoke(messages)\n",
+    "print(response.content)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "6271d032-da40-44d4-9b52-58370e164be3",
+   "metadata": {},
+   "source": [
+    "Note that the model exposes its thought process in addition to its final response."
+   ]
+  },
   {
    "cell_type": "markdown",
    "id": "3a5bb5ca-c3ae-4a58-be67-2cd18574b9a3",
@@ -465,7 +528,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-      "version": "3.12.4"
+   "version": "3.10.4"
   }
  },
  "nbformat": 4,

libs/partners/ollama/langchain_ollama/chat_models.py

@@ -26,6 +26,7 @@ from langchain_core.messages import (
     AIMessageChunk,
     BaseMessage,
     BaseMessageChunk,
+    ChatMessage,
     HumanMessage,
     SystemMessage,
     ToolCall,
@@ -511,7 +512,7 @@ class ChatOllama(BaseChatModel):
     ) -> Sequence[Message]:
         ollama_messages: list = []
         for message in messages:
-            role: Literal["user", "assistant", "system", "tool"]
+            role: str
             tool_call_id: Optional[str] = None
             tool_calls: Optional[list[dict[str, Any]]] = None
             if isinstance(message, HumanMessage):
@@ -528,6 +529,8 @@ class ChatOllama(BaseChatModel):
                 )
             elif isinstance(message, SystemMessage):
                 role = "system"
+            elif isinstance(message, ChatMessage):
+                role = message.role
             elif isinstance(message, ToolMessage):
                 role = "tool"
                 tool_call_id = message.tool_call_id

libs/partners/ollama/pyproject.toml

@@ -6,7 +6,10 @@ build-backend = "pdm.backend"
 authors = []
 license = { text = "MIT" }
 requires-python = "<4.0,>=3.9"
-dependencies = ["ollama<1,>=0.4.4", "langchain-core<1.0.0,>=0.3.52"]
+dependencies = [
+    "ollama>=0.4.8,<1.0.0",
+    "langchain-core<1.0.0,>=0.3.52",
+]
 name = "langchain-ollama"
 version = "0.3.2"
 description = "An integration package connecting Ollama and LangChain"

libs/partners/ollama/tests/unit_tests/test_chat_models.py

@@ -1,7 +1,13 @@
 """Test chat model integration."""
 
 import json
+from collections.abc import Generator
+from contextlib import contextmanager
+from typing import Any
 
+import pytest
+from httpx import Client, Request, Response
+from langchain_core.messages import ChatMessage
 from langchain_tests.unit_tests import ChatModelUnitTests
 
 from langchain_ollama.chat_models import ChatOllama, _parse_arguments_from_tool_call
@@ -23,3 +29,38 @@ def test__parse_arguments_from_tool_call() -> None:
     response = _parse_arguments_from_tool_call(raw_tool_calls[0])
     assert response is not None
     assert isinstance(response["arg_1"], str)
+
+
+@contextmanager
+def _mock_httpx_client_stream(
+    *args: Any, **kwargs: Any
+) -> Generator[Response, Any, Any]:
+    yield Response(
+        status_code=200,
+        content='{"message": {"role": "assistant", "content": "The meaning ..."}}',
+        request=Request(method="POST", url="http://whocares:11434"),
+    )
+
+
+def test_arbitrary_roles_accepted_in_chatmessages(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(Client, "stream", _mock_httpx_client_stream)
+
+    llm = ChatOllama(
+        base_url="http://whocares:11434",
+        model="granite3.2",
+        verbose=True,
+        format=None,
+    )
+
+    messages = [
+        ChatMessage(
+            role="somerandomrole",
+            content="I'm ok with you adding any role message now!",
+        ),
+        ChatMessage(role="control", content="thinking"),
+        ChatMessage(role="user", content="What is the meaning of life?"),
+    ]
+
+    llm.invoke(messages)

libs/partners/ollama/uv.lock

@@ -288,7 +288,7 @@ wheels = [
 
 [[package]]
 name = "langchain-core"
-version = "0.3.52"
+version = "0.3.54"
 source = { editable = "../../core" }
 dependencies = [
     { name = "jsonpatch" },
@@ -381,7 +381,7 @@ typing = [
 [package.metadata]
 requires-dist = [
     { name = "langchain-core", editable = "../../core" },
-    { name = "ollama", specifier = ">=0.4.4,<1" },
+    { name = "ollama", specifier = ">=0.4.8,<1.0.0" },
 ]
 
 [package.metadata.requires-dev]
@@ -405,7 +405,7 @@ typing = [
 
 [[package]]
 name = "langchain-tests"
-version = "0.3.18"
+version = "0.3.19"
 source = { editable = "../../standard-tests" }
 dependencies = [
     { name = "httpx" },
@@ -625,15 +625,15 @@ wheels = [
 
 [[package]]
 name = "ollama"
-version = "0.4.7"
+version = "0.4.8"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "httpx" },
     { name = "pydantic" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/b0/6d/dc77539c735bbed5d0c873fb029fb86aa9f0163df169b34152914331c369/ollama-0.4.7.tar.gz", hash = "sha256:891dcbe54f55397d82d289c459de0ea897e103b86a3f1fad0fdb1895922a75ff", size = 12843 }
+sdist = { url = "https://files.pythonhosted.org/packages/e2/64/709dc99030f8f46ec552f0a7da73bbdcc2da58666abfec4742ccdb2e800e/ollama-0.4.8.tar.gz", hash = "sha256:1121439d49b96fa8339842965d0616eba5deb9f8c790786cdf4c0b3df4833802", size = 12972 }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/31/83/c3ffac86906c10184c88c2e916460806b072a2cfe34cdcaf3a0c0e836d39/ollama-0.4.7-py3-none-any.whl", hash = "sha256:85505663cca67a83707be5fb3aeff0ea72e67846cea5985529d8eca4366564a1", size = 13210 },
+    { url = "https://files.pythonhosted.org/packages/33/3f/164de150e983b3a16e8bf3d4355625e51a357e7b3b1deebe9cc1f7cb9af8/ollama-0.4.8-py3-none-any.whl", hash = "sha256:04312af2c5e72449aaebac4a2776f52ef010877c554103419d3f36066fe8af4c", size = 13325 },
 ]
 
 [[package]]