Mistype issue using MLX Chat Model via MLXPipeline

[kozachynskyi]

Checked other resources

• I added a very descriptive title to this issue.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.
• I am sure that this is a bug in LangChain rather than my code.
• The bug is not resolved by updating to the latest stable version of LangChain (or the specific integration package).

Example Code

from langchain_community.llms import MLXPipeline
from langchain_community.chat_models.mlx import ChatMLX
from langchain.agents import AgentExecutor, load_tools

from langchain_core.prompts.chat import (
    ChatPromptTemplate,
    HumanMessagePromptTemplate,
    SystemMessagePromptTemplate,
)
from langchain.tools.render import render_text_description
from langchain.agents.format_scratchpad import format_log_to_str
from langchain.agents.output_parsers import (
    ReActJsonSingleInputOutputParser,
)

system = '''a'''
human = '''{input}
{agent_scratchpad}
'''
def get_custom_prompt():
    messages = [
            SystemMessagePromptTemplate.from_template(system),
            HumanMessagePromptTemplate.from_template(human),
    ]
    input_variables = ["agent_scratchpad", "input", "tool_names", "tools"]
    return ChatPromptTemplate(input_variables=input_variables, messages=messages)

llm = MLXPipeline.from_model_id(
    model_id="mlx-community/Meta-Llama-3-8B-Instruct-4bit",
)
chat_model = ChatMLX(llm=llm)

prompt = get_custom_prompt()
prompt = prompt.partial(
    tools=render_text_description([]),
    tool_names=", ".join([t.name for t in []]),
)

chat_model_with_stop = chat_model.bind(stop=["\nObservation"])
agent = (
    {
        "input": lambda x: x["input"],
        "agent_scratchpad": lambda x: format_log_to_str(x["intermediate_steps"]),
    }
    | prompt
    | chat_model_with_stop
    | ReActJsonSingleInputOutputParser()
)

# instantiate AgentExecutor
agent_executor = AgentExecutor(agent=agent, tools=[], verbose=True)

agent_executor.invoke(
    {
        "input": "What is your name?"
    }
)

Error Message and Stack Trace (if applicable)

File "/Users/==/.pyenv/versions/hack/lib/python3.11/site-packages/langchain_community/chat_models/mlx.py", line 184, in _stream
text = self.tokenizer.decode(token.item())
^^^^^^^^^^
AttributeError: 'int' object has no attribute 'item'
Uncaught exception. Entering post mortem debugging

Description

Hi there,

I assume this bug is similar to this issue #20561. Why is that? Because if you locally apply changes from this patch ad48f77 starting from line 174 to langchain_community/chat_models/mlx.py, the bug disappears.

Best wishes

System Info

langchain==0.2.12
langchain-community==0.2.11
langchain-core==0.2.28
langchain-experimental==0.0.64
langchain-huggingface==0.0.3
langchain-text-splitters==0.2.2

mac

3.11.9