[BUG]: The mistral pipeline command is missing

[clafollett]

Bug description

I attempted to run the advertised demo on the Modular site for the Mistral pipeline but receive the following output stating the mistral command does not exist:

23:39:53.958 INFO: MainThread: root: Logging initialized: Console: 20, File: None, Telemetry: None
Usage: pipelines.py [OPTIONS] COMMAND [ARGS]...
Try 'pipelines.py --help' for help.

Error: Command not supported: mistral
Supported commands: generate, list, llama3, replit, serve

I have a potential fix in a fork but I do not believe I have it configured correctly. I get a bf16 datatype error on my Macbook Pro M3 Max. I'm fairly new in the AI world and assume it's some missing transform config?

error: The bf16 data type is not supported on device 'cpu:0'.

@main.command(name="mistral")
@pipeline_config_options
@common_server_options
@click.option(
    "--prompt",
    type=str,
    default="Why is the sky blue?",
    help="The text prompt to use for further generation.",
)
@click.option(
    "--num-warmups",
    type=int,
    default=0,
    show_default=True,
    help="# of warmup iterations to run before the final timed run.",
)
@click.option(
    "--serve",
    type=bool,
    default=False,
    is_flag=True,
    show_default=True,
    help="Whether to serve an OpenAI HTTP endpoint on port 8000.",
)
def run_mistral(
    prompt,
    num_warmups,
    serve,
    profile_serve,
    performance_fake,
    batch_timeout,
    model_name,
    **config_kwargs,
):
    """Runs the Mistral pipeline."""

    # Update basic parameters.
    if config_kwargs["architecture"] is None:
        config_kwargs["architecture"] = "MistralForCausalLM"

    if config_kwargs["architecture"] != "MistralForCausalLM":
        msg = (
            f"provided architecture '{config_kwargs['architecture']}' not"
            " compatible with Mistral."
        )
        raise ValueError(msg)

    config_kwargs["trust_remote_code"] = True

    config = PipelineConfig(**config_kwargs)

    # if config.quantization_encoding not in [
    #     SupportedEncoding.bfloat16
    # ]:
    #     config.cache_strategy = KVCacheStrategy.NAIVE

    if serve:
        serve_pipeline(
            pipeline_config=config,
            profile=profile_serve,
            performance_fake=performance_fake,
            batch_timeout=batch_timeout,
            model_name=model_name,
        )
    else:
        generate_text_for_pipeline(
            pipeline_config=config, prompt=prompt, num_warmups=num_warmups
        )

Steps to reproduce

• Include relevant code snippet or link to code that did not work as expected.
• If applicable, add screenshots to help explain the problem.
• Include anything else that might help us debug the issue.

Simply follow the steps outlined here:
Mistral NeMo

Review the code file for the pipeline definitions and see the command definition is not there:
https://github.com/modular/max/blob/main/pipelines/python/pipelines.py#L141-L266

System information

- What OS did you do install MAX on ?
macOS Sequoia 15.2

- Provide version information for MAX by pasting the output of max -v`
- Provide version information for Mojo by pasting the output of mojo -v`
- Provide Magic CLI version by pasting the output of `magic -v`
magic 0.6.2 - (based on pixi 0.40.0)