together 2.0.0a14__py3-none-any.whl → 2.0.0a16__py3-none-any.whl

together/lib/cli/api/endpoints.py

@@ -1,467 +0,0 @@
-from __future__ import annotations
-
-import sys
-import json
-from typing import Any, Dict, Literal, TypeVar, Callable, cast
-from functools import wraps
-
-import click
-
-from together import Together, omit
-from together.types import DedicatedEndpoint
-from together._exceptions import APIError
-from together.lib.utils.serializer import datetime_serializer
-from together.types.endpoint_list_response import Data as DedicatedEndpointListItem
-
-
-def print_endpoint(endpoint: DedicatedEndpoint | DedicatedEndpointListItem) -> None:
-    """Print endpoint details in a Docker-like format or JSON."""
-
-    # Print header info
-    click.echo(f"ID:\t\t{endpoint.id}")
-    click.echo(f"Name:\t\t{endpoint.name}")
-
-    if isinstance(endpoint, DedicatedEndpoint):
-        click.echo(f"Display Name:\t{endpoint.display_name}")
-        click.echo(f"Hardware:\t{endpoint.hardware}")
-        click.echo(f"Autoscaling:\tMin={endpoint.autoscaling.min_replicas}, Max={endpoint.autoscaling.max_replicas}")
-
-    click.echo(f"Model:\t\t{endpoint.model}")
-    click.echo(f"Type:\t\t{endpoint.type}")
-    click.echo(f"Owner:\t\t{endpoint.owner}")
-    click.echo(f"State:\t\t{endpoint.state}")
-    click.echo(f"Created:\t{endpoint.created_at}")
-
-
-F = TypeVar("F", bound=Callable[..., Any])
-
-
-def print_api_error(e: APIError) -> None:
-    error_details = cast(Dict[str, Any], e.body)["error"]["message"]
-
-    if error_details and ("credentials" in error_details.lower() or "authentication" in error_details.lower()):
-        click.echo("Error: Invalid API key or authentication failed", err=True)
-    else:
-        click.echo(f"Error: {error_details}", err=True)
-
-
-def handle_api_errors(f: F) -> F:
-    """Decorator to handle common API errors in CLI commands."""
-
-    @wraps(f)
-    def wrapper(*args: Any, **kwargs: Any) -> Any:
-        try:
-            return f(*args, **kwargs)
-        except APIError as e:
-            print_api_error(e)
-            sys.exit(1)
-        except Exception as e:
-            click.echo(f"Error: An unexpected error occurred - {str(e)}", err=True)
-            sys.exit(1)
-
-    return wrapper  # type: ignore
-
-
-@click.group()
-@click.pass_context
-def endpoints(ctx: click.Context) -> None:
-    """Endpoints API commands"""
-    pass
-
-
-@endpoints.command()
-@click.option(
-    "--model",
-    required=True,
-    help="The model to deploy (e.g. meta-llama/Llama-4-Scout-17B-16E-Instruct)",
-)
-@click.option(
-    "--min-replicas",
-    type=int,
-    default=1,
-    help="Minimum number of replicas to deploy",
-)
-@click.option(
-    "--max-replicas",
-    type=int,
-    default=1,
-    help="Maximum number of replicas to deploy",
-)
-@click.option(
-    "--gpu",
-    type=click.Choice(["h100", "a100", "l40", "l40s", "rtx-6000"]),
-    required=True,
-    help="GPU type to use for inference",
-)
-@click.option(
-    "--gpu-count",
-    type=int,
-    default=1,
-    help="Number of GPUs to use per replica",
-)
-@click.option(
-    "--display-name",
-    help="A human-readable name for the endpoint",
-)
-@click.option(
-    "--no-prompt-cache",
-    is_flag=True,
-    help="Disable the prompt cache for this endpoint",
-)
-@click.option(
-    "--no-speculative-decoding",
-    is_flag=True,
-    help="Disable speculative decoding for this endpoint",
-)
-@click.option(
-    "--no-auto-start",
-    is_flag=True,
-    help="Create the endpoint in STOPPED state instead of auto-starting it",
-)
-@click.option(
-    "--inactive-timeout",
-    type=int,
-    help="Number of minutes of inactivity after which the endpoint will be automatically stopped. Set to 0 to disable.",
-)
-@click.option(
-    "--availability-zone",
-    help="Start endpoint in specified availability zone (e.g., us-central-4b)",
-)
-@click.option(
-    "--wait/--no-wait",
-    default=True,
-    help="Wait for the endpoint to be ready after creation",
-)
-@click.pass_obj
-@handle_api_errors
-def create(
-    client: Together,
-    model: str,
-    min_replicas: int,
-    max_replicas: int,
-    gpu: str,
-    gpu_count: int,
-    display_name: str | None,
-    no_prompt_cache: bool | None,
-    no_speculative_decoding: bool | None,
-    no_auto_start: bool,
-    inactive_timeout: int | None,
-    availability_zone: str | None,
-    wait: bool,
-) -> None:
-    """Create a new dedicated inference endpoint."""
-    # Map GPU types to their full hardware ID names
-    gpu_map = {
-        "h100": "nvidia_h100_80gb_sxm",
-        "a100": "nvidia_a100_80gb_pcie" if gpu_count == 1 else "nvidia_a100_80gb_sxm",
-        "l40": "nvidia_l40",
-        "l40s": "nvidia_l40s",
-        "rtx-6000": "nvidia_rtx_6000_ada",
-    }
-
-    hardware_id = f"{gpu_count}x_{gpu_map[gpu]}"
-
-    try:
-        response = client.endpoints.create(
-            model=model,
-            hardware=hardware_id,
-            autoscaling={
-                "min_replicas": min_replicas,
-                "max_replicas": max_replicas,
-            },
-            display_name=display_name or omit,
-            disable_prompt_cache=no_prompt_cache or omit,
-            disable_speculative_decoding=no_speculative_decoding or omit,
-            state="STOPPED" if no_auto_start else "STARTED",
-            inactive_timeout=inactive_timeout,
-            extra_query={"availability_zone": availability_zone or omit},
-        )
-    except APIError as e:
-        print_api_error(e)
-        if "check the hardware api" in str(e.args[0]).lower() or "invalid hardware provided" in str(e.args[0]).lower():
-            fetch_and_print_hardware_options(client=client, model=model, print_json=False, available=True)
-
-        sys.exit(1)
-
-    # Print detailed information to stderr
-    click.echo("Created dedicated endpoint with:", err=True)
-    click.echo(f"  Model: {model}", err=True)
-    click.echo(f"  Min replicas: {min_replicas}", err=True)
-    click.echo(f"  Max replicas: {max_replicas}", err=True)
-    click.echo(f"  Hardware: {hardware_id}", err=True)
-    if display_name:
-        click.echo(f"  Display name: {display_name}", err=True)
-    if no_prompt_cache:
-        click.echo("  Prompt cache: disabled", err=True)
-    if no_speculative_decoding:
-        click.echo("  Speculative decoding: disabled", err=True)
-    if no_auto_start:
-        click.echo("  Auto-start: disabled", err=True)
-    if inactive_timeout is not None:
-        click.echo(f"  Inactive timeout: {inactive_timeout} minutes", err=True)
-    if availability_zone:
-        click.echo(f"  Availability zone: {availability_zone}", err=True)
-
-    click.echo(f"Endpoint created successfully, id: {response.id}", err=True)
-
-    if wait:
-        import time
-
-        click.echo("Waiting for endpoint to be ready...", err=True)
-        while client.endpoints.retrieve(response.id).state != "STARTED":
-            time.sleep(1)
-        click.echo("Endpoint ready", err=True)
-
-    # Print only the endpoint ID to stdout
-    click.echo(response.id)
-
-
-@endpoints.command()
-@click.argument("endpoint-id", required=True)
-@click.option("--json", is_flag=True, help="Print output in JSON format")
-@click.pass_obj
-@handle_api_errors
-def get(client: Together, endpoint_id: str, json: bool) -> None:
-    """Get a dedicated inference endpoint."""
-    endpoint = client.endpoints.retrieve(endpoint_id)
-    if json:
-        import json as json_lib
-
-        click.echo(json_lib.dumps(endpoint.model_dump(), indent=2, default=datetime_serializer))
-    else:
-        print_endpoint(endpoint)
-
-
-@endpoints.command()
-@click.option("--model", help="Filter hardware options by model")
-@click.option("--json", is_flag=True, help="Print output in JSON format")
-@click.option(
-    "--available",
-    is_flag=True,
-    help="Print only available hardware options (can only be used if model is passed in)",
-)
-@click.pass_obj
-@handle_api_errors
-def hardware(client: Together, model: str | None, json: bool, available: bool) -> None:
-    """List all available hardware options, optionally filtered by model."""
-    fetch_and_print_hardware_options(client, model, json, available)
-
-
-def fetch_and_print_hardware_options(client: Together, model: str | None, print_json: bool, available: bool) -> None:
-    """Print hardware options for a model."""
-
-    message = "Available hardware options:" if available else "All hardware options:"
-    click.echo(message, err=True)
-    hardware_options = client.hardware.list(model=model or omit)
-    # hardware_options = client.endpoints.list_hardware(model)
-    if available:
-        hardware_options.data = [
-            hardware
-            for hardware in hardware_options.data
-            if hardware.availability is not None and hardware.availability.status == "available"
-        ]
-
-    if print_json:
-        json_output = [hardware.model_dump() for hardware in hardware_options.data]
-        click.echo(json.dumps(json_output, default=datetime_serializer, indent=2))
-    else:
-        for hardware in hardware_options.data:
-            click.echo(f"  {hardware.id}", err=True)
-
-
-@endpoints.command()
-@click.argument("endpoint-id", required=True)
-@click.option("--wait/--no-wait", default=True, help="Wait for the endpoint to stop")
-@click.pass_obj
-@handle_api_errors
-def stop(client: Together, endpoint_id: str, wait: bool) -> None:
-    """Stop a dedicated inference endpoint."""
-    client.endpoints.update(endpoint_id, state="STOPPED")
-    click.echo("Successfully marked endpoint as stopping", err=True)
-
-    if wait:
-        import time
-
-        click.echo("Waiting for endpoint to stop...", err=True)
-        while client.endpoints.retrieve(endpoint_id).state != "STOPPED":
-            time.sleep(1)
-        click.echo("Endpoint stopped", err=True)
-
-    click.echo(endpoint_id)
-
-
-@endpoints.command()
-@click.argument("endpoint-id", required=True)
-@click.option("--wait/--no-wait", default=True, help="Wait for the endpoint to start")
-@click.pass_obj
-@handle_api_errors
-def start(client: Together, endpoint_id: str, wait: bool) -> None:
-    """Start a dedicated inference endpoint."""
-    client.endpoints.update(endpoint_id, state="STARTED")
-    click.echo("Successfully marked endpoint as starting", err=True)
-
-    if wait:
-        import time
-
-        click.echo("Waiting for endpoint to start...", err=True)
-        while client.endpoints.retrieve(endpoint_id).state != "STARTED":
-            time.sleep(1)
-        click.echo("Endpoint started", err=True)
-
-    click.echo(endpoint_id)
-
-
-@endpoints.command()
-@click.argument("endpoint-id", required=True)
-@click.pass_obj
-@handle_api_errors
-def delete(client: Together, endpoint_id: str) -> None:
-    """Delete a dedicated inference endpoint."""
-    client.endpoints.delete(endpoint_id)
-    click.echo("Successfully deleted endpoint", err=True)
-    click.echo(endpoint_id)
-
-
-@endpoints.command()
-@click.option("--json", is_flag=True, help="Print output in JSON format")
-@click.option(
-    "--type",
-    type=click.Choice(["dedicated", "serverless"]),
-    help="Filter by endpoint type",
-)
-@click.option(
-    "--mine",
-    type=click.BOOL,
-    default=None,
-    help="true (only mine), default=all",
-)
-@click.option(
-    "--usage-type",
-    type=click.Choice(["on-demand", "reserved"]),
-    help="Filter by endpoint usage type",
-)
-@click.pass_obj
-@handle_api_errors
-def list(
-    client: Together,
-    json: bool,
-    type: Literal["dedicated", "serverless"] | None,
-    usage_type: Literal["on-demand", "reserved"] | None,
-    mine: bool | None,
-) -> None:
-    """List all inference endpoints (includes both dedicated and serverless endpoints)."""
-    endpoints = client.endpoints.list(
-        type=type or omit,
-        usage_type=usage_type or omit,
-        mine=mine if mine is not None else omit,
-    )
-
-    if not endpoints:
-        click.echo("No dedicated endpoints found", err=True)
-        return
-
-    click.echo("Endpoints:", err=True)
-    if json:
-        import json as json_lib
-
-        click.echo(
-            json_lib.dumps(
-                [endpoint.model_dump() for endpoint in endpoints.data], default=datetime_serializer, indent=2
-            )
-        )
-    else:
-        for endpoint in endpoints.data:
-            print_endpoint(
-                endpoint,
-            )
-            click.echo()
-
-
-@endpoints.command()
-@click.argument("endpoint-id", required=True)
-@click.option(
-    "--display-name",
-    help="A new human-readable name for the endpoint",
-)
-@click.option(
-    "--min-replicas",
-    type=int,
-    help="New minimum number of replicas to maintain",
-)
-@click.option(
-    "--max-replicas",
-    type=int,
-    help="New maximum number of replicas to scale up to",
-)
-@click.option(
-    "--inactive-timeout",
-    type=int,
-    help="Number of minutes of inactivity after which the endpoint will be automatically stopped. Set to 0 to disable.",
-)
-@click.pass_obj
-@handle_api_errors
-def update(
-    client: Together,
-    endpoint_id: str,
-    display_name: str | None,
-    min_replicas: int | None,
-    max_replicas: int | None,
-    inactive_timeout: int | None,
-) -> None:
-    """Update a dedicated inference endpoint's configuration."""
-    if not any([display_name, min_replicas, max_replicas, inactive_timeout]):
-        click.echo("Error: At least one update option must be specified", err=True)
-        sys.exit(1)
-
-    # Build kwargs for the update
-    kwargs: Dict[str, Any] = {}
-    if display_name is not None:
-        kwargs["display_name"] = display_name
-
-    if min_replicas is not None or max_replicas is not None:
-        kwargs["autoscaling"] = {}
-        if min_replicas is not None:
-            kwargs["autoscaling"]["min_replicas"] = min_replicas
-        if max_replicas is not None:
-            kwargs["autoscaling"]["max_replicas"] = max_replicas
-
-    if inactive_timeout is not None:
-        kwargs["inactive_timeout"] = inactive_timeout
-
-    client.endpoints.update(endpoint_id, **kwargs)
-
-    # Print what was updated
-    click.echo("Updated endpoint configuration:", err=True)
-    if display_name:
-        click.echo(f"  Display name: {display_name}", err=True)
-    if min_replicas:
-        click.echo(f"  Min replicas: {min_replicas}", err=True)
-    if max_replicas:
-        click.echo(f"  Max replicas: {max_replicas}", err=True)
-    if inactive_timeout is not None:
-        click.echo(f"  Inactive timeout: {inactive_timeout} minutes", err=True)
-
-    click.echo("Successfully updated endpoint", err=True)
-    click.echo(endpoint_id)
-
-
-@endpoints.command()
-@click.option("--json", is_flag=True, help="Print output in JSON format")
-@click.pass_obj
-@handle_api_errors
-def availability_zones(client: Together, json: bool) -> None:
-    """List all availability zones."""
-    avzones = client.endpoints.list_avzones()
-
-    if not avzones:
-        click.echo("No availability zones found", err=True)
-        return
-
-    if json:
-        import json as json_lib
-
-        click.echo(json_lib.dumps(avzones.model_dump(), indent=2))
-    else:
-        click.echo("Available zones:", err=True)
-        for availability_zone in sorted(avzones.avzones):
-            click.echo(f"  {availability_zone}")

together/lib/cli/api/files.py

@@ -1,133 +0,0 @@
-import json
-import pathlib
-from typing import Any, Dict, List, get_args
-from textwrap import wrap
-
-import click
-from tabulate import tabulate
-
-from together import Together
-from together.types import FilePurpose
-
-# from together.utils import check_file, convert_bytes, convert_unix_timestamp
-from ...utils import check_file, convert_bytes, convert_unix_timestamp
-
-
-@click.group()
-@click.pass_context
-def files(ctx: click.Context) -> None:
-    """File API commands"""
-    pass
-
-
-@files.command()
-@click.pass_context
-@click.argument(
-    "file",
-    type=click.Path(exists=True, file_okay=True, resolve_path=True, readable=True, dir_okay=False),
-    required=True,
-)
-@click.option(
-    "--purpose",
-    type=click.Choice(get_args(FilePurpose)),
-    default="fine-tune",
-    help="Purpose of file upload. Acceptable values in enum `together.types.FilePurpose`. Defaults to `fine-tunes`.",
-)
-@click.option(
-    "--check/--no-check",
-    default=True,
-    help="Whether to check the file before uploading.",
-)
-def upload(ctx: click.Context, file: pathlib.Path, purpose: FilePurpose, check: bool) -> None:
-    """Upload file"""
-
-    client: Together = ctx.obj
-
-    response = client.files.upload(file=file, purpose=purpose, check=check)
-
-    click.echo(json.dumps(response.model_dump(exclude_none=True), indent=4))
-
-
-@files.command()
-@click.pass_context
-def list(ctx: click.Context) -> None:
-    """List files"""
-    client: Together = ctx.obj
-
-    response = client.files.list()
-
-    display_list: List[Dict[str, Any]] = []
-    for i in response.data or []:
-        display_list.append(
-            {
-                "File name": "\n".join(wrap(i.filename or "", width=30)),
-                "File ID": i.id,
-                "Size": convert_bytes(float(str(i.bytes))),  # convert to string for mypy typing
-                "Created At": convert_unix_timestamp(i.created_at or 0),
-                "Line Count": i.line_count,
-            }
-        )
-    table = tabulate(display_list, headers="keys", tablefmt="grid", showindex=True)
-
-    click.echo(table)
-
-
-@files.command()
-@click.pass_context
-@click.argument("id", type=str, required=True)
-def retrieve(ctx: click.Context, id: str) -> None:
-    """Upload file"""
-
-    client: Together = ctx.obj
-
-    response = client.files.retrieve(id=id)
-
-    click.echo(json.dumps(response.model_dump(exclude_none=True), indent=4))
-
-
-@files.command()
-@click.pass_context
-@click.argument("id", type=str, required=True)
-@click.option("--output", type=str, default=None, help="Output filename")
-def retrieve_content(ctx: click.Context, id: str, output: str) -> None:
-    """Retrieve file content and output to file"""
-
-    client: Together = ctx.obj
-
-    response = client.files.content(id=id)
-
-    if output:
-        with open(output, "wb") as f:
-            f.write(response.read())
-        click.echo(f"File saved to {output}")
-
-    else:
-        click.echo(response.read().decode("utf-8"))
-
-
-@files.command()
-@click.pass_context
-@click.argument("id", type=str, required=True)
-def delete(ctx: click.Context, id: str) -> None:
-    """Delete remote file"""
-
-    client: Together = ctx.obj
-
-    response = client.files.delete(id=id)
-
-    click.echo(json.dumps(response.model_dump(exclude_none=True), indent=4))
-
-
-@files.command()
-@click.pass_context
-@click.argument(
-    "file",
-    type=click.Path(exists=True, file_okay=True, resolve_path=True, readable=True, dir_okay=False),
-    required=True,
-)
-def check(_ctx: click.Context, file: pathlib.Path) -> None:
-    """Check file for issues"""
-
-    report = check_file(file)
-
-    click.echo(json.dumps(report, indent=4))

together/lib/cli/cli.py

@@ -1,73 +0,0 @@
-from __future__ import annotations
-
-import os
-from typing import Any
-
-import click
-
-import together
-from together._version import __version__
-from together._constants import DEFAULT_TIMEOUT
-from together.lib.cli.api.evals import evals
-from together.lib.cli.api.files import files
-from together.lib.cli.api.models import models
-from together.lib.cli.api.beta.beta import beta
-from together.lib.cli.api.endpoints import endpoints
-from together.lib.cli.api.fine_tuning import fine_tuning
-
-
-def print_version(ctx: click.Context, _params: Any, value: Any) -> None:
-    if not value or ctx.resilient_parsing:
-        return
-    click.echo(f"Version {__version__}")
-    ctx.exit()
-
-
-@click.group()
-@click.pass_context
-@click.option(
-    "--api-key",
-    type=str,
-    help="API Key. Defaults to environment variable `TOGETHER_API_KEY`",
-    default=os.getenv("TOGETHER_API_KEY"),
-)
-@click.option("--base-url", type=str, help="API Base URL. Defaults to Together AI endpoint.")
-@click.option("--timeout", type=int, help=f"Request timeout. Defaults to {DEFAULT_TIMEOUT} seconds")
-@click.option(
-    "--max-retries",
-    type=int,
-    help=f"Maximum number of HTTP retries.",
-)
-@click.option(
-    "--version",
-    is_flag=True,
-    callback=print_version,
-    expose_value=False,
-    is_eager=True,
-    help="Print version",
-)
-@click.option("--debug", help="Debug mode", is_flag=True)
-def main(
-    ctx: click.Context,
-    api_key: str | None,
-    base_url: str | None,
-    timeout: int | None,
-    debug: bool | None,
-    max_retries: int | None,
-) -> None:
-    """This is a sample CLI tool."""
-    os.environ.setdefault("TOGETHER_LOG", "debug" if debug else "info")
-    ctx.obj = together.Together(
-        api_key=api_key, base_url=base_url, timeout=timeout, max_retries=max_retries if max_retries is not None else 0
-    )
-
-
-main.add_command(files)
-main.add_command(fine_tuning)
-main.add_command(models)
-main.add_command(endpoints)
-main.add_command(evals)
-main.add_command(beta)
-
-if __name__ == "__main__":
-    main()

together/types/beta/cluster_create_response.py

@@ -1,9 +0,0 @@
-# File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
-
-from ..._models import BaseModel
-
-__all__ = ["ClusterCreateResponse"]
-
-
-class ClusterCreateResponse(BaseModel):
-    cluster_id: str

together/types/beta/cluster_update_response.py

@@ -1,9 +0,0 @@
-# File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
-
-from ..._models import BaseModel
-
-__all__ = ["ClusterUpdateResponse"]
-
-
-class ClusterUpdateResponse(BaseModel):
-    cluster_id: str

together/types/beta/clusters/storage_create_response.py

@@ -1,9 +0,0 @@
-# File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
-
-from ...._models import BaseModel
-
-__all__ = ["StorageCreateResponse"]
-
-
-class StorageCreateResponse(BaseModel):
-    volume_id: str

together-2.0.0a14.dist-info/entry_points.txt

@@ -1,2 +0,0 @@
-[console_scripts]
-together = together.lib.cli.cli:main