Merge branch 'main' of https://github.com/meta-llama/llama-stack into register_custom_model

This commit is contained in:
raspawar 2025-04-24 21:44:32 +05:30
commit 0990f60dad
74 changed files with 4854 additions and 1869 deletions

View file

@ -0,0 +1,40 @@
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
from pathlib import Path
from llama_stack.cli.stack._build import (
_run_stack_build_command_from_build_config,
)
from llama_stack.distribution.datatypes import BuildConfig, DistributionSpec
from llama_stack.distribution.utils.image_types import LlamaStackImageType
def test_container_build_passes_path(monkeypatch, tmp_path):
called_with = {}
def spy_build_image(cfg, build_file_path, image_name, template_or_config, run_config=None):
called_with["path"] = template_or_config
called_with["run_config"] = run_config
return 0
monkeypatch.setattr(
"llama_stack.cli.stack._build.build_image",
spy_build_image,
raising=True,
)
cfg = BuildConfig(
image_type=LlamaStackImageType.CONTAINER.value,
distribution_spec=DistributionSpec(providers={}, description=""),
)
_run_stack_build_command_from_build_config(cfg, image_name="dummy")
assert "path" in called_with
assert isinstance(called_with["path"], str)
assert Path(called_with["path"]).exists()
assert called_with["run_config"] is None

View file

@ -216,35 +216,48 @@ class TestNvidiaPostTraining(unittest.TestCase):
)
def test_get_training_job_status(self):
self.mock_make_request.return_value = {
"created_at": "2024-12-09T04:06:28.580220",
"updated_at": "2024-12-09T04:21:19.852832",
"status": "completed",
"steps_completed": 1210,
"epochs_completed": 2,
"percentage_done": 100.0,
"best_epoch": 2,
"train_loss": 1.718016266822815,
"val_loss": 1.8661999702453613,
}
customizer_status_to_job_status = [
("running", "in_progress"),
("completed", "completed"),
("failed", "failed"),
("cancelled", "cancelled"),
("pending", "scheduled"),
("unknown", "scheduled"),
]
job_id = "cust-JGTaMbJMdqjJU8WbQdN9Q2"
for customizer_status, expected_status in customizer_status_to_job_status:
with self.subTest(customizer_status=customizer_status, expected_status=expected_status):
self.mock_make_request.return_value = {
"created_at": "2024-12-09T04:06:28.580220",
"updated_at": "2024-12-09T04:21:19.852832",
"status": customizer_status,
"steps_completed": 1210,
"epochs_completed": 2,
"percentage_done": 100.0,
"best_epoch": 2,
"train_loss": 1.718016266822815,
"val_loss": 1.8661999702453613,
}
status = self.run_async(self.adapter.get_training_job_status(job_uuid=job_id))
job_id = "cust-JGTaMbJMdqjJU8WbQdN9Q2"
assert isinstance(status, NvidiaPostTrainingJobStatusResponse)
assert status.status.value == "completed"
assert status.steps_completed == 1210
assert status.epochs_completed == 2
assert status.percentage_done == 100.0
assert status.best_epoch == 2
assert status.train_loss == 1.718016266822815
assert status.val_loss == 1.8661999702453613
status = self.run_async(self.adapter.get_training_job_status(job_uuid=job_id))
self.mock_make_request.assert_called_once()
self._assert_request(
self.mock_make_request, "GET", f"/v1/customization/jobs/{job_id}/status", expected_params={"job_id": job_id}
)
assert isinstance(status, NvidiaPostTrainingJobStatusResponse)
assert status.status.value == expected_status
assert status.steps_completed == 1210
assert status.epochs_completed == 2
assert status.percentage_done == 100.0
assert status.best_epoch == 2
assert status.train_loss == 1.718016266822815
assert status.val_loss == 1.8661999702453613
self._assert_request(
self.mock_make_request,
"GET",
f"/v1/customization/jobs/{job_id}/status",
expected_params={"job_id": job_id},
)
def test_get_training_jobs(self):
job_id = "cust-JGTaMbJMdqjJU8WbQdN9Q2"

View file

@ -0,0 +1,55 @@
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
import asyncio
import pytest
from llama_stack.distribution.server.server import create_sse_event, sse_generator
@pytest.mark.asyncio
async def test_sse_generator_basic():
# An AsyncIterator wrapped in an Awaitable, just like our web methods
async def async_event_gen():
async def event_gen():
yield "Test event 1"
yield "Test event 2"
return event_gen()
sse_gen = sse_generator(async_event_gen())
assert sse_gen is not None
# Test that the events are streamed correctly
seen_events = []
async for event in sse_gen:
seen_events.append(event)
assert len(seen_events) == 2
assert seen_events[0] == create_sse_event("Test event 1")
assert seen_events[1] == create_sse_event("Test event 2")
@pytest.mark.asyncio
async def test_sse_generator_client_disconnected():
# An AsyncIterator wrapped in an Awaitable, just like our web methods
async def async_event_gen():
async def event_gen():
yield "Test event 1"
# Simulate a client disconnect before emitting event 2
raise asyncio.CancelledError()
return event_gen()
sse_gen = sse_generator(async_event_gen())
assert sse_gen is not None
# Start reading the events, ensuring this doesn't raise an exception
seen_events = []
async for event in sse_gen:
seen_events.append(event)
assert len(seen_events) == 1
assert seen_events[0] == create_sse_event("Test event 1")