report

tests/client-sdk/agents/test_agents.py

@@ -80,7 +80,7 @@ class TestClientTool(ClientTool):
 
 
 @pytest.fixture(scope="session")
-def model_id(llama_stack_client):
+def text_model_id(llama_stack_client):
     available_models = [
         model.identifier
         for model in llama_stack_client.models.list()
@@ -92,14 +92,14 @@ def model_id(llama_stack_client):
 
 
 @pytest.fixture(scope="session")
-def agent_config(llama_stack_client, model_id):
+def agent_config(llama_stack_client, text_model_id):
     available_shields = [
         shield.identifier for shield in llama_stack_client.shields.list()
     ]
     available_shields = available_shields[:1]
     print(f"Using shield: {available_shields}")
     agent_config = AgentConfig(
-        model=model_id,
+        model=text_model_id,
         instructions="You are a helpful assistant",
         sampling_params={
             "strategy": {

tests/client-sdk/conftest.py

@@ -10,12 +10,28 @@ import pytest
 from llama_stack import LlamaStackAsLibraryClient
 from llama_stack.providers.tests.env import get_env_or_fail
 from llama_stack_client import LlamaStackClient
+from report import Report
 
 
 def pytest_configure(config):
     config.option.tbstyle = "short"
     config.option.disable_warnings = True
 
+    if config.getoption("--output"):
+        config.pluginmanager.register(Report())
+
+
+def pytest_addoption(parser):
+    parser.addoption(
+        "--output",
+        action="store_false",
+        help="Knob to determine if we should generate report, e.g. --output=True",
+    )
+
+
+TEXT_MODEL = "meta-llama/Llama-3.1-8B-Instruct"
+INFERENCE_MODEL = "meta-llama/Llama-3.2-11B-Vision-Instruct"
+
 
 @pytest.fixture(scope="session")
 def provider_data():

tests/client-sdk/inference/test_inference.py

@@ -82,7 +82,7 @@ def base64_image_url():
         return base64_url
 
 
-def test_completion_non_streaming(llama_stack_client, text_model_id):
+def test_text_completion_non_streaming(llama_stack_client, text_model_id):
     response = llama_stack_client.inference.completion(
         content="Complete the sentence using one word: Roses are red, violets are ",
         stream=False,
@@ -94,7 +94,7 @@ def test_completion_non_streaming(llama_stack_client, text_model_id):
     assert "blue" in response.content.lower().strip()
 
 
-def test_completion_streaming(llama_stack_client, text_model_id):
+def test_text_completion_streaming(llama_stack_client, text_model_id):
     response = llama_stack_client.inference.completion(
         content="Complete the sentence using one word: Roses are red, violets are ",
         stream=True,
@@ -147,7 +147,7 @@ def test_completion_log_probs_streaming(llama_stack_client, text_model_id):
             assert not chunk.logprobs, "Logprobs should be empty"
 
 
-def test_completion_structured_output(
+def test_text_completion_structured_output(
     llama_stack_client, text_model_id, inference_provider_type
 ):
     user_input = """

tests/client-sdk/metadata.py

@@ -0,0 +1,50 @@
+# 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.
+
+
+INFERENCE_API_CAPA_TEST_MAP = {
+    "chat_completion": {
+        "streaming": [
+            "test_text_chat_completion_streaming",
+            "test_image_chat_completion_streaming",
+        ],
+        "non_streaming": [
+            "test_image_chat_completion_non_streaming",
+            "test_text_chat_completion_non_streaming",
+        ],
+        "tool_calling": [
+            "test_text_chat_completion_with_tool_calling_and_streaming",
+            "test_text_chat_completion_with_tool_calling_and_non_streaming",
+        ],
+    },
+    "completion": {
+        "streaming": ["test_text_completion_streaming"],
+        "non_streaming": ["test_text_completion_non_streaming"],
+        "structured_output": ["test_text_completion_structured_output"],
+    },
+}
+
+MEMORY_API_CAPA_TEST_MAP = {
+    "insert and query": {
+        "inline": ["test_memory_bank_insert_inline_and_query"],
+        "url": ["test_memory_bank_insert_from_url_and_query"],
+    }
+}
+
+AGENTS_API_CAPA_TEST_MAP = {
+    "create_agent_turn": {
+        "rag": ["test_rag_agent"],
+        "custom_tool": ["test_custom_tool"],
+        "code_execution": ["test_code_execution"],
+    }
+}
+
+
+API_CAPA_MAPS = {
+    "inference": INFERENCE_API_CAPA_TEST_MAP,
+    "memory": MEMORY_API_CAPA_TEST_MAP,
+    "agents": AGENTS_API_CAPA_TEST_MAP,
+}

tests/client-sdk/report.py

@@ -0,0 +1,217 @@
+# 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 os
+from collections import defaultdict
+from pathlib import Path
+
+import pytest
+from llama_models.datatypes import CoreModelId
+from llama_models.sku_list import all_registered_models
+
+from llama_stack.distribution.library_client import LlamaStackAsLibraryClient
+from metadata import API_CAPA_MAPS
+
+from pytest import CollectReport
+
+
+SUPPORTED_MODELS = {
+    "ollama": set(
+        [
+            CoreModelId.llama3_1_8b_instruct.value,
+            CoreModelId.llama3_1_8b_instruct.value,
+            CoreModelId.llama3_1_70b_instruct.value,
+            CoreModelId.llama3_1_70b_instruct.value,
+            CoreModelId.llama3_1_405b_instruct.value,
+            CoreModelId.llama3_1_405b_instruct.value,
+            CoreModelId.llama3_2_1b_instruct.value,
+            CoreModelId.llama3_2_1b_instruct.value,
+            CoreModelId.llama3_2_3b_instruct.value,
+            CoreModelId.llama3_2_3b_instruct.value,
+            CoreModelId.llama3_2_11b_vision_instruct.value,
+            CoreModelId.llama3_2_11b_vision_instruct.value,
+            CoreModelId.llama3_2_90b_vision_instruct.value,
+            CoreModelId.llama3_2_90b_vision_instruct.value,
+            CoreModelId.llama3_3_70b_instruct.value,
+            CoreModelId.llama_guard_3_8b.value,
+            CoreModelId.llama_guard_3_1b.value,
+        ]
+    ),
+    "fireworks": set(
+        [
+            CoreModelId.llama3_1_8b_instruct.value,
+            CoreModelId.llama3_1_70b_instruct.value,
+            CoreModelId.llama3_1_405b_instruct.value,
+            CoreModelId.llama3_2_1b_instruct.value,
+            CoreModelId.llama3_2_3b_instruct.value,
+            CoreModelId.llama3_2_11b_vision_instruct.value,
+            CoreModelId.llama3_2_90b_vision_instruct.value,
+            CoreModelId.llama3_3_70b_instruct.value,
+            CoreModelId.llama_guard_3_8b.value,
+            CoreModelId.llama_guard_3_11b_vision.value,
+        ]
+    ),
+    "together": set(
+        [
+            CoreModelId.llama3_1_8b_instruct.value,
+            CoreModelId.llama3_1_70b_instruct.value,
+            CoreModelId.llama3_1_405b_instruct.value,
+            CoreModelId.llama3_2_3b_instruct.value,
+            CoreModelId.llama3_2_11b_vision_instruct.value,
+            CoreModelId.llama3_2_90b_vision_instruct.value,
+            CoreModelId.llama3_3_70b_instruct.value,
+            CoreModelId.llama_guard_3_8b.value,
+            CoreModelId.llama_guard_3_11b_vision.value,
+        ]
+    ),
+}
+
+
+class Report:
+
+    def __init__(self):
+        config_file = os.environ.get("LLAMA_STACK_CONFIG")
+        if not config_file:
+            raise ValueError(
+                "Currently we only support generating report for LlamaStackClientLibrary distributions"
+            )
+        config_path = Path(config_file)
+        self.output_path = Path(config_path.parent / "report.md")
+        self.client = LlamaStackAsLibraryClient(
+            config_file,
+            provider_data=None,
+            skip_logger_removal=True,
+        )
+        self.image_name = self.client.async_client.config.image_name
+        self.report_data = defaultdict(dict)
+        # test function -> test nodeid
+        self.test_data = dict()
+        self.test_name_to_nodeid = defaultdict(list)
+
+    @pytest.hookimpl(tryfirst=True)
+    def pytest_runtest_logreport(self, report):
+        # This hook is called in several phases, including setup, call and teardown
+        # The test is considered failed / error if any of the outcomes is not "Passed"
+        outcome = self._process_outcome(report)
+        if report.nodeid not in self.test_data:
+            self.test_data[report.nodeid] = outcome
+        elif self.test_data[report.nodeid] != outcome and outcome != "Passed":
+            self.test_data[report.nodeid] = outcome
+
+    def pytest_sessionfinish(self, session):
+        report = []
+        report.append(f"# Report for {self.image_name} distribution")
+        report.append("\n## Supported Models: ")
+
+        header = f"| Model Descriptor | {self.image_name} |"
+        dividor = "|:---|"
+
+        report.append(header)
+        report.append(dividor)
+
+        rows = []
+        for model in all_registered_models():
+            if (
+                "Instruct" not in model.core_model_id.value
+                and "Guard" not in model.core_model_id.value
+            ):
+                continue
+            row = f"| {model.core_model_id.value} |"
+            if model.core_model_id.value in SUPPORTED_MODELS[self.image_name]:
+                row += " ✅ |"
+            else:
+                row += " ❌ |"
+            rows.append(row)
+        report.extend(rows)
+
+        report.append("\n## Inference: ")
+        test_table = [
+            "| Model | API | Capability | Test | Status |",
+            "|:----- |:-----|:-----|:-----|:-----|",
+        ]
+        for api, capa_map in API_CAPA_MAPS["inference"].items():
+            for capa, tests in capa_map.items():
+                vision_tests = filter(lambda test_name: "image" in test_name, tests)
+                text_tests = filter(lambda test_name: "text" in test_name, tests)
+
+                for test_name in text_tests:
+                    test_nodeids = self.test_name_to_nodeid[test_name]
+                    assert len(test_nodeids) > 0
+                    # There might be more than one parametrizations for the same test function. We take
+                    # the result of the first one for now. Ideally we should mark the test as failed if
+                    # any of the parametrizations failed.
+                    test_table.append(
+                        f"| Text | /{api} | {capa} | {test_name} | {self.test_data[test_nodeids[0]]} |"
+                    )
+
+                for test_name in vision_tests:
+                    test_nodeids = self.test_name_to_nodeid[test_name]
+                    assert len(test_nodeids) > 0
+                    test_table.append(
+                        f"| Vision | /{api} | {capa} | {test_name} | {self.test_data[test_nodeids[0]]} |"
+                    )
+
+        report.extend(test_table)
+
+        report.append("\n## Memory: ")
+        test_table = [
+            "| API | Capability | Test | Status |",
+            "|:-----|:-----|:-----|:-----|",
+        ]
+        for api, capa_map in API_CAPA_MAPS["memory"].items():
+            for capa, tests in capa_map.items():
+                for test_name in tests:
+                    test_nodeids = self.test_name_to_nodeid[test_name]
+                    assert len(test_nodeids) > 0
+                    test_table.append(
+                        f"| {api} | {capa} | {test_name} | {self.test_data[test_nodeids[0]]} |"
+                    )
+        report.extend(test_table)
+        report.append("\n## Agent: ")
+        test_table = [
+            "| API | Capability | Test | Status |",
+            "|:-----|:-----|:-----|:-----|",
+        ]
+        for api, capa_map in API_CAPA_MAPS["agents"].items():
+            for capa, tests in capa_map.items():
+                for test_name in tests:
+                    test_nodeids = self.test_name_to_nodeid[test_name]
+                    assert len(test_nodeids) > 0
+                    test_table.append(
+                        f"| {api} | {capa} | {test_name} | {self.test_data[test_nodeids[0]]} |"
+                    )
+        report.extend(test_table)
+        output_file = self.output_path
+        output_file.write_text("\n".join(report))
+        print(f"\nReport generated: {output_file.absolute()}")
+
+    def pytest_runtest_makereport(self, item, call):
+        func_name = getattr(item, "originalname", item.name)
+        self.test_name_to_nodeid[func_name].append(item.nodeid)
+
+    def _print_result_icon(self, result):
+        if result == "Passed":
+            return "✅"
+        else:
+            #  result == "Failed" or result == "Error":
+            return "❌"
+
+    def _process_outcome(self, report: CollectReport):
+        if self._is_error(report):
+            return "Error"
+        if hasattr(report, "wasxfail"):
+            if report.outcome in ["passed", "failed"]:
+                return "XPassed"
+            if report.outcome == "skipped":
+                return "XFailed"
+        return report.outcome.capitalize()
+
+    def _is_error(self, report: CollectReport):
+        return (
+            report.when in ["setup", "teardown", "collect"]
+            and report.outcome == "failed"
+        )