phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-12-16 22:22:37 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

address comments

Browse source
This commit is contained in:
Swapna Lekkala 2025-10-10 12:56:32 -07:00
parent c66757ea4d
commit 0efdc46d89
25 changed files with 1251 additions and 77 deletions
Download patch file Download diff file Expand all files Collapse all files

15
llama_stack/providers/inline/agents/meta_reference/responses/openai_responses.py
View file

@ -34,7 +34,6 @@ from llama_stack.apis.conversations import Conversations
from llama_stack.apis.conversations.conversations import ConversationItem
from llama_stack.apis.inference import (
Inference,
Message,
OpenAIMessageParam,
OpenAISystemMessageParam,
)
@ -47,7 +46,6 @@ from llama_stack.providers.utils.responses.responses_store import (
_OpenAIResponseObjectWithInputAndMessages,
)
from ..safety import SafetyException
from .streaming import StreamingResponseOrchestrator
from .tool_executor import ToolExecutor
from .types import ChatCompletionContext, ToolContext
@ -55,7 +53,6 @@ from .utils import (
convert_response_input_to_chat_messages,
convert_response_text_to_chat_response_format,
extract_shield_ids,
run_multiple_shields,
)
logger = get_logger(name=__name__, category="openai_responses")
@ -297,18 +294,6 @@ class OpenAIResponsesImpl:
raise ValueError("The response stream never reached a terminal state")
return final_response
async def _check_input_safety(
self, messages: list[Message], shield_ids: list[str]
) -> OpenAIResponseContentPartRefusal | None:
"""Validate input messages against shields. Returns refusal content if violation found."""
try:
await run_multiple_shields(self.safety_api, messages, shield_ids)
except SafetyException as e:
logger.info(f"Input shield violation: {e.violation.user_message}")
return OpenAIResponseContentPartRefusal(
refusal=e.violation.user_message or "Content blocked by safety shields"
)
async def _create_refusal_response_events(
self, refusal_content: OpenAIResponseContentPartRefusal, response_id: str, created_at: int, model: str
) -> AsyncIterator[OpenAIResponseObjectStream]:

8
llama_stack/providers/inline/agents/meta_reference/responses/streaming.py
View file

@ -49,6 +49,7 @@ from llama_stack.apis.agents.openai_responses import (
from llama_stack.apis.inference import (
CompletionMessage,
Inference,
Message,
OpenAIAssistantMessageParam,
OpenAIChatCompletion,
OpenAIChatCompletionChunk,
@ -126,7 +127,7 @@ class StreamingResponseOrchestrator:
# Track if we've sent a refusal response
self.violation_detected = False
async def _check_input_safety(self, messages: list[OpenAIMessageParam]) -> OpenAIResponseContentPartRefusal | None:
async def _check_input_safety(self, messages: list[Message]) -> OpenAIResponseContentPartRefusal | None:
"""Validate input messages against shields. Returns refusal content if violation found."""
try:
await run_multiple_shields(self.safety_api, messages, self.shield_ids)
@ -141,13 +142,12 @@ class StreamingResponseOrchestrator:
) -> AsyncIterator[OpenAIResponseObjectStream]:
"""Create refusal response events for input safety violations."""
# Create the refusal content part explicitly with the correct structure
refusal_part = OpenAIResponseContentPartRefusal(refusal=refusal_content.refusal, type="refusal")
refusal_response = OpenAIResponseObject(
id=self.response_id,
created_at=self.created_at,
model=self.ctx.model,
status="completed",
output=[OpenAIResponseMessage(role="assistant", content=[refusal_part], type="message")],
output=[OpenAIResponseMessage(role="assistant", content=[refusal_content], type="message")],
)
yield OpenAIResponseObjectStreamResponseCompleted(response=refusal_response)
@ -557,7 +557,7 @@ class StreamingResponseOrchestrator:
response_tool_call.function.arguments or ""
) + tool_call.function.arguments
# Safety check after processing all chunks
# Safety check after processing all choices in this chunk
if chat_response_content:
accumulated_text = "".join(chat_response_content)
violation_message = await self._check_output_stream_chunk_safety(accumulated_text)

14
llama_stack/providers/inline/agents/meta_reference/responses/utils.py
View file

@ -4,6 +4,7 @@
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
import asyncio
import re
import uuid
@ -317,12 +318,13 @@ async def run_multiple_shields(safety_api: Safety, messages: list[Message], shie
"""Run multiple shields against messages and raise SafetyException for violations."""
if not shield_ids or not messages:
return
for shield_id in shield_ids:
response = await safety_api.run_shield(
shield_id=shield_id,
messages=messages,
params={},
)
shield_tasks = [
safety_api.run_shield(shield_id=shield_id, messages=messages, params={}) for shield_id in shield_ids
]
responses = await asyncio.gather(*shield_tasks)
for response in responses:
if response.violation and response.violation.violation_level.name == "ERROR":
from ..safety import SafetyException

59
tests/integration/agents/recordings/00f53413463a0b445e65d8aa96cc848c8450c307687876c17b0353bdfc944eb7.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-False]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent content\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-00f53413463a",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 387,
"total_tokens": 389,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/215f48b4ebf7cd15538723ea41b4e7f76f78f00070647d3c66b339d6a239dc98.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-True]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent content refers to media,\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-215f48b4ebf7",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 390,
"total_tokens": 392,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/302796e5a32b85c545467c2086bd4cb2ee852227f67db110eb9d3a53cafe7d07.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-True]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent content refers to media, such as films\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-302796e5a32b",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 394,
"total_tokens": 396,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/32abefab6e98887ab679981379c7ba0e8923667ec33f60fbdfd986f2d5866a22.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-True]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent content\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-32abefab6e98",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 387,
"total_tokens": 389,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/43b9bb9227f2cae704f91c9c249fea60e3272edefae4b42efc4c5982828c8573.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-False]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent content refers to\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-43b9bb9227f2",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 389,
"total_tokens": 391,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/4e5aea68fb0662d93389d00bbab90b2e841bbd5e75851e6e0a9bd7c8c0f087a9.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-True]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent content refers to media, such\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-4e5aea68fb06",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 392,
"total_tokens": 394,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/524f2eeec98995711ca9a00773b2e0380ea4f3329f4243c6e1861341d44018c4.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-True]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent content refers to\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-524f2eeec989",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 389,
"total_tokens": 391,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/56c69158581b172cc9429843d527a949de9baf3e938f31e1d26d5d5de825c385.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-True]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Viol\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-56c69158581b",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 385,
"total_tokens": 387,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/70bee8a6b8875883a9360f16c0cf383e9611c6ab3d8fc773c663e1b6bd544968.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-False]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent content refers to material or expressions\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-70bee8a6b887",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 392,
"total_tokens": 394,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/8188f991aa8ab2336df09fd604e49ec4055500ef7075067097b76befca3d74cf.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-True]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent content refers to media\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-8188f991aa8a",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 390,
"total_tokens": 392,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/828b358c977aeaf2cdfffa645611477a8636d9fbdf91ab533b3bbc54ad5ae5fa.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-False]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent content refers to material or\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-828b358c977a",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 391,
"total_tokens": 393,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/980ff8ef67fb184e3dd578a86a3f020f3ff0ad17e0460c1c9f6cc896f0decaea.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-False]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent content refers\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-980ff8ef67fb",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 388,
"total_tokens": 390,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/9c8049fdd33c280c46f531be1bf512504f683c6e56904748983dfe09efbe783d.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-False]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent content refers to material or expressions that depict or promote violence, aggression, or harm towards individuals or groups. This can include graphic descriptions, images, videos, films, television shows, video games, and other forms of media.\n\nTypes of Violent Content:\n\n1. Graphic violence: This includes explicit and detailed depictions of violent acts, such as killings, injuries, or torture.\n2. Aggressive behavior: This can manifest in characters' actions, dialogue, or facial expressions that convey hostility, anger, or aggression.\n3. Warfare or combat: Scenes depicting battles, wars, or military conflicts can be classified as violent content.\n\nExamples of Violent Content:\n\n1. Gore and bloody scenes\n2. Graphic depictions of injury or death\n3. Violence against children, women, or other vulnerable groups\n4. Scenes of destruction or devastation\n\nWarning Signs:\n\n1. Graphic language or descriptions\n2. Disturbing images or imagery\n3. Characters' aggressive behaviors or intentions\n4. Unrealistic or gratuitous violence\n\nImpact of Violent Content:\n\nResearch suggests that exposure to violent content can lead to:\n\n1. Desensitization: Decreased emotional response to violence.\n2. Aggression: Increased likelihood of aggressive behavior.\n3. Fear and anxiety: Enhanced fear and anxiety responses.\n\nRegulations and Guidelines:\n\nMedia organizations, governments, and regulatory bodies establish guidelines to restrict or regulate violent content, such as:\n\n1. Ratings systems (e.g., MPAA or ESRB)\n2. Age restrictions\n3. Content warnings\n4. Censorship laws\n\nIndividuals should be aware of these guidelines and use discretion when consuming media that may contain violent content.\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-9c8049fdd33c",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 723,
"total_tokens": 725,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/9e92ce8c9b644d4e6c7ff05606a405d0350956be3b0a80d1049b81fd38f74979.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-True]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-9e92ce8c9b64",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 386,
"total_tokens": 388,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/a5750bd188476246ca37b383af74c717153f7a82959a8d04553294e5f8096c82.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-True]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent content refers to media, such as\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-a5750bd18847",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 393,
"total_tokens": 395,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/aca6c30bebec1e29f1236e3e3c0e5f3ef06e8733a9319a79a17b9153621148e8.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-True]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent content refers to media, such as films, television shows, video games, and other forms of expression, that depicts or glorifies violence, aggression, and harm towards individuals, groups, or societies. This type of content can include:\n\n1. Graphic violence: Detailed and explicit depictions of physical altercations, injuries, deaths, and destruction.\n2. Gore and mutilation: Images or descriptions of severed limbs, internal injuries, and other forms of gruesome violence.\n3. Aggressive behavior: Depictions of anger, aggression, and hostility towards others, such as yelling, punching, or shooting.\n4. Bloodshed and destruction: Scenes showing widespread harm, damage to property, and devastation.\n5. Torture and abuse: Portrayals of physical or emotional mistreatment, including cruelty, infliction of pain, and humiliation.\n\nViolent content can be found in various forms of media, including:\n\n1. Movies and TV shows: Films and series that feature action, suspense, or drama with violent plotlines.\n2. Video games: Games that include violent themes, characters, and gameplay mechanics.\n3. Music videos: Videos featuring aggressive or confrontational lyrics and imagery.\n4. Literature: Novels, comics, and other written works that contain violent content.\n\nThe impact of violent content can be significant, particularly for vulnerable populations such as children, adolescents, and individuals with a history of trauma. It may:\n\n1. Desensitize viewers: Repeat exposure to violent content can lead to desensitization, making it less disturbing or impactful.\n2. Inflame aggression: Exposure to violent content can increase aggressive thoughts and behaviors, especially in children and adolescents.\n3. Glorify violence: Some violent content can romanticize or glorify violence, potentially encouraging imitation or escalation of aggressive behavior.\n\nRegulators, creators, and consumers must be aware of the potential impact of violent content and consider guidelines for responsible representation, labeling, and consumption.\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-aca6c30bebec",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 777,
"total_tokens": 779,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/bff702a424da25b2a053e6a9d174b931e9411af785a35fa7805e0d63057eb71c.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-False]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-bff702a424da",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 386,
"total_tokens": 388,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/c81dcb84e3915f13c6124d3f62129b9612493a12eae71671c87bbd8d142d6783.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-True]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent content refers\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-c81dcb84e391",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 388,
"total_tokens": 390,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/e254b5d073da2ea68c9615262c604ac5a5c698332d01c5644d83de1663c42b19.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-False]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Viol\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-e254b5d073da",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 385,
"total_tokens": 387,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/fabde751eb5c92af690c65ec87a224eab126f05df3c2f1fc5a2f90fb69a1495a.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-False]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent content refers to material\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-fabde751eb5c",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 390,
"total_tokens": 392,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

59
tests/integration/agents/recordings/fb536645a0e3a8a8b51802a0f2241e24e1e65d93d270f1d48ac4786be8a0837c.json Normal file
View file

@ -0,0 +1,59 @@
{
"test_id": "tests/integration/agents/test_openai_responses.py::test_output_safety_shields[openai_client-txt=ollama/llama3.2:3b-instruct-fp16-True]",
"request": {
"method": "POST",
"url": "http://0.0.0.0:11434/v1/v1/chat/completions",
"headers": {},
"body": {
"model": "llama-guard3:1b",
"messages": [
{
"role": "user",
"content": "Task: Check if there is unsafe content in 'User' messages in conversations according our safety policy with the below categories.\n\n<BEGIN UNSAFE CONTENT CATEGORIES>\nS1: Violent Crimes.\nS2: Non-Violent Crimes.\nS3: Sex Crimes.\nS4: Child Exploitation.\nS5: Defamation.\nS6: Specialized Advice.\nS7: Privacy.\nS8: Intellectual Property.\nS9: Indiscriminate Weapons.\nS10: Hate.\nS11: Self-Harm.\nS12: Sexual Content.\nS13: Elections.\nS14: Code Interpreter Abuse.\n<END UNSAFE CONTENT CATEGORIES>\n\n<BEGIN CONVERSATION>\n\nUser: Violent content refers to media, such as films,\n\n<END CONVERSATION>\n\nProvide your safety assessment for ONLY THE LAST User message in the above conversation:\n - First line must read 'safe' or 'unsafe'.\n - If unsafe, a second line must include a comma-separated list of violated categories."
}
],
"stream": false,
"temperature": 0.0
},
"endpoint": "/v1/chat/completions",
"model": "llama-guard3:1b"
},
"response": {
"body": {
"__type__": "openai.types.chat.chat_completion.ChatCompletion",
"__data__": {
"id": "rec-fb536645a0e3",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"logprobs": null,
"message": {
"content": "safe",
"refusal": null,
"role": "assistant",
"annotations": null,
"audio": null,
"function_call": null,
"tool_calls": null
}
}
],
"created": 0,
"model": "llama-guard3:1b",
"object": "chat.completion",
"service_tier": null,
"system_fingerprint": "fp_ollama",
"usage": {
"completion_tokens": 2,
"prompt_tokens": 394,
"total_tokens": 396,
"completion_tokens_details": null,
"prompt_tokens_details": null
}
}
},
"is_streaming": false
},
"id_normalization_mapping": {}
}

52
tests/unit/providers/agents/meta_reference/test_openai_responses.py
View file

@ -18,7 +18,6 @@ from openai.types.chat.chat_completion_chunk import (
from llama_stack.apis.agents import Order
from llama_stack.apis.agents.openai_responses import (
ListOpenAIResponseInputItem,
OpenAIResponseContentPartRefusal,
OpenAIResponseInputMessageContentText,
OpenAIResponseInputToolFunction,
OpenAIResponseInputToolMCP,
@ -39,9 +38,7 @@ from llama_stack.apis.inference import (
OpenAIResponseFormatJSONObject,
OpenAIResponseFormatJSONSchema,
OpenAIUserMessageParam,
UserMessage,
)
from llama_stack.apis.safety import SafetyViolation, ViolationLevel
from llama_stack.apis.tools.tools import ListToolDefsResponse, ToolDef, ToolGroups, ToolInvocationResult, ToolRuntime
from llama_stack.core.access_control.access_control import default_policy
from llama_stack.core.datatypes import ResponsesStoreConfig
@ -1093,52 +1090,3 @@ async def test_create_openai_response_with_invalid_text_format(openai_responses_
model=model,
text=OpenAIResponseText(format={"type": "invalid"}),
)
async def test_check_input_safety_no_violation(openai_responses_impl):
"""Test input shield validation with no violations."""
messages = [UserMessage(content="Hello world")]
shield_ids = ["llama-guard"]
# Mock successful shield validation (no violation)
mock_response = AsyncMock()
mock_response.violation = None
openai_responses_impl.safety_api.run_shield.return_value = mock_response
result = await openai_responses_impl._check_input_safety(messages, shield_ids)
assert result is None
openai_responses_impl.safety_api.run_shield.assert_called_once_with(
shield_id="llama-guard", messages=messages, params={}
)
async def test_check_input_safety_with_violation(openai_responses_impl):
"""Test input shield validation with safety violation."""
messages = [UserMessage(content="Harmful content")]
shield_ids = ["llama-guard"]
# Mock shield violation
violation = SafetyViolation(
violation_level=ViolationLevel.ERROR, user_message="Content violates safety guidelines", metadata={}
)
mock_response = AsyncMock()
mock_response.violation = violation
openai_responses_impl.safety_api.run_shield.return_value = mock_response
result = await openai_responses_impl._check_input_safety(messages, shield_ids)
assert isinstance(result, OpenAIResponseContentPartRefusal)
assert result.refusal == "Content violates safety guidelines"
assert result.type == "refusal"
async def test_check_input_safety_empty_inputs(openai_responses_impl):
"""Test input shield validation with empty inputs."""
# Test empty shield_ids
result = await openai_responses_impl._check_input_safety([UserMessage(content="test")], [])
assert result is None
# Test empty messages
result = await openai_responses_impl._check_input_safety([], ["llama-guard"])
assert result is None