Compare commits
3 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
|
9d05f77a9b | ||
|
|
23e093fc75 | ||
|
|
f77ce4dff2 |
@@ -1396,8 +1396,6 @@ def normalize_anthropic_response(
|
||||
"tool_use": "tool_calls",
|
||||
"max_tokens": "length",
|
||||
"stop_sequence": "stop",
|
||||
"refusal": "content_filter",
|
||||
"model_context_window_exceeded": "length",
|
||||
}
|
||||
finish_reason = stop_reason_map.get(response.stop_reason, "stop")
|
||||
|
||||
@@ -1411,42 +1409,3 @@ def normalize_anthropic_response(
|
||||
),
|
||||
finish_reason,
|
||||
)
|
||||
|
||||
|
||||
def normalize_anthropic_response_v2(
|
||||
response,
|
||||
strip_tool_prefix: bool = False,
|
||||
) -> "NormalizedResponse":
|
||||
"""Normalize Anthropic response to NormalizedResponse.
|
||||
|
||||
Wraps the existing normalize_anthropic_response() and maps its output
|
||||
to the shared transport types. This allows incremental migration
|
||||
without disturbing the legacy call sites.
|
||||
"""
|
||||
from agent.transports.types import NormalizedResponse, build_tool_call
|
||||
|
||||
assistant_msg, finish_reason = normalize_anthropic_response(response, strip_tool_prefix)
|
||||
|
||||
tool_calls = None
|
||||
if assistant_msg.tool_calls:
|
||||
tool_calls = [
|
||||
build_tool_call(
|
||||
id=tc.id,
|
||||
name=tc.function.name,
|
||||
arguments=tc.function.arguments,
|
||||
)
|
||||
for tc in assistant_msg.tool_calls
|
||||
]
|
||||
|
||||
provider_data = {}
|
||||
if getattr(assistant_msg, "reasoning_details", None):
|
||||
provider_data["reasoning_details"] = assistant_msg.reasoning_details
|
||||
|
||||
return NormalizedResponse(
|
||||
content=assistant_msg.content,
|
||||
tool_calls=tool_calls,
|
||||
finish_reason=finish_reason,
|
||||
reasoning=getattr(assistant_msg, "reasoning", None),
|
||||
usage=None,
|
||||
provider_data=provider_data or None,
|
||||
)
|
||||
|
||||
@@ -1,57 +0,0 @@
|
||||
"""Transport layer types and registry for provider response normalization.
|
||||
|
||||
Usage:
|
||||
from agent.transports import get_transport
|
||||
transport = get_transport("anthropic_messages")
|
||||
result = transport.normalize_response(raw_response)
|
||||
"""
|
||||
|
||||
from agent.transports.types import ( # noqa: F401
|
||||
NormalizedResponse,
|
||||
ToolCall,
|
||||
Usage,
|
||||
build_tool_call,
|
||||
map_finish_reason,
|
||||
)
|
||||
|
||||
_REGISTRY: dict = {}
|
||||
|
||||
|
||||
def register_transport(api_mode: str, transport_cls: type) -> None:
|
||||
"""Register a transport class for an api_mode string."""
|
||||
_REGISTRY[api_mode] = transport_cls
|
||||
|
||||
|
||||
def get_transport(api_mode: str):
|
||||
"""Get a transport instance for the given api_mode.
|
||||
|
||||
Returns None if no transport is registered for this api_mode.
|
||||
This allows gradual migration — call sites can check for None
|
||||
and fall back to the legacy code path.
|
||||
"""
|
||||
if not _REGISTRY:
|
||||
_discover_transports()
|
||||
cls = _REGISTRY.get(api_mode)
|
||||
if cls is None:
|
||||
return None
|
||||
return cls()
|
||||
|
||||
|
||||
def _discover_transports() -> None:
|
||||
"""Import all transport modules to trigger auto-registration."""
|
||||
try:
|
||||
import agent.transports.anthropic # noqa: F401
|
||||
except ImportError:
|
||||
pass
|
||||
try:
|
||||
import agent.transports.codex # noqa: F401
|
||||
except ImportError:
|
||||
pass
|
||||
try:
|
||||
import agent.transports.chat_completions # noqa: F401
|
||||
except ImportError:
|
||||
pass
|
||||
try:
|
||||
import agent.transports.bedrock # noqa: F401
|
||||
except ImportError:
|
||||
pass
|
||||
@@ -1,95 +0,0 @@
|
||||
"""Anthropic Messages API transport.
|
||||
|
||||
Delegates to the existing adapter functions in agent/anthropic_adapter.py.
|
||||
This transport owns format conversion and normalization — NOT client lifecycle.
|
||||
"""
|
||||
|
||||
from typing import Any, Dict, List, Optional
|
||||
|
||||
from agent.transports.base import ProviderTransport
|
||||
from agent.transports.types import NormalizedResponse
|
||||
|
||||
|
||||
class AnthropicTransport(ProviderTransport):
|
||||
"""Transport for api_mode='anthropic_messages'."""
|
||||
|
||||
@property
|
||||
def api_mode(self) -> str:
|
||||
return "anthropic_messages"
|
||||
|
||||
def convert_messages(self, messages: List[Dict[str, Any]], **kwargs) -> Any:
|
||||
from agent.anthropic_adapter import convert_messages_to_anthropic
|
||||
|
||||
base_url = kwargs.get("base_url")
|
||||
return convert_messages_to_anthropic(messages, base_url=base_url)
|
||||
|
||||
def convert_tools(self, tools: List[Dict[str, Any]]) -> Any:
|
||||
from agent.anthropic_adapter import convert_tools_to_anthropic
|
||||
|
||||
return convert_tools_to_anthropic(tools)
|
||||
|
||||
def build_kwargs(
|
||||
self,
|
||||
model: str,
|
||||
messages: List[Dict[str, Any]],
|
||||
tools: Optional[List[Dict[str, Any]]] = None,
|
||||
**params,
|
||||
) -> Dict[str, Any]:
|
||||
from agent.anthropic_adapter import build_anthropic_kwargs
|
||||
|
||||
return build_anthropic_kwargs(
|
||||
model=model,
|
||||
messages=messages,
|
||||
tools=tools,
|
||||
max_tokens=params.get("max_tokens", 16384),
|
||||
reasoning_config=params.get("reasoning_config"),
|
||||
tool_choice=params.get("tool_choice"),
|
||||
is_oauth=params.get("is_oauth", False),
|
||||
preserve_dots=params.get("preserve_dots", False),
|
||||
context_length=params.get("context_length"),
|
||||
base_url=params.get("base_url"),
|
||||
fast_mode=params.get("fast_mode", False),
|
||||
)
|
||||
|
||||
def normalize_response(self, response: Any, **kwargs) -> NormalizedResponse:
|
||||
from agent.anthropic_adapter import normalize_anthropic_response_v2
|
||||
|
||||
strip_tool_prefix = kwargs.get("strip_tool_prefix", False)
|
||||
return normalize_anthropic_response_v2(response, strip_tool_prefix=strip_tool_prefix)
|
||||
|
||||
def validate_response(self, response: Any) -> bool:
|
||||
if response is None:
|
||||
return False
|
||||
content_blocks = getattr(response, "content", None)
|
||||
if not isinstance(content_blocks, list):
|
||||
return False
|
||||
if not content_blocks:
|
||||
return False
|
||||
return True
|
||||
|
||||
def extract_cache_stats(self, response: Any):
|
||||
usage = getattr(response, "usage", None)
|
||||
if usage is None:
|
||||
return None
|
||||
cached = getattr(usage, "cache_read_input_tokens", 0) or 0
|
||||
written = getattr(usage, "cache_creation_input_tokens", 0) or 0
|
||||
if cached or written:
|
||||
return {"cached_tokens": cached, "creation_tokens": written}
|
||||
return None
|
||||
|
||||
_STOP_REASON_MAP = {
|
||||
"end_turn": "stop",
|
||||
"tool_use": "tool_calls",
|
||||
"max_tokens": "length",
|
||||
"stop_sequence": "stop",
|
||||
"refusal": "content_filter",
|
||||
"model_context_window_exceeded": "length",
|
||||
}
|
||||
|
||||
def map_finish_reason(self, raw_reason: str) -> str:
|
||||
return self._STOP_REASON_MAP.get(raw_reason, "stop")
|
||||
|
||||
|
||||
from agent.transports import register_transport # noqa: E402
|
||||
|
||||
register_transport("anthropic_messages", AnthropicTransport)
|
||||
@@ -1,61 +0,0 @@
|
||||
"""Abstract base for provider transports.
|
||||
|
||||
A transport owns the data path for one api_mode:
|
||||
convert_messages → convert_tools → build_kwargs → normalize_response
|
||||
|
||||
It does NOT own: client construction, streaming, credential refresh,
|
||||
prompt caching, interrupt handling, or retry logic. Those stay on AIAgent.
|
||||
"""
|
||||
|
||||
from abc import ABC, abstractmethod
|
||||
from typing import Any, Dict, List, Optional
|
||||
|
||||
from agent.transports.types import NormalizedResponse
|
||||
|
||||
|
||||
class ProviderTransport(ABC):
|
||||
"""Base class for provider-specific format conversion and normalization."""
|
||||
|
||||
@property
|
||||
@abstractmethod
|
||||
def api_mode(self) -> str:
|
||||
"""The api_mode string this transport handles."""
|
||||
...
|
||||
|
||||
@abstractmethod
|
||||
def convert_messages(self, messages: List[Dict[str, Any]], **kwargs) -> Any:
|
||||
"""Convert OpenAI-format messages to provider-native format."""
|
||||
...
|
||||
|
||||
@abstractmethod
|
||||
def convert_tools(self, tools: List[Dict[str, Any]]) -> Any:
|
||||
"""Convert OpenAI-format tool definitions to provider-native format."""
|
||||
...
|
||||
|
||||
@abstractmethod
|
||||
def build_kwargs(
|
||||
self,
|
||||
model: str,
|
||||
messages: List[Dict[str, Any]],
|
||||
tools: Optional[List[Dict[str, Any]]] = None,
|
||||
**params,
|
||||
) -> Dict[str, Any]:
|
||||
"""Build the complete provider kwargs dict."""
|
||||
...
|
||||
|
||||
@abstractmethod
|
||||
def normalize_response(self, response: Any, **kwargs) -> NormalizedResponse:
|
||||
"""Normalize a raw provider response to the shared NormalizedResponse type."""
|
||||
...
|
||||
|
||||
def validate_response(self, response: Any) -> bool:
|
||||
"""Optional structural validation for raw responses."""
|
||||
return True
|
||||
|
||||
def extract_cache_stats(self, response: Any) -> Optional[Dict[str, int]]:
|
||||
"""Optional cache stats extraction."""
|
||||
return None
|
||||
|
||||
def map_finish_reason(self, raw_reason: str) -> str:
|
||||
"""Optional stop-reason mapping. Defaults to passthrough."""
|
||||
return raw_reason
|
||||
@@ -1,58 +0,0 @@
|
||||
"""Shared types for normalized provider responses."""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import json
|
||||
from dataclasses import dataclass, field
|
||||
from typing import Any, Dict, List, Optional
|
||||
|
||||
|
||||
@dataclass
|
||||
class ToolCall:
|
||||
"""A normalized tool call from any provider."""
|
||||
|
||||
id: Optional[str]
|
||||
name: str
|
||||
arguments: str
|
||||
provider_data: Optional[Dict[str, Any]] = field(default=None, repr=False)
|
||||
|
||||
|
||||
@dataclass
|
||||
class Usage:
|
||||
"""Token usage from an API response."""
|
||||
|
||||
prompt_tokens: int = 0
|
||||
completion_tokens: int = 0
|
||||
total_tokens: int = 0
|
||||
cached_tokens: int = 0
|
||||
|
||||
|
||||
@dataclass
|
||||
class NormalizedResponse:
|
||||
"""Normalized API response from any provider."""
|
||||
|
||||
content: Optional[str]
|
||||
tool_calls: Optional[List[ToolCall]]
|
||||
finish_reason: str
|
||||
reasoning: Optional[str] = None
|
||||
usage: Optional[Usage] = None
|
||||
provider_data: Optional[Dict[str, Any]] = field(default=None, repr=False)
|
||||
|
||||
|
||||
def build_tool_call(
|
||||
id: Optional[str],
|
||||
name: str,
|
||||
arguments: Any,
|
||||
**provider_fields: Any,
|
||||
) -> ToolCall:
|
||||
"""Build a ToolCall, auto-serialising dict arguments."""
|
||||
args_str = json.dumps(arguments) if isinstance(arguments, dict) else str(arguments)
|
||||
provider_data = dict(provider_fields) if provider_fields else None
|
||||
return ToolCall(id=id, name=name, arguments=args_str, provider_data=provider_data)
|
||||
|
||||
|
||||
def map_finish_reason(reason: Optional[str], mapping: Dict[str, str]) -> str:
|
||||
"""Translate a provider-specific stop reason to the normalized set."""
|
||||
if reason is None:
|
||||
return "stop"
|
||||
return mapping.get(reason, "stop")
|
||||
@@ -1,194 +1,757 @@
|
||||
[
|
||||
{
|
||||
"id": "screenshot_github_home",
|
||||
"id": "screenshot_github_mark",
|
||||
"url": "https://github.githubassets.com/images/modules/logos_page/GitHub-Mark.png",
|
||||
"category": "screenshot",
|
||||
"expected_keywords": ["github", "logo", "mark"],
|
||||
"expected_keywords": [
|
||||
"github",
|
||||
"logo",
|
||||
"mark"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 30, "min_sentences": 1, "has_numbers": false}
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "diagram_mermaid_flow",
|
||||
"url": "https://mermaid.ink/img/pako:eNpdkE9PwzAMxb-K5VOl7gc7sAOIIDuAw9gptnRaSJLSJttQStmXs9LCH-ymBOI1ef_42U6cUSae4IkDxbAAWtB6siSZXVhjQTlgl1nigHg5fRBOzSfebopROCu_cytObSfgLSE1ANOeZWkO2IH5upZxYot8m1hqAdpD_63WRl0xdUG1jdl9kPiOb_EWk2JBtPaiKkF4eVIYgO0EtkW-RSgC4gJ6HJYRG1UNdN0HNVd0Bftjj7X8P92qPj-F8l8T3w",
|
||||
"id": "screenshot_github_social",
|
||||
"url": "https://github.githubassets.com/images/modules/site/social-cards.png",
|
||||
"category": "screenshot",
|
||||
"expected_keywords": [
|
||||
"github",
|
||||
"page",
|
||||
"web"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "screenshot_github_code_search",
|
||||
"url": "https://github.githubassets.com/images/modules/site/features-code-search.png",
|
||||
"category": "screenshot",
|
||||
"expected_keywords": [
|
||||
"search",
|
||||
"code",
|
||||
"feature"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "screenshot_terminal_capture",
|
||||
"url": "https://raw.githubusercontent.com/nicehash/nicehash-quick-start/main/images/nicehash-terminal.png",
|
||||
"category": "screenshot",
|
||||
"expected_keywords": [
|
||||
"terminal",
|
||||
"command",
|
||||
"output"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "screenshot_http_404",
|
||||
"url": "https://http.cat/404.jpg",
|
||||
"category": "screenshot",
|
||||
"expected_keywords": [
|
||||
"404",
|
||||
"error",
|
||||
"cat"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "screenshot_dummy_cli_01",
|
||||
"url": "https://dummyimage.com/1280x720/111827/f9fafb.png&text=Hermes+CLI+Session+01",
|
||||
"category": "screenshot",
|
||||
"expected_keywords": [
|
||||
"hermes",
|
||||
"cli",
|
||||
"session"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "screenshot_dummy_cli_02",
|
||||
"url": "https://dummyimage.com/1280x720/0f172a/e2e8f0.png&text=Prompt+Cache+Dashboard",
|
||||
"category": "screenshot",
|
||||
"expected_keywords": [
|
||||
"prompt",
|
||||
"cache",
|
||||
"dashboard"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "screenshot_dummy_ui_01",
|
||||
"url": "https://dummyimage.com/1280x720/1f2937/f3f4f6.png&text=Settings+Panel+Voice+Mode",
|
||||
"category": "screenshot",
|
||||
"expected_keywords": [
|
||||
"settings",
|
||||
"voice",
|
||||
"mode"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "screenshot_dummy_ui_02",
|
||||
"url": "https://dummyimage.com/1280x720/334155/f8fafc.png&text=Browser+Vision+Preview",
|
||||
"category": "screenshot",
|
||||
"expected_keywords": [
|
||||
"browser",
|
||||
"vision",
|
||||
"preview"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "screenshot_dummy_ui_03",
|
||||
"url": "https://dummyimage.com/1280x720/111827/ffffff.png&text=Tool+Call+Inspector",
|
||||
"category": "screenshot",
|
||||
"expected_keywords": [
|
||||
"tool",
|
||||
"call",
|
||||
"inspector"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "diagram_flow_a",
|
||||
"url": "https://dummyimage.com/1200x800/f8fafc/0f172a.png&text=Flowchart+API+Gateway+Queue+Worker",
|
||||
"category": "diagram",
|
||||
"expected_keywords": ["flow", "diagram", "process"],
|
||||
"expected_keywords": [
|
||||
"flowchart",
|
||||
"api",
|
||||
"worker"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 50, "min_sentences": 2, "has_numbers": false}
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "photo_random_1",
|
||||
"url": "https://picsum.photos/seed/vision1/400/300",
|
||||
"category": "photo",
|
||||
"expected_keywords": [],
|
||||
"id": "diagram_flow_b",
|
||||
"url": "https://dummyimage.com/1200x800/f1f5f9/0f172a.png&text=Architecture+Diagram+Database+Cache+Client",
|
||||
"category": "diagram",
|
||||
"expected_keywords": [
|
||||
"architecture",
|
||||
"diagram",
|
||||
"cache"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 30, "min_sentences": 1, "has_numbers": false}
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "photo_random_2",
|
||||
"url": "https://picsum.photos/seed/vision2/400/300",
|
||||
"category": "photo",
|
||||
"expected_keywords": [],
|
||||
"id": "diagram_uml_a",
|
||||
"url": "https://dummyimage.com/1200x800/e2e8f0/0f172a.png&text=Class+Diagram+User+Session+Message",
|
||||
"category": "diagram",
|
||||
"expected_keywords": [
|
||||
"class",
|
||||
"diagram",
|
||||
"session"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 30, "min_sentences": 1, "has_numbers": false}
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "chart_simple_bar",
|
||||
"url": "https://quickchart.io/chart?c={type:'bar',data:{labels:['Q1','Q2','Q3','Q4'],datasets:[{label:'Revenue',data:[100,150,200,250]}]}}",
|
||||
"category": "chart",
|
||||
"expected_keywords": ["bar", "chart", "revenue"],
|
||||
"id": "diagram_uml_b",
|
||||
"url": "https://dummyimage.com/1200x800/cbd5e1/0f172a.png&text=Sequence+Diagram+Request+Response",
|
||||
"category": "diagram",
|
||||
"expected_keywords": [
|
||||
"sequence",
|
||||
"diagram",
|
||||
"response"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 50, "min_sentences": 2, "has_numbers": true}
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "chart_pie",
|
||||
"url": "https://quickchart.io/chart?c={type:'pie',data:{labels:['A','B','C'],datasets:[{data:[30,50,20]}]}}",
|
||||
"category": "chart",
|
||||
"expected_keywords": ["pie", "chart", "percentage"],
|
||||
"id": "diagram_network_a",
|
||||
"url": "https://dummyimage.com/1200x800/ffffff/111827.png&text=Network+Nodes+Edges+Router",
|
||||
"category": "diagram",
|
||||
"expected_keywords": [
|
||||
"network",
|
||||
"node",
|
||||
"router"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 50, "min_sentences": 2, "has_numbers": true}
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "diagram_network_b",
|
||||
"url": "https://dummyimage.com/1200x800/ffffff/1e293b.png&text=Service+Mesh+Proxy+Auth",
|
||||
"category": "diagram",
|
||||
"expected_keywords": [
|
||||
"service",
|
||||
"mesh",
|
||||
"auth"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "diagram_state_machine",
|
||||
"url": "https://dummyimage.com/1200x800/f8fafc/334155.png&text=State+Machine+Idle+Run+Stop",
|
||||
"category": "diagram",
|
||||
"expected_keywords": [
|
||||
"state",
|
||||
"machine",
|
||||
"idle"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "diagram_mind_map",
|
||||
"url": "https://dummyimage.com/1200x800/fefce8/1f2937.png&text=Mind+Map+Memory+Recall+Tools",
|
||||
"category": "diagram",
|
||||
"expected_keywords": [
|
||||
"mind",
|
||||
"memory",
|
||||
"tools"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "diagram_pipeline",
|
||||
"url": "https://dummyimage.com/1200x800/ecfeff/155e75.png&text=Pipeline+Ingest+Rank+Summarize",
|
||||
"category": "diagram",
|
||||
"expected_keywords": [
|
||||
"pipeline",
|
||||
"ingest",
|
||||
"summarize"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "diagram_org_chart",
|
||||
"url": "https://mermaid.ink/img/pako:eNpdkE9PwzAMxb-K5VOl7gc7sAOIIDuAw9gptnRaSJLSJttQStmXs9LCH-ymBOI1ef_42U6cUSae4IkDxbAAWtB6iuyIWyrLgXLALrPEAfFy-iCcmk-83RSjcFZ-51ac2k7AW0JqAKY9y9IcsAPzdS3jxBb5NrHUAraH_lutjbpi6oJqG7P7IPEd3-ItJsWCaO1FVYLw8qQwANsJbIt8i1AExAX0OCwjNqoa6LoPaq7oCvbHHmv5f7pVfX4K5b8mvg",
|
||||
"url": "https://dummyimage.com/1200x800/fdf2f8/831843.png&text=Org+Chart+Lead+Review+Ops",
|
||||
"category": "diagram",
|
||||
"expected_keywords": ["organization", "hierarchy", "chart"],
|
||||
"expected_keywords": [
|
||||
"org",
|
||||
"chart",
|
||||
"review"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 50, "min_sentences": 2, "has_numbers": false}
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "screenshot_terminal",
|
||||
"url": "https://raw.githubusercontent.com/nicehash/nicehash-quick-start/main/images/nicehash-terminal.png",
|
||||
"category": "screenshot",
|
||||
"expected_keywords": ["terminal", "command", "output"],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 30, "min_sentences": 1, "has_numbers": false}
|
||||
},
|
||||
{
|
||||
"id": "photo_random_3",
|
||||
"url": "https://picsum.photos/seed/vision3/400/300",
|
||||
"id": "photo_random_01",
|
||||
"url": "https://picsum.photos/seed/vision-bench-1/640/480",
|
||||
"category": "photo",
|
||||
"expected_keywords": [],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 30, "min_sentences": 1, "has_numbers": false}
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "chart_line",
|
||||
"id": "photo_random_02",
|
||||
"url": "https://picsum.photos/seed/vision-bench-2/640/480",
|
||||
"category": "photo",
|
||||
"expected_keywords": [],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "photo_random_03",
|
||||
"url": "https://picsum.photos/seed/vision-bench-3/640/480",
|
||||
"category": "photo",
|
||||
"expected_keywords": [],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "photo_random_04",
|
||||
"url": "https://picsum.photos/seed/vision-bench-4/640/480",
|
||||
"category": "photo",
|
||||
"expected_keywords": [],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "photo_random_05",
|
||||
"url": "https://picsum.photos/seed/vision-bench-5/640/480",
|
||||
"category": "photo",
|
||||
"expected_keywords": [],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "photo_random_06",
|
||||
"url": "https://picsum.photos/seed/vision-bench-6/640/480",
|
||||
"category": "photo",
|
||||
"expected_keywords": [],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "photo_random_07",
|
||||
"url": "https://picsum.photos/seed/vision-bench-7/640/480",
|
||||
"category": "photo",
|
||||
"expected_keywords": [],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "photo_random_08",
|
||||
"url": "https://picsum.photos/seed/vision-bench-8/640/480",
|
||||
"category": "photo",
|
||||
"expected_keywords": [],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "photo_random_09",
|
||||
"url": "https://picsum.photos/seed/vision-bench-9/640/480",
|
||||
"category": "photo",
|
||||
"expected_keywords": [],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "photo_random_10",
|
||||
"url": "https://picsum.photos/seed/vision-bench-10/640/480",
|
||||
"category": "photo",
|
||||
"expected_keywords": [],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 30,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "chart_bar_quarterly",
|
||||
"url": "https://quickchart.io/chart?c={type:'bar',data:{labels:['Q1','Q2','Q3','Q4'],datasets:[{label:'Revenue',data:[100,150,200,250]}]}}",
|
||||
"category": "chart",
|
||||
"expected_keywords": [
|
||||
"bar",
|
||||
"chart",
|
||||
"revenue"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "chart_pie_market",
|
||||
"url": "https://quickchart.io/chart?c={type:'pie',data:{labels:['A','B','C'],datasets:[{data:[30,50,20]}]}}",
|
||||
"category": "chart",
|
||||
"expected_keywords": [
|
||||
"pie",
|
||||
"chart",
|
||||
"percentage"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "chart_line_temp",
|
||||
"url": "https://quickchart.io/chart?c={type:'line',data:{labels:['Jan','Feb','Mar','Apr'],datasets:[{label:'Temperature',data:[5,8,12,18]}]}}",
|
||||
"category": "chart",
|
||||
"expected_keywords": ["line", "chart", "temperature"],
|
||||
"expected_keywords": [
|
||||
"line",
|
||||
"chart",
|
||||
"temperature"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 50, "min_sentences": 2, "has_numbers": true}
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "diagram_sequence",
|
||||
"url": "https://mermaid.ink/img/pako:eNpdkE9PwzAMxb-K5VOl7gc7sAOIIDuAw9gptnRaSJLSJttQStmXs9LCH-ymBOI1ef_42U6cUSae4IkDxbAAWtB6iuyIWyrLgXLALrPEAfFy-iCcmk-83RSjcFZ-51ac2k7AW0JqAKY9y9IcsAPzdS3jxBb5NrHUAraH_lutjbpi6oJqG7P7IPEd3-ItJsWCaO1FVYLw8qQwANsJbIt8i1AExAX0OCwjNqoa6LoPaq7oCvbHHmv5f7pVfX4K5b8mvg",
|
||||
"category": "diagram",
|
||||
"expected_keywords": ["sequence", "interaction", "message"],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 50, "min_sentences": 2, "has_numbers": false}
|
||||
},
|
||||
{
|
||||
"id": "photo_random_4",
|
||||
"url": "https://picsum.photos/seed/vision4/400/300",
|
||||
"category": "photo",
|
||||
"expected_keywords": [],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 30, "min_sentences": 1, "has_numbers": false}
|
||||
},
|
||||
{
|
||||
"id": "screenshot_webpage",
|
||||
"url": "https://github.githubassets.com/images/modules/site/social-cards.png",
|
||||
"category": "screenshot",
|
||||
"expected_keywords": ["github", "page", "web"],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 30, "min_sentences": 1, "has_numbers": false}
|
||||
},
|
||||
{
|
||||
"id": "chart_radar",
|
||||
"id": "chart_radar_skill",
|
||||
"url": "https://quickchart.io/chart?c={type:'radar',data:{labels:['Speed','Power','Defense','Magic'],datasets:[{label:'Hero',data:[80,60,70,90]}]}}",
|
||||
"category": "chart",
|
||||
"expected_keywords": ["radar", "chart", "skill"],
|
||||
"expected_keywords": [
|
||||
"radar",
|
||||
"chart",
|
||||
"skill"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 50, "min_sentences": 2, "has_numbers": true}
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "photo_random_5",
|
||||
"url": "https://picsum.photos/seed/vision5/400/300",
|
||||
"category": "photo",
|
||||
"expected_keywords": [],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 30, "min_sentences": 1, "has_numbers": false}
|
||||
},
|
||||
{
|
||||
"id": "diagram_class",
|
||||
"url": "https://mermaid.ink/img/pako:eNpdkE9PwzAMxb-K5VOl7gc7sAOIIDuAw9gptnRaSJLSJttQStmXs9LCH-ymBOI1ef_42U6cUSae4IkDxbAAWtB6iuyIWyrLgXLALrPEAfFy-iCcmk-83RSjcFZ-51ac2k7AW0JqAKY9y9IcsAPzdS3jxBb5NrHUAraH_lutjbpi6oJqG7P7IPEd3-ItJsWCaO1FVYLw8qQwANsJbIt8i1AExAX0OCwjNqoa6LoPaq7oCvbHHmv5f7pVfX4K5b8mvg",
|
||||
"category": "diagram",
|
||||
"expected_keywords": ["class", "object", "attribute"],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 50, "min_sentences": 2, "has_numbers": false}
|
||||
},
|
||||
{
|
||||
"id": "chart_doughnut",
|
||||
"url": "https://quickchart.io/chart?c={type:'doughnut',data:{labels:['Desktop','Mobile','Tablet'],datasets:[{data:[60,30,10]}]}}",
|
||||
"category": "chart",
|
||||
"expected_keywords": ["doughnut", "chart", "device"],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 50, "min_sentences": 2, "has_numbers": true}
|
||||
},
|
||||
{
|
||||
"id": "photo_random_6",
|
||||
"url": "https://picsum.photos/seed/vision6/400/300",
|
||||
"category": "photo",
|
||||
"expected_keywords": [],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 30, "min_sentences": 1, "has_numbers": false}
|
||||
},
|
||||
{
|
||||
"id": "screenshot_error",
|
||||
"url": "https://http.cat/404.jpg",
|
||||
"category": "screenshot",
|
||||
"expected_keywords": ["404", "error", "cat"],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 30, "min_sentences": 1, "has_numbers": true}
|
||||
},
|
||||
{
|
||||
"id": "diagram_network",
|
||||
"url": "https://mermaid.ink/img/pako:eNpdkE9PwzAMxb-K5VOl7gc7sAOIIDuAw9gptnRaSJLSJttQStmXs9LCH-ymBOI1ef_42U6cUSae4IkDxbAAWtB6iuyIWyrLgXLALrPEAfFy-iCcmk-83RSjcFZ-51ac2k7AW0JqAKY9y9IcsAPzdS3jxBb5NrHUAraH_lutjbpi6oJqG7P7IPEd3-ItJsWCaO1FVYLw8qQwANsJbIt8i1AExAX0OCwjNqoa6LoPaq7oCvbHHmv5f7pVfX4K5b8mvg",
|
||||
"category": "diagram",
|
||||
"expected_keywords": ["network", "node", "connection"],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 50, "min_sentences": 2, "has_numbers": false}
|
||||
},
|
||||
{
|
||||
"id": "photo_random_7",
|
||||
"url": "https://picsum.photos/seed/vision7/400/300",
|
||||
"category": "photo",
|
||||
"expected_keywords": [],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 30, "min_sentences": 1, "has_numbers": false}
|
||||
},
|
||||
{
|
||||
"id": "chart_stacked_bar",
|
||||
"id": "chart_stacked_cloud",
|
||||
"url": "https://quickchart.io/chart?c={type:'bar',data:{labels:['2022','2023','2024'],datasets:[{label:'Cloud',data:[100,150,200]},{label:'On-prem',data:[200,180,160]}]},options:{scales:{x:{stacked:true},y:{stacked:true}}}}",
|
||||
"category": "chart",
|
||||
"expected_keywords": ["stacked", "bar", "chart"],
|
||||
"expected_keywords": [
|
||||
"stacked",
|
||||
"bar",
|
||||
"chart"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 50, "min_sentences": 2, "has_numbers": true}
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "screenshot_dashboard",
|
||||
"url": "https://github.githubassets.com/images/modules/site/features-code-search.png",
|
||||
"category": "screenshot",
|
||||
"expected_keywords": ["search", "code", "feature"],
|
||||
"id": "chart_area_growth",
|
||||
"url": "https://quickchart.io/chart?c={type:'line',data:{labels:['W1','W2','W3','W4'],datasets:[{label:'Growth',data:[10,15,18,24],fill:true}]}}",
|
||||
"category": "chart",
|
||||
"expected_keywords": [
|
||||
"line",
|
||||
"growth",
|
||||
"chart"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 30, "min_sentences": 1, "has_numbers": false}
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "photo_random_8",
|
||||
"url": "https://picsum.photos/seed/vision8/400/300",
|
||||
"category": "photo",
|
||||
"expected_keywords": [],
|
||||
"id": "chart_scatter_eval",
|
||||
"url": "https://quickchart.io/chart?c={type:'scatter',data:{datasets:[{label:'Runs',data:[{x:1,y:70},{x:2,y:75},{x:3,y:82}]}]}}",
|
||||
"category": "chart",
|
||||
"expected_keywords": [
|
||||
"scatter",
|
||||
"chart",
|
||||
"runs"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 30, "min_sentences": 1, "has_numbers": false}
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "chart_horizontal_bar",
|
||||
"url": "https://quickchart.io/chart?c={type:'bar',data:{labels:['UI','OCR','Docs'],datasets:[{label:'Score',data:[88,76,91]}]},options:{indexAxis:'y'}}",
|
||||
"category": "chart",
|
||||
"expected_keywords": [
|
||||
"bar",
|
||||
"score",
|
||||
"ocr"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "chart_bubble_usage",
|
||||
"url": "https://quickchart.io/chart?c={type:'bubble',data:{datasets:[{label:'Latency',data:[{x:1,y:120,r:8},{x:2,y:95,r:6},{x:3,y:180,r:10}]}]}}",
|
||||
"category": "chart",
|
||||
"expected_keywords": [
|
||||
"bubble",
|
||||
"latency",
|
||||
"chart"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "chart_doughnut_devices",
|
||||
"url": "https://quickchart.io/chart?c={type:'doughnut',data:{labels:['Desktop','Mobile','Tablet'],datasets:[{data:[60,30,10]}]}}",
|
||||
"category": "chart",
|
||||
"expected_keywords": [
|
||||
"doughnut",
|
||||
"chart",
|
||||
"device"
|
||||
],
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {
|
||||
"min_length": 50,
|
||||
"min_sentences": 2,
|
||||
"has_numbers": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "ocr_text_01",
|
||||
"url": "https://dummyimage.com/1200x320/ffffff/000000.png&text=Hermes+OCR+Alpha+01",
|
||||
"category": "ocr",
|
||||
"expected_keywords": [
|
||||
"hermes",
|
||||
"ocr"
|
||||
],
|
||||
"ground_truth_ocr": "Hermes OCR Alpha 01",
|
||||
"expected_structure": {
|
||||
"min_length": 10,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "ocr_text_02",
|
||||
"url": "https://dummyimage.com/1200x320/ffffff/000000.png&text=Prompt+Cache+Hit+87%",
|
||||
"category": "ocr",
|
||||
"expected_keywords": [
|
||||
"prompt",
|
||||
"cache"
|
||||
],
|
||||
"ground_truth_ocr": "Prompt Cache Hit 87%",
|
||||
"expected_structure": {
|
||||
"min_length": 10,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "ocr_text_03",
|
||||
"url": "https://dummyimage.com/1200x320/ffffff/000000.png&text=Session+42+Ready",
|
||||
"category": "ocr",
|
||||
"expected_keywords": [
|
||||
"session",
|
||||
"42"
|
||||
],
|
||||
"ground_truth_ocr": "Session 42 Ready",
|
||||
"expected_structure": {
|
||||
"min_length": 10,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "ocr_text_04",
|
||||
"url": "https://dummyimage.com/1200x320/ffffff/000000.png&text=Latency+118+ms",
|
||||
"category": "ocr",
|
||||
"expected_keywords": [
|
||||
"latency",
|
||||
"118"
|
||||
],
|
||||
"ground_truth_ocr": "Latency 118 ms",
|
||||
"expected_structure": {
|
||||
"min_length": 10,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "ocr_text_05",
|
||||
"url": "https://dummyimage.com/1200x320/ffffff/000000.png&text=Voice+Mode+Enabled",
|
||||
"category": "ocr",
|
||||
"expected_keywords": [
|
||||
"voice",
|
||||
"mode"
|
||||
],
|
||||
"ground_truth_ocr": "Voice Mode Enabled",
|
||||
"expected_structure": {
|
||||
"min_length": 10,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "document_text_01",
|
||||
"url": "https://dummyimage.com/1400x900/f8fafc/0f172a.png&text=Invoice+1001+Total+42+Due+2026-04-22",
|
||||
"category": "document",
|
||||
"expected_keywords": [
|
||||
"invoice",
|
||||
"1001",
|
||||
"total"
|
||||
],
|
||||
"ground_truth_ocr": "Invoice 1001 Total 42 Due 2026-04-22",
|
||||
"expected_structure": {
|
||||
"min_length": 20,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "document_text_02",
|
||||
"url": "https://dummyimage.com/1400x900/f8fafc/0f172a.png&text=Form+A+Name+Alice+Status+Approved",
|
||||
"category": "document",
|
||||
"expected_keywords": [
|
||||
"form",
|
||||
"a",
|
||||
"name"
|
||||
],
|
||||
"ground_truth_ocr": "Form A Name Alice Status Approved",
|
||||
"expected_structure": {
|
||||
"min_length": 20,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "document_text_03",
|
||||
"url": "https://dummyimage.com/1400x900/f8fafc/0f172a.png&text=Report+Memory+Recall+Score+91+Percent",
|
||||
"category": "document",
|
||||
"expected_keywords": [
|
||||
"report",
|
||||
"memory",
|
||||
"recall"
|
||||
],
|
||||
"ground_truth_ocr": "Report Memory Recall Score 91 Percent",
|
||||
"expected_structure": {
|
||||
"min_length": 20,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "document_text_04",
|
||||
"url": "https://dummyimage.com/1400x900/f8fafc/0f172a.png&text=Checklist+Crisis+Escalation+Call+988+Now",
|
||||
"category": "document",
|
||||
"expected_keywords": [
|
||||
"checklist",
|
||||
"crisis",
|
||||
"escalation"
|
||||
],
|
||||
"ground_truth_ocr": "Checklist Crisis Escalation Call 988 Now",
|
||||
"expected_structure": {
|
||||
"min_length": 20,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "document_text_05",
|
||||
"url": "https://dummyimage.com/1400x900/f8fafc/0f172a.png&text=Meeting+Notes+Vision+Benchmark+Run+Pending",
|
||||
"category": "document",
|
||||
"expected_keywords": [
|
||||
"meeting",
|
||||
"notes",
|
||||
"vision"
|
||||
],
|
||||
"ground_truth_ocr": "Meeting Notes Vision Benchmark Run Pending",
|
||||
"expected_structure": {
|
||||
"min_length": 20,
|
||||
"min_sentences": 1,
|
||||
"has_numbers": false
|
||||
}
|
||||
}
|
||||
]
|
||||
]
|
||||
@@ -22,10 +22,12 @@ import argparse
|
||||
import asyncio
|
||||
import base64
|
||||
import json
|
||||
import mimetypes
|
||||
import os
|
||||
import statistics
|
||||
import sys
|
||||
import time
|
||||
import urllib.request
|
||||
from datetime import datetime, timezone
|
||||
from pathlib import Path
|
||||
from typing import Any, Dict, List, Optional
|
||||
@@ -41,12 +43,16 @@ MODELS = {
|
||||
"model_id": "google/gemma-4-27b-it",
|
||||
"display_name": "Gemma 4 27B",
|
||||
"provider": "nous",
|
||||
"fallback_provider": "ollama",
|
||||
"fallback_model_id": "gemma4:latest",
|
||||
"description": "Google's multimodal Gemma 4 model",
|
||||
},
|
||||
"gemini3_flash": {
|
||||
"model_id": "google/gemini-3-flash-preview",
|
||||
"display_name": "Gemini 3 Flash Preview",
|
||||
"provider": "openrouter",
|
||||
"fallback_provider": "gemini",
|
||||
"fallback_model_id": "gemini-2.5-flash",
|
||||
"description": "Current default vision model",
|
||||
},
|
||||
}
|
||||
@@ -84,91 +90,150 @@ async def analyze_with_model(
|
||||
"""
|
||||
import httpx
|
||||
|
||||
def _load_image_bytes_cached() -> tuple[bytes, str]:
|
||||
nonlocal _image_bytes, _mime_type
|
||||
if _image_bytes is not None:
|
||||
return _image_bytes, _mime_type
|
||||
if image_url.startswith(("http://", "https://")):
|
||||
with urllib.request.urlopen(image_url, timeout=30) as resp:
|
||||
_image_bytes = resp.read()
|
||||
_mime_type = resp.headers.get_content_type() or mimetypes.guess_type(image_url)[0] or "image/png"
|
||||
else:
|
||||
path = Path(image_url).expanduser()
|
||||
_image_bytes = path.read_bytes()
|
||||
_mime_type = mimetypes.guess_type(str(path))[0] or "image/png"
|
||||
return _image_bytes, _mime_type
|
||||
|
||||
def _data_url() -> str:
|
||||
image_bytes, mime_type = _load_image_bytes_cached()
|
||||
return f"data:{mime_type};base64,{base64.b64encode(image_bytes).decode()}"
|
||||
|
||||
def _provider_key(provider: str) -> str:
|
||||
if provider == "openrouter":
|
||||
return os.getenv("OPENROUTER_API_KEY", "")
|
||||
if provider == "nous":
|
||||
return os.getenv("NOUS_API_KEY", "") or os.getenv("NOUS_INFERENCE_API_KEY", "")
|
||||
if provider == "gemini":
|
||||
return os.getenv("GEMINI_API_KEY", "") or os.getenv("GOOGLE_API_KEY", "")
|
||||
return os.getenv(f"{provider.upper()}_API_KEY", "")
|
||||
|
||||
provider = model_config["provider"]
|
||||
model_id = model_config["model_id"]
|
||||
candidates = [(provider, model_id)]
|
||||
if model_config.get("fallback_provider") and model_config.get("fallback_model_id"):
|
||||
candidates.append((model_config["fallback_provider"], model_config["fallback_model_id"]))
|
||||
|
||||
# Prepare messages
|
||||
messages = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{"type": "text", "text": prompt},
|
||||
{"type": "image_url", "image_url": {"url": image_url}},
|
||||
],
|
||||
}
|
||||
]
|
||||
_image_bytes: Optional[bytes] = None
|
||||
_mime_type = "image/png"
|
||||
failures = []
|
||||
|
||||
# Route to provider
|
||||
if provider == "openrouter":
|
||||
api_url = "https://openrouter.ai/api/v1/chat/completions"
|
||||
api_key = os.getenv("OPENROUTER_API_KEY", "")
|
||||
elif provider == "nous":
|
||||
api_url = "https://inference.nousresearch.com/v1/chat/completions"
|
||||
api_key = os.getenv("NOUS_API_KEY", "") or os.getenv("NOUS_INFERENCE_API_KEY", "")
|
||||
else:
|
||||
api_url = os.getenv(f"{provider.upper()}_API_URL", "")
|
||||
api_key = os.getenv(f"{provider.upper()}_API_KEY", "")
|
||||
for candidate_provider, candidate_model in candidates:
|
||||
api_key = _provider_key(candidate_provider)
|
||||
start = time.perf_counter()
|
||||
try:
|
||||
if candidate_provider in {"openrouter", "nous"}:
|
||||
api_url = (
|
||||
"https://openrouter.ai/api/v1/chat/completions"
|
||||
if candidate_provider == "openrouter"
|
||||
else "https://inference.nousresearch.com/v1/chat/completions"
|
||||
)
|
||||
if not api_key:
|
||||
raise RuntimeError(f"No API key for provider {candidate_provider}")
|
||||
payload = {
|
||||
"model": candidate_model,
|
||||
"messages": [{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{"type": "text", "text": prompt},
|
||||
{"type": "image_url", "image_url": {"url": _data_url() if not image_url.startswith(("http://", "https://")) else image_url}},
|
||||
],
|
||||
}],
|
||||
"max_tokens": 2000,
|
||||
"temperature": 0.1,
|
||||
}
|
||||
headers = {
|
||||
"Authorization": f"Bearer {api_key}",
|
||||
"Content-Type": "application/json",
|
||||
}
|
||||
async with httpx.AsyncClient(timeout=timeout) as client:
|
||||
resp = await client.post(api_url, json=payload, headers=headers)
|
||||
resp.raise_for_status()
|
||||
data = resp.json()
|
||||
analysis = data.get("choices", [{}])[0].get("message", {}).get("content", "")
|
||||
usage = data.get("usage", {})
|
||||
tokens = {
|
||||
"prompt_tokens": usage.get("prompt_tokens", 0),
|
||||
"completion_tokens": usage.get("completion_tokens", 0),
|
||||
"total_tokens": usage.get("total_tokens", 0),
|
||||
}
|
||||
elif candidate_provider == "gemini":
|
||||
if not api_key:
|
||||
raise RuntimeError("No API key for provider gemini")
|
||||
image_bytes, mime_type = _load_image_bytes_cached()
|
||||
api_url = f"https://generativelanguage.googleapis.com/v1beta/models/{candidate_model}:generateContent?key={api_key}"
|
||||
payload = {
|
||||
"contents": [{"parts": [
|
||||
{"text": prompt},
|
||||
{"inline_data": {"mime_type": mime_type, "data": base64.b64encode(image_bytes).decode()}},
|
||||
]}],
|
||||
"generationConfig": {"temperature": 0.1, "maxOutputTokens": 2000},
|
||||
}
|
||||
async with httpx.AsyncClient(timeout=timeout) as client:
|
||||
resp = await client.post(api_url, json=payload)
|
||||
resp.raise_for_status()
|
||||
data = resp.json()
|
||||
parts = data.get("candidates", [{}])[0].get("content", {}).get("parts", [])
|
||||
analysis = "\n".join(part.get("text", "") for part in parts if isinstance(part, dict) and part.get("text"))
|
||||
usage = data.get("usageMetadata", {})
|
||||
tokens = {
|
||||
"prompt_tokens": usage.get("promptTokenCount", 0),
|
||||
"completion_tokens": usage.get("candidatesTokenCount", 0),
|
||||
"total_tokens": usage.get("totalTokenCount", 0),
|
||||
}
|
||||
elif candidate_provider == "ollama":
|
||||
image_bytes, _ = _load_image_bytes_cached()
|
||||
payload = {
|
||||
"model": candidate_model,
|
||||
"stream": False,
|
||||
"messages": [{"role": "user", "content": prompt, "images": [base64.b64encode(image_bytes).decode()]}],
|
||||
"options": {"temperature": 0.1},
|
||||
}
|
||||
async with httpx.AsyncClient(timeout=timeout) as client:
|
||||
resp = await client.post("http://localhost:11434/api/chat", json=payload)
|
||||
resp.raise_for_status()
|
||||
data = resp.json()
|
||||
analysis = data.get("message", {}).get("content", "")
|
||||
tokens = {
|
||||
"prompt_tokens": data.get("prompt_eval_count", 0),
|
||||
"completion_tokens": data.get("eval_count", 0),
|
||||
"total_tokens": (data.get("prompt_eval_count", 0) or 0) + (data.get("eval_count", 0) or 0),
|
||||
}
|
||||
else:
|
||||
raise RuntimeError(f"Unsupported provider {candidate_provider}")
|
||||
|
||||
if not api_key:
|
||||
return {
|
||||
"analysis": "",
|
||||
"latency_ms": 0,
|
||||
"tokens": {},
|
||||
"success": False,
|
||||
"error": f"No API key for provider {provider}",
|
||||
}
|
||||
latency_ms = (time.perf_counter() - start) * 1000
|
||||
return {
|
||||
"analysis": analysis,
|
||||
"latency_ms": round(latency_ms, 1),
|
||||
"tokens": tokens,
|
||||
"success": True,
|
||||
"error": "",
|
||||
"provider_used": candidate_provider,
|
||||
"model_used": candidate_model,
|
||||
}
|
||||
except Exception as e:
|
||||
failures.append(f"{candidate_provider}:{candidate_model} => {e}")
|
||||
|
||||
headers = {
|
||||
"Authorization": f"Bearer {api_key}",
|
||||
"Content-Type": "application/json",
|
||||
return {
|
||||
"analysis": "",
|
||||
"latency_ms": 0,
|
||||
"tokens": {},
|
||||
"success": False,
|
||||
"error": " | ".join(failures) if failures else "No runs",
|
||||
"provider_used": candidates[-1][0] if candidates else provider,
|
||||
"model_used": candidates[-1][1] if candidates else model_id,
|
||||
}
|
||||
|
||||
payload = {
|
||||
"model": model_id,
|
||||
"messages": messages,
|
||||
"max_tokens": 2000,
|
||||
"temperature": 0.1,
|
||||
}
|
||||
|
||||
start = time.perf_counter()
|
||||
try:
|
||||
async with httpx.AsyncClient(timeout=timeout) as client:
|
||||
resp = await client.post(api_url, json=payload, headers=headers)
|
||||
resp.raise_for_status()
|
||||
data = resp.json()
|
||||
|
||||
latency_ms = (time.perf_counter() - start) * 1000
|
||||
|
||||
analysis = ""
|
||||
choices = data.get("choices", [])
|
||||
if choices:
|
||||
msg = choices[0].get("message", {})
|
||||
analysis = msg.get("content", "")
|
||||
|
||||
usage = data.get("usage", {})
|
||||
tokens = {
|
||||
"prompt_tokens": usage.get("prompt_tokens", 0),
|
||||
"completion_tokens": usage.get("completion_tokens", 0),
|
||||
"total_tokens": usage.get("total_tokens", 0),
|
||||
}
|
||||
|
||||
return {
|
||||
"analysis": analysis,
|
||||
"latency_ms": round(latency_ms, 1),
|
||||
"tokens": tokens,
|
||||
"success": True,
|
||||
"error": "",
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
return {
|
||||
"analysis": "",
|
||||
"latency_ms": round((time.perf_counter() - start) * 1000, 1),
|
||||
"tokens": {},
|
||||
"success": False,
|
||||
"error": str(e),
|
||||
}
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Evaluation metrics
|
||||
@@ -398,7 +463,13 @@ def aggregate_results(results: List[dict], models: dict) -> dict:
|
||||
failed = [r[model_name] for r in results if not r[model_name]["success"]]
|
||||
|
||||
if not model_results:
|
||||
summary[model_name] = {"success_rate": 0, "error": "All runs failed"}
|
||||
summary[model_name] = {
|
||||
"success_rate": 0,
|
||||
"error": "All runs failed",
|
||||
"total_runs": 0,
|
||||
"total_failures": len(failed),
|
||||
"failure_examples": sorted({f.get("error", "unknown failure") for f in failed})[:3],
|
||||
}
|
||||
continue
|
||||
|
||||
latencies = [r["avg_latency_ms"] for r in model_results]
|
||||
@@ -410,6 +481,7 @@ def aggregate_results(results: List[dict], models: dict) -> dict:
|
||||
"success_rate": round(len(model_results) / (len(model_results) + len(failed)), 4),
|
||||
"total_runs": len(model_results),
|
||||
"total_failures": len(failed),
|
||||
"failure_examples": sorted({f.get("error", "unknown failure") for f in failed})[:3],
|
||||
"latency": {
|
||||
"mean_ms": round(statistics.mean(latencies), 1),
|
||||
"median_ms": round(statistics.median(latencies), 1),
|
||||
@@ -495,6 +567,23 @@ def to_markdown(report: dict) -> str:
|
||||
f"| {mname} | {tok['mean_total']:.0f} | {tok['total_used']} |"
|
||||
)
|
||||
|
||||
lines += ["", "## Failure Modes", ""]
|
||||
had_failures = False
|
||||
for mkey, mname in config["models"].items():
|
||||
model_summary = summary.get(mkey, {})
|
||||
failure_examples = model_summary.get("failure_examples", [])
|
||||
if not failure_examples and not model_summary.get("error"):
|
||||
continue
|
||||
had_failures = True
|
||||
lines.append(f"### {mname}")
|
||||
if model_summary.get("error"):
|
||||
lines.append(f"- Summary: {model_summary['error']}")
|
||||
for err in failure_examples:
|
||||
lines.append(f"- {err}")
|
||||
lines.append("")
|
||||
if not had_failures:
|
||||
lines.append("- No provider/runtime failures recorded.")
|
||||
|
||||
# Verdict
|
||||
lines += ["", "## Verdict", ""]
|
||||
|
||||
@@ -516,8 +605,12 @@ def to_markdown(report: dict) -> str:
|
||||
|
||||
if best_model:
|
||||
lines.append(f"**Best overall: {best_model}** (composite score: {best_score:.1%})")
|
||||
lines.append("")
|
||||
lines.append("Recommendation: keep the best-performing Gemma/Gemini lane from this run and only switch if repeated runs disagree.")
|
||||
else:
|
||||
lines.append("No clear winner — insufficient data.")
|
||||
lines.append("Benchmark blocked or insufficient data for a trustworthy winner.")
|
||||
lines.append("")
|
||||
lines.append("Recommendation: repair provider/runtime availability, rerun the benchmark, and keep the current implementation unchanged until comparative results exist.")
|
||||
|
||||
return "\n".join(lines)
|
||||
|
||||
@@ -528,44 +621,124 @@ def to_markdown(report: dict) -> str:
|
||||
|
||||
|
||||
def generate_sample_dataset() -> List[dict]:
|
||||
"""Generate a sample test dataset with diverse public images.
|
||||
"""Generate a larger benchmark dataset aligned with issue #817.
|
||||
|
||||
Returns list of test image definitions.
|
||||
Returns 50+ images across screenshots, diagrams, photos, OCR, charts,
|
||||
and document-like images so the harness matches the issue contract.
|
||||
"""
|
||||
return [
|
||||
# Screenshots
|
||||
{
|
||||
"id": "screenshot_github",
|
||||
"url": "https://github.githubassets.com/images/modules/logos_page/GitHub-Mark.png",
|
||||
dataset: List[dict] = []
|
||||
|
||||
screenshots = [
|
||||
("github_mark", "https://github.githubassets.com/images/modules/logos_page/GitHub-Mark.png", ["github", "logo", "mark"]),
|
||||
("github_social", "https://github.githubassets.com/images/modules/site/social-cards.png", ["github", "page", "web"]),
|
||||
("github_code_search", "https://github.githubassets.com/images/modules/site/features-code-search.png", ["search", "code", "feature"]),
|
||||
("terminal_capture", "https://raw.githubusercontent.com/nicehash/nicehash-quick-start/main/images/nicehash-terminal.png", ["terminal", "command", "output"]),
|
||||
("http_404", "https://http.cat/404.jpg", ["404", "error", "cat"]),
|
||||
("dummy_cli_01", "https://dummyimage.com/1280x720/111827/f9fafb.png&text=Hermes+CLI+Session+01", ["hermes", "cli", "session"]),
|
||||
("dummy_cli_02", "https://dummyimage.com/1280x720/0f172a/e2e8f0.png&text=Prompt+Cache+Dashboard", ["prompt", "cache", "dashboard"]),
|
||||
("dummy_ui_01", "https://dummyimage.com/1280x720/1f2937/f3f4f6.png&text=Settings+Panel+Voice+Mode", ["settings", "voice", "mode"]),
|
||||
("dummy_ui_02", "https://dummyimage.com/1280x720/334155/f8fafc.png&text=Browser+Vision+Preview", ["browser", "vision", "preview"]),
|
||||
("dummy_ui_03", "https://dummyimage.com/1280x720/111827/ffffff.png&text=Tool+Call+Inspector", ["tool", "call", "inspector"]),
|
||||
]
|
||||
for ident, url, keywords in screenshots:
|
||||
dataset.append({
|
||||
"id": f"screenshot_{ident}",
|
||||
"url": url,
|
||||
"category": "screenshot",
|
||||
"expected_keywords": ["github", "logo", "octocat"],
|
||||
"expected_structure": {"min_length": 50, "min_sentences": 2},
|
||||
},
|
||||
# Diagrams
|
||||
{
|
||||
"id": "diagram_architecture",
|
||||
"url": "https://mermaid.ink/img/pako:eNp9kMtOwzAQRX_F8hKpJbhJFVJBi1QJiMWCG8eZNsGJLdlOiqIid5RdufiHnZRA7GbuzJwZe4ZGH2SCBPYUwgxoQKvJnCR2YY0F5YBdJJkD4uX0oXB6PnF3U4zCWcWdW3FqOwGvCKkBmHKSTB2gJeRrLTeJLfJdJKkBGYf9P1sTNdUXVJqY3YNJK7xLVwR0mxJFU6rCgEKnhSGIL2Eq8BdEERAX0OGwEiVQ1R0MaNFR8QfqKxmHigbX8VLjDz_Q0L8Wc_qPxDw",
|
||||
"expected_keywords": keywords,
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 30, "min_sentences": 1, "has_numbers": False},
|
||||
})
|
||||
|
||||
diagrams = [
|
||||
("flow_a", "https://dummyimage.com/1200x800/f8fafc/0f172a.png&text=Flowchart+API+Gateway+Queue+Worker", ["flowchart", "api", "worker"]),
|
||||
("flow_b", "https://dummyimage.com/1200x800/f1f5f9/0f172a.png&text=Architecture+Diagram+Database+Cache+Client", ["architecture", "diagram", "cache"]),
|
||||
("uml_a", "https://dummyimage.com/1200x800/e2e8f0/0f172a.png&text=Class+Diagram+User+Session+Message", ["class", "diagram", "session"]),
|
||||
("uml_b", "https://dummyimage.com/1200x800/cbd5e1/0f172a.png&text=Sequence+Diagram+Request+Response", ["sequence", "diagram", "response"]),
|
||||
("network_a", "https://dummyimage.com/1200x800/ffffff/111827.png&text=Network+Nodes+Edges+Router", ["network", "node", "router"]),
|
||||
("network_b", "https://dummyimage.com/1200x800/ffffff/1e293b.png&text=Service+Mesh+Proxy+Auth", ["service", "mesh", "auth"]),
|
||||
("state_machine", "https://dummyimage.com/1200x800/f8fafc/334155.png&text=State+Machine+Idle+Run+Stop", ["state", "machine", "idle"]),
|
||||
("mind_map", "https://dummyimage.com/1200x800/fefce8/1f2937.png&text=Mind+Map+Memory+Recall+Tools", ["mind", "memory", "tools"]),
|
||||
("pipeline", "https://dummyimage.com/1200x800/ecfeff/155e75.png&text=Pipeline+Ingest+Rank+Summarize", ["pipeline", "ingest", "summarize"]),
|
||||
("org_chart", "https://dummyimage.com/1200x800/fdf2f8/831843.png&text=Org+Chart+Lead+Review+Ops", ["org", "chart", "review"]),
|
||||
]
|
||||
for ident, url, keywords in diagrams:
|
||||
dataset.append({
|
||||
"id": f"diagram_{ident}",
|
||||
"url": url,
|
||||
"category": "diagram",
|
||||
"expected_keywords": ["architecture", "component", "service"],
|
||||
"expected_structure": {"min_length": 100, "min_sentences": 3},
|
||||
},
|
||||
# Photos
|
||||
{
|
||||
"id": "photo_nature",
|
||||
"url": "https://picsum.photos/seed/bench1/400/300",
|
||||
"expected_keywords": keywords,
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 50, "min_sentences": 2, "has_numbers": False},
|
||||
})
|
||||
|
||||
for idx in range(1, 11):
|
||||
dataset.append({
|
||||
"id": f"photo_random_{idx:02d}",
|
||||
"url": f"https://picsum.photos/seed/vision-bench-{idx}/640/480",
|
||||
"category": "photo",
|
||||
"expected_keywords": [],
|
||||
"expected_structure": {"min_length": 30, "min_sentences": 1},
|
||||
},
|
||||
# Charts
|
||||
{
|
||||
"id": "chart_bar",
|
||||
"url": "https://quickchart.io/chart?c={type:'bar',data:{labels:['Q1','Q2','Q3','Q4'],datasets:[{label:'Users',data:[50,60,70,80]}]}}",
|
||||
"category": "chart",
|
||||
"expected_keywords": ["bar", "chart", "data"],
|
||||
"expected_structure": {"min_length": 50, "min_sentences": 2},
|
||||
},
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 30, "min_sentences": 1, "has_numbers": False},
|
||||
})
|
||||
|
||||
charts = [
|
||||
("bar_quarterly", "https://quickchart.io/chart?c={type:'bar',data:{labels:['Q1','Q2','Q3','Q4'],datasets:[{label:'Revenue',data:[100,150,200,250]}]}}", ["bar", "chart", "revenue"]),
|
||||
("pie_market", "https://quickchart.io/chart?c={type:'pie',data:{labels:['A','B','C'],datasets:[{data:[30,50,20]}]}}", ["pie", "chart", "percentage"]),
|
||||
("line_temp", "https://quickchart.io/chart?c={type:'line',data:{labels:['Jan','Feb','Mar','Apr'],datasets:[{label:'Temperature',data:[5,8,12,18]}]}}", ["line", "chart", "temperature"]),
|
||||
("radar_skill", "https://quickchart.io/chart?c={type:'radar',data:{labels:['Speed','Power','Defense','Magic'],datasets:[{label:'Hero',data:[80,60,70,90]}]}}", ["radar", "chart", "skill"]),
|
||||
("stacked_cloud", "https://quickchart.io/chart?c={type:'bar',data:{labels:['2022','2023','2024'],datasets:[{label:'Cloud',data:[100,150,200]},{label:'On-prem',data:[200,180,160]}]},options:{scales:{x:{stacked:true},y:{stacked:true}}}}", ["stacked", "bar", "chart"]),
|
||||
("area_growth", "https://quickchart.io/chart?c={type:'line',data:{labels:['W1','W2','W3','W4'],datasets:[{label:'Growth',data:[10,15,18,24],fill:true}]}}", ["line", "growth", "chart"]),
|
||||
("scatter_eval", "https://quickchart.io/chart?c={type:'scatter',data:{datasets:[{label:'Runs',data:[{x:1,y:70},{x:2,y:75},{x:3,y:82}]}]}}", ["scatter", "chart", "runs"]),
|
||||
("horizontal_bar", "https://quickchart.io/chart?c={type:'bar',data:{labels:['UI','OCR','Docs'],datasets:[{label:'Score',data:[88,76,91]}]},options:{indexAxis:'y'}}", ["bar", "score", "ocr"]),
|
||||
("bubble_usage", "https://quickchart.io/chart?c={type:'bubble',data:{datasets:[{label:'Latency',data:[{x:1,y:120,r:8},{x:2,y:95,r:6},{x:3,y:180,r:10}]}]}}", ["bubble", "latency", "chart"]),
|
||||
("doughnut_devices", "https://quickchart.io/chart?c={type:'doughnut',data:{labels:['Desktop','Mobile','Tablet'],datasets:[{data:[60,30,10]}]}}", ["doughnut", "chart", "device"]),
|
||||
]
|
||||
for ident, url, keywords in charts:
|
||||
dataset.append({
|
||||
"id": f"chart_{ident}",
|
||||
"url": url,
|
||||
"category": "chart",
|
||||
"expected_keywords": keywords,
|
||||
"ground_truth_ocr": "",
|
||||
"expected_structure": {"min_length": 50, "min_sentences": 2, "has_numbers": True},
|
||||
})
|
||||
|
||||
ocr_texts = [
|
||||
"Hermes OCR Alpha 01",
|
||||
"Prompt Cache Hit 87%",
|
||||
"Session 42 Ready",
|
||||
"Latency 118 ms",
|
||||
"Voice Mode Enabled",
|
||||
]
|
||||
for idx, text in enumerate(ocr_texts, start=1):
|
||||
dataset.append({
|
||||
"id": f"ocr_text_{idx:02d}",
|
||||
"url": f"https://dummyimage.com/1200x320/ffffff/000000.png&text={text.replace(' ', '+')}",
|
||||
"category": "ocr",
|
||||
"expected_keywords": text.lower().split()[:2],
|
||||
"ground_truth_ocr": text,
|
||||
"expected_structure": {"min_length": 10, "min_sentences": 1, "has_numbers": any(ch.isdigit() for ch in text)},
|
||||
})
|
||||
|
||||
documents = [
|
||||
"Invoice 1001 Total 42 Due 2026-04-22",
|
||||
"Form A Name Alice Status Approved",
|
||||
"Report Memory Recall Score 91 Percent",
|
||||
"Checklist Crisis Escalation Call 988 Now",
|
||||
"Meeting Notes Vision Benchmark Run Pending",
|
||||
]
|
||||
for idx, text in enumerate(documents, start=1):
|
||||
dataset.append({
|
||||
"id": f"document_text_{idx:02d}",
|
||||
"url": f"https://dummyimage.com/1400x900/f8fafc/0f172a.png&text={text.replace(' ', '+')}",
|
||||
"category": "document",
|
||||
"expected_keywords": text.lower().split()[:3],
|
||||
"ground_truth_ocr": text,
|
||||
"expected_structure": {"min_length": 20, "min_sentences": 1, "has_numbers": any(ch.isdigit() for ch in text)},
|
||||
})
|
||||
|
||||
return dataset
|
||||
|
||||
|
||||
def load_dataset(path: str) -> List[dict]:
|
||||
@@ -585,7 +758,9 @@ async def main():
|
||||
parser.add_argument("--url", help="Single image URL to test")
|
||||
parser.add_argument("--category", default="photo", help="Category for single URL")
|
||||
parser.add_argument("--output", default=None, help="Output JSON file")
|
||||
parser.add_argument("--markdown-output", default=None, help="Optional markdown report output path")
|
||||
parser.add_argument("--runs", type=int, default=1, help="Runs per model per image")
|
||||
parser.add_argument("--limit", type=int, default=0, help="Limit to the first N images for smoke runs")
|
||||
parser.add_argument("--models", nargs="+", default=None,
|
||||
help="Models to test (default: all)")
|
||||
parser.add_argument("--markdown", action="store_true", help="Output markdown report")
|
||||
@@ -617,9 +792,14 @@ async def main():
|
||||
print("ERROR: Provide --images or --url")
|
||||
sys.exit(1)
|
||||
|
||||
if args.limit and args.limit > 0:
|
||||
images = images[:args.limit]
|
||||
|
||||
# Run benchmark
|
||||
report = await run_benchmark_suite(images, selected, args.runs)
|
||||
|
||||
markdown_report = to_markdown(report)
|
||||
|
||||
# Output
|
||||
if args.output:
|
||||
os.makedirs(os.path.dirname(args.output) or ".", exist_ok=True)
|
||||
@@ -627,8 +807,14 @@ async def main():
|
||||
json.dump(report, f, indent=2)
|
||||
print(f"\nResults saved to {args.output}")
|
||||
|
||||
if args.markdown_output:
|
||||
os.makedirs(os.path.dirname(args.markdown_output) or ".", exist_ok=True)
|
||||
with open(args.markdown_output, "w", encoding="utf-8") as f:
|
||||
f.write(markdown_report)
|
||||
print(f"Markdown report saved to {args.markdown_output}")
|
||||
|
||||
if args.markdown or not args.output:
|
||||
print("\n" + to_markdown(report))
|
||||
print("\n" + markdown_report)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
|
||||
67
metrics/vision-benchmark-smoke-2026-04-22.json
Normal file
67
metrics/vision-benchmark-smoke-2026-04-22.json
Normal file
@@ -0,0 +1,67 @@
|
||||
{
|
||||
"generated_at": "2026-04-22T16:21:56.271426+00:00",
|
||||
"config": {
|
||||
"total_images": 2,
|
||||
"runs_per_model": 1,
|
||||
"models": {
|
||||
"gemma4": "Gemma 4 27B",
|
||||
"gemini3_flash": "Gemini 3 Flash Preview"
|
||||
}
|
||||
},
|
||||
"results": [
|
||||
{
|
||||
"gemma4": {
|
||||
"success": false,
|
||||
"error": "nous:google/gemma-4-27b-it => No API key for provider nous | ollama:gemma4:latest => Server error '500 Internal Server Error' for url 'http://localhost:11434/api/chat'\nFor more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/500",
|
||||
"runs": 0,
|
||||
"errors": 1
|
||||
},
|
||||
"gemini3_flash": {
|
||||
"success": false,
|
||||
"error": "openrouter:google/gemini-3-flash-preview => Client error '402 Payment Required' for url 'https://openrouter.ai/api/v1/chat/completions'\nFor more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/402 | gemini:gemini-2.5-flash => Client error '429 Too Many Requests' for url 'https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent?key=AIzaSyAmIctJQG_b4VKV1sMLebBnouq6yCckEf0'\nFor more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/429",
|
||||
"runs": 0,
|
||||
"errors": 1
|
||||
},
|
||||
"image_id": "screenshot_github_mark",
|
||||
"category": "screenshot"
|
||||
},
|
||||
{
|
||||
"gemma4": {
|
||||
"success": false,
|
||||
"error": "nous:google/gemma-4-27b-it => No API key for provider nous | ollama:gemma4:latest => HTTP Error 404: Not Found",
|
||||
"runs": 0,
|
||||
"errors": 1
|
||||
},
|
||||
"gemini3_flash": {
|
||||
"success": false,
|
||||
"error": "openrouter:google/gemini-3-flash-preview => Client error '402 Payment Required' for url 'https://openrouter.ai/api/v1/chat/completions'\nFor more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/402 | gemini:gemini-2.5-flash => HTTP Error 404: Not Found",
|
||||
"runs": 0,
|
||||
"errors": 1
|
||||
},
|
||||
"image_id": "screenshot_github_social",
|
||||
"category": "screenshot"
|
||||
}
|
||||
],
|
||||
"summary": {
|
||||
"gemma4": {
|
||||
"success_rate": 0,
|
||||
"error": "All runs failed",
|
||||
"total_runs": 0,
|
||||
"total_failures": 2,
|
||||
"failure_examples": [
|
||||
"nous:google/gemma-4-27b-it => No API key for provider nous | ollama:gemma4:latest => HTTP Error 404: Not Found",
|
||||
"nous:google/gemma-4-27b-it => No API key for provider nous | ollama:gemma4:latest => Server error '500 Internal Server Error' for url 'http://localhost:11434/api/chat'\nFor more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/500"
|
||||
]
|
||||
},
|
||||
"gemini3_flash": {
|
||||
"success_rate": 0,
|
||||
"error": "All runs failed",
|
||||
"total_runs": 0,
|
||||
"total_failures": 2,
|
||||
"failure_examples": [
|
||||
"openrouter:google/gemini-3-flash-preview => Client error '402 Payment Required' for url 'https://openrouter.ai/api/v1/chat/completions'\nFor more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/402 | gemini:gemini-2.5-flash => Client error '429 Too Many Requests' for url 'https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent?key=AIzaSyAmIctJQG_b4VKV1sMLebBnouq6yCckEf0'\nFor more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/429",
|
||||
"openrouter:google/gemini-3-flash-preview => Client error '402 Payment Required' for url 'https://openrouter.ai/api/v1/chat/completions'\nFor more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/402 | gemini:gemini-2.5-flash => HTTP Error 404: Not Found"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
44
metrics/vision-benchmark-smoke-2026-04-22.md
Normal file
44
metrics/vision-benchmark-smoke-2026-04-22.md
Normal file
@@ -0,0 +1,44 @@
|
||||
# Vision Benchmark Report
|
||||
|
||||
Generated: 2026-04-22T16:21
|
||||
Images tested: 2
|
||||
Runs per model: 1
|
||||
Models: Gemma 4 27B, Gemini 3 Flash Preview
|
||||
|
||||
## Latency Comparison
|
||||
|
||||
| Model | Mean (ms) | Median | P95 | Std Dev |
|
||||
|-------|-----------|--------|-----|---------|
|
||||
|
||||
## Accuracy Comparison
|
||||
|
||||
| Model | OCR Accuracy | Keyword Coverage | Success Rate |
|
||||
|-------|-------------|-----------------|--------------|
|
||||
|
||||
## Token Usage
|
||||
|
||||
| Model | Mean Tokens/Image | Total Tokens |
|
||||
|-------|------------------|--------------|
|
||||
|
||||
## Failure Modes
|
||||
|
||||
### Gemma 4 27B
|
||||
- Summary: All runs failed
|
||||
- nous:google/gemma-4-27b-it => No API key for provider nous | ollama:gemma4:latest => HTTP Error 404: Not Found
|
||||
- nous:google/gemma-4-27b-it => No API key for provider nous | ollama:gemma4:latest => Server error '500 Internal Server Error' for url 'http://localhost:11434/api/chat'
|
||||
For more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/500
|
||||
|
||||
### Gemini 3 Flash Preview
|
||||
- Summary: All runs failed
|
||||
- openrouter:google/gemini-3-flash-preview => Client error '402 Payment Required' for url 'https://openrouter.ai/api/v1/chat/completions'
|
||||
For more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/402 | gemini:gemini-2.5-flash => Client error '429 Too Many Requests' for url 'https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent?key=AIzaSyAmIctJQG_b4VKV1sMLebBnouq6yCckEf0'
|
||||
For more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/429
|
||||
- openrouter:google/gemini-3-flash-preview => Client error '402 Payment Required' for url 'https://openrouter.ai/api/v1/chat/completions'
|
||||
For more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/402 | gemini:gemini-2.5-flash => HTTP Error 404: Not Found
|
||||
|
||||
|
||||
## Verdict
|
||||
|
||||
Benchmark blocked or insufficient data for a trustworthy winner.
|
||||
|
||||
Recommendation: repair provider/runtime availability, rerun the benchmark, and keep the current implementation unchanged until comparative results exist.
|
||||
@@ -1,213 +0,0 @@
|
||||
"""Regression tests: normalize_anthropic_response_v2 vs v1.
|
||||
|
||||
Constructs mock Anthropic responses and asserts that the v2 function
|
||||
(returning NormalizedResponse) produces identical field values to the
|
||||
original v1 function (returning SimpleNamespace + finish_reason).
|
||||
"""
|
||||
|
||||
from types import SimpleNamespace
|
||||
|
||||
import pytest
|
||||
|
||||
from agent.anthropic_adapter import (
|
||||
normalize_anthropic_response,
|
||||
normalize_anthropic_response_v2,
|
||||
)
|
||||
from agent.transports.types import NormalizedResponse
|
||||
|
||||
|
||||
def _text_block(text: str):
|
||||
return SimpleNamespace(type="text", text=text)
|
||||
|
||||
|
||||
def _thinking_block(thinking: str, signature: str = "sig_abc"):
|
||||
return SimpleNamespace(type="thinking", thinking=thinking, signature=signature)
|
||||
|
||||
|
||||
def _tool_use_block(id: str, name: str, input: dict):
|
||||
return SimpleNamespace(type="tool_use", id=id, name=name, input=input)
|
||||
|
||||
|
||||
def _response(content_blocks, stop_reason="end_turn"):
|
||||
return SimpleNamespace(
|
||||
content=content_blocks,
|
||||
stop_reason=stop_reason,
|
||||
usage=SimpleNamespace(input_tokens=10, output_tokens=5),
|
||||
)
|
||||
|
||||
|
||||
class TestTextOnly:
|
||||
def setup_method(self):
|
||||
self.resp = _response([_text_block("Hello world")])
|
||||
self.v1_msg, self.v1_finish = normalize_anthropic_response(self.resp)
|
||||
self.v2 = normalize_anthropic_response_v2(self.resp)
|
||||
|
||||
def test_type(self):
|
||||
assert isinstance(self.v2, NormalizedResponse)
|
||||
|
||||
def test_content_matches(self):
|
||||
assert self.v2.content == self.v1_msg.content
|
||||
|
||||
def test_finish_reason_matches(self):
|
||||
assert self.v2.finish_reason == self.v1_finish
|
||||
|
||||
def test_no_tool_calls(self):
|
||||
assert self.v2.tool_calls is None
|
||||
assert self.v1_msg.tool_calls is None
|
||||
|
||||
def test_no_reasoning(self):
|
||||
assert self.v2.reasoning is None
|
||||
assert self.v1_msg.reasoning is None
|
||||
|
||||
|
||||
class TestWithToolCalls:
|
||||
def setup_method(self):
|
||||
self.resp = _response(
|
||||
[
|
||||
_text_block("I'll check that"),
|
||||
_tool_use_block("toolu_abc", "terminal", {"command": "ls"}),
|
||||
_tool_use_block("toolu_def", "read_file", {"path": "/tmp"}),
|
||||
],
|
||||
stop_reason="tool_use",
|
||||
)
|
||||
self.v1_msg, self.v1_finish = normalize_anthropic_response(self.resp)
|
||||
self.v2 = normalize_anthropic_response_v2(self.resp)
|
||||
|
||||
def test_finish_reason(self):
|
||||
assert self.v2.finish_reason == "tool_calls"
|
||||
assert self.v1_finish == "tool_calls"
|
||||
|
||||
def test_tool_call_count(self):
|
||||
assert len(self.v2.tool_calls) == 2
|
||||
assert len(self.v1_msg.tool_calls) == 2
|
||||
|
||||
def test_tool_call_ids_match(self):
|
||||
for i in range(2):
|
||||
assert self.v2.tool_calls[i].id == self.v1_msg.tool_calls[i].id
|
||||
|
||||
def test_tool_call_names_match(self):
|
||||
assert self.v2.tool_calls[0].name == "terminal"
|
||||
assert self.v2.tool_calls[1].name == "read_file"
|
||||
for i in range(2):
|
||||
assert self.v2.tool_calls[i].name == self.v1_msg.tool_calls[i].function.name
|
||||
|
||||
def test_tool_call_arguments_match(self):
|
||||
for i in range(2):
|
||||
assert self.v2.tool_calls[i].arguments == self.v1_msg.tool_calls[i].function.arguments
|
||||
|
||||
def test_content_preserved(self):
|
||||
assert self.v2.content == self.v1_msg.content
|
||||
assert "check that" in self.v2.content
|
||||
|
||||
|
||||
class TestWithThinking:
|
||||
def setup_method(self):
|
||||
self.resp = _response([
|
||||
_thinking_block("Let me think about this carefully..."),
|
||||
_text_block("The answer is 42."),
|
||||
])
|
||||
self.v1_msg, self.v1_finish = normalize_anthropic_response(self.resp)
|
||||
self.v2 = normalize_anthropic_response_v2(self.resp)
|
||||
|
||||
def test_reasoning_matches(self):
|
||||
assert self.v2.reasoning == self.v1_msg.reasoning
|
||||
assert "think about this" in self.v2.reasoning
|
||||
|
||||
def test_reasoning_details_in_provider_data(self):
|
||||
v1_details = self.v1_msg.reasoning_details
|
||||
v2_details = self.v2.provider_data.get("reasoning_details") if self.v2.provider_data else None
|
||||
assert v1_details is not None
|
||||
assert v2_details is not None
|
||||
assert len(v2_details) == len(v1_details)
|
||||
|
||||
def test_content_excludes_thinking(self):
|
||||
assert self.v2.content == "The answer is 42."
|
||||
|
||||
|
||||
class TestMixed:
|
||||
def setup_method(self):
|
||||
self.resp = _response(
|
||||
[
|
||||
_thinking_block("Planning my approach..."),
|
||||
_text_block("I'll run the command"),
|
||||
_tool_use_block("toolu_xyz", "terminal", {"command": "pwd"}),
|
||||
],
|
||||
stop_reason="tool_use",
|
||||
)
|
||||
self.v1_msg, self.v1_finish = normalize_anthropic_response(self.resp)
|
||||
self.v2 = normalize_anthropic_response_v2(self.resp)
|
||||
|
||||
def test_all_fields_present(self):
|
||||
assert self.v2.content is not None
|
||||
assert self.v2.tool_calls is not None
|
||||
assert self.v2.reasoning is not None
|
||||
assert self.v2.finish_reason == "tool_calls"
|
||||
|
||||
def test_content_matches(self):
|
||||
assert self.v2.content == self.v1_msg.content
|
||||
|
||||
def test_reasoning_matches(self):
|
||||
assert self.v2.reasoning == self.v1_msg.reasoning
|
||||
|
||||
def test_tool_call_matches(self):
|
||||
assert self.v2.tool_calls[0].id == self.v1_msg.tool_calls[0].id
|
||||
assert self.v2.tool_calls[0].name == self.v1_msg.tool_calls[0].function.name
|
||||
|
||||
|
||||
class TestStopReasons:
|
||||
@pytest.mark.parametrize("stop_reason,expected", [
|
||||
("end_turn", "stop"),
|
||||
("tool_use", "tool_calls"),
|
||||
("max_tokens", "length"),
|
||||
("stop_sequence", "stop"),
|
||||
("refusal", "content_filter"),
|
||||
("model_context_window_exceeded", "length"),
|
||||
("unknown_future_reason", "stop"),
|
||||
])
|
||||
def test_stop_reason_mapping(self, stop_reason, expected):
|
||||
resp = _response([_text_block("x")], stop_reason=stop_reason)
|
||||
_v1_msg, v1_finish = normalize_anthropic_response(resp)
|
||||
v2 = normalize_anthropic_response_v2(resp)
|
||||
assert v2.finish_reason == v1_finish == expected
|
||||
|
||||
|
||||
class TestStripToolPrefix:
|
||||
def test_prefix_stripped(self):
|
||||
resp = _response(
|
||||
[_tool_use_block("toolu_1", "mcp_terminal", {"cmd": "ls"})],
|
||||
stop_reason="tool_use",
|
||||
)
|
||||
v1_msg, _ = normalize_anthropic_response(resp, strip_tool_prefix=True)
|
||||
v2 = normalize_anthropic_response_v2(resp, strip_tool_prefix=True)
|
||||
assert v1_msg.tool_calls[0].function.name == "terminal"
|
||||
assert v2.tool_calls[0].name == "terminal"
|
||||
|
||||
def test_prefix_kept(self):
|
||||
resp = _response(
|
||||
[_tool_use_block("toolu_1", "mcp_terminal", {"cmd": "ls"})],
|
||||
stop_reason="tool_use",
|
||||
)
|
||||
v1_msg, _ = normalize_anthropic_response(resp, strip_tool_prefix=False)
|
||||
v2 = normalize_anthropic_response_v2(resp, strip_tool_prefix=False)
|
||||
assert v1_msg.tool_calls[0].function.name == "mcp_terminal"
|
||||
assert v2.tool_calls[0].name == "mcp_terminal"
|
||||
|
||||
|
||||
class TestEdgeCases:
|
||||
def test_empty_content_blocks(self):
|
||||
resp = _response([])
|
||||
v1_msg, _v1_finish = normalize_anthropic_response(resp)
|
||||
v2 = normalize_anthropic_response_v2(resp)
|
||||
assert v2.content == v1_msg.content
|
||||
assert v2.content is None
|
||||
|
||||
def test_no_reasoning_details_means_none_provider_data(self):
|
||||
resp = _response([_text_block("hi")])
|
||||
v2 = normalize_anthropic_response_v2(resp)
|
||||
assert v2.provider_data is None
|
||||
|
||||
def test_v2_returns_dataclass_not_namespace(self):
|
||||
resp = _response([_text_block("hi")])
|
||||
v2 = normalize_anthropic_response_v2(resp)
|
||||
assert isinstance(v2, NormalizedResponse)
|
||||
assert not isinstance(v2, SimpleNamespace)
|
||||
@@ -1,208 +0,0 @@
|
||||
"""Tests for the transport ABC, registry, and AnthropicTransport."""
|
||||
|
||||
from types import SimpleNamespace
|
||||
|
||||
import pytest
|
||||
|
||||
from agent.transports import _REGISTRY, get_transport, register_transport
|
||||
from agent.transports.base import ProviderTransport
|
||||
from agent.transports.types import NormalizedResponse
|
||||
|
||||
|
||||
class TestProviderTransportABC:
|
||||
def test_cannot_instantiate_abc(self):
|
||||
with pytest.raises(TypeError):
|
||||
ProviderTransport()
|
||||
|
||||
def test_concrete_must_implement_all_abstract(self):
|
||||
class Incomplete(ProviderTransport):
|
||||
@property
|
||||
def api_mode(self):
|
||||
return "test"
|
||||
|
||||
with pytest.raises(TypeError):
|
||||
Incomplete()
|
||||
|
||||
def test_minimal_concrete(self):
|
||||
class Minimal(ProviderTransport):
|
||||
@property
|
||||
def api_mode(self):
|
||||
return "test_minimal"
|
||||
|
||||
def convert_messages(self, messages, **kw):
|
||||
return messages
|
||||
|
||||
def convert_tools(self, tools):
|
||||
return tools
|
||||
|
||||
def build_kwargs(self, model, messages, tools=None, **params):
|
||||
return {"model": model, "messages": messages}
|
||||
|
||||
def normalize_response(self, response, **kw):
|
||||
return NormalizedResponse(content="ok", tool_calls=None, finish_reason="stop")
|
||||
|
||||
t = Minimal()
|
||||
assert t.api_mode == "test_minimal"
|
||||
assert t.validate_response(None) is True
|
||||
assert t.extract_cache_stats(None) is None
|
||||
assert t.map_finish_reason("end_turn") == "end_turn"
|
||||
|
||||
|
||||
class TestTransportRegistry:
|
||||
def test_get_unregistered_returns_none(self):
|
||||
assert get_transport("nonexistent_mode") is None
|
||||
|
||||
def test_anthropic_registered_on_import(self):
|
||||
import agent.transports.anthropic # noqa: F401
|
||||
|
||||
t = get_transport("anthropic_messages")
|
||||
assert t is not None
|
||||
assert t.api_mode == "anthropic_messages"
|
||||
|
||||
def test_register_and_get(self):
|
||||
class DummyTransport(ProviderTransport):
|
||||
@property
|
||||
def api_mode(self):
|
||||
return "dummy_test"
|
||||
|
||||
def convert_messages(self, messages, **kw):
|
||||
return messages
|
||||
|
||||
def convert_tools(self, tools):
|
||||
return tools
|
||||
|
||||
def build_kwargs(self, model, messages, tools=None, **params):
|
||||
return {}
|
||||
|
||||
def normalize_response(self, response, **kw):
|
||||
return NormalizedResponse(content=None, tool_calls=None, finish_reason="stop")
|
||||
|
||||
register_transport("dummy_test", DummyTransport)
|
||||
t = get_transport("dummy_test")
|
||||
assert t.api_mode == "dummy_test"
|
||||
_REGISTRY.pop("dummy_test", None)
|
||||
|
||||
|
||||
class TestAnthropicTransport:
|
||||
@pytest.fixture
|
||||
def transport(self):
|
||||
import agent.transports.anthropic # noqa: F401
|
||||
|
||||
return get_transport("anthropic_messages")
|
||||
|
||||
def test_api_mode(self, transport):
|
||||
assert transport.api_mode == "anthropic_messages"
|
||||
|
||||
def test_convert_tools_simple(self, transport):
|
||||
tools = [{
|
||||
"type": "function",
|
||||
"function": {
|
||||
"name": "test_tool",
|
||||
"description": "A test",
|
||||
"parameters": {"type": "object", "properties": {}},
|
||||
},
|
||||
}]
|
||||
result = transport.convert_tools(tools)
|
||||
assert len(result) == 1
|
||||
assert result[0]["name"] == "test_tool"
|
||||
assert "input_schema" in result[0]
|
||||
|
||||
def test_validate_response_none(self, transport):
|
||||
assert transport.validate_response(None) is False
|
||||
|
||||
def test_validate_response_empty_content(self, transport):
|
||||
r = SimpleNamespace(content=[])
|
||||
assert transport.validate_response(r) is False
|
||||
|
||||
def test_validate_response_valid(self, transport):
|
||||
r = SimpleNamespace(content=[SimpleNamespace(type="text", text="hello")])
|
||||
assert transport.validate_response(r) is True
|
||||
|
||||
def test_map_finish_reason(self, transport):
|
||||
assert transport.map_finish_reason("end_turn") == "stop"
|
||||
assert transport.map_finish_reason("tool_use") == "tool_calls"
|
||||
assert transport.map_finish_reason("max_tokens") == "length"
|
||||
assert transport.map_finish_reason("stop_sequence") == "stop"
|
||||
assert transport.map_finish_reason("refusal") == "content_filter"
|
||||
assert transport.map_finish_reason("model_context_window_exceeded") == "length"
|
||||
assert transport.map_finish_reason("unknown") == "stop"
|
||||
|
||||
def test_extract_cache_stats_none_usage(self, transport):
|
||||
r = SimpleNamespace(usage=None)
|
||||
assert transport.extract_cache_stats(r) is None
|
||||
|
||||
def test_extract_cache_stats_with_cache(self, transport):
|
||||
usage = SimpleNamespace(cache_read_input_tokens=100, cache_creation_input_tokens=50)
|
||||
r = SimpleNamespace(usage=usage)
|
||||
result = transport.extract_cache_stats(r)
|
||||
assert result == {"cached_tokens": 100, "creation_tokens": 50}
|
||||
|
||||
def test_extract_cache_stats_zero(self, transport):
|
||||
usage = SimpleNamespace(cache_read_input_tokens=0, cache_creation_input_tokens=0)
|
||||
r = SimpleNamespace(usage=usage)
|
||||
assert transport.extract_cache_stats(r) is None
|
||||
|
||||
def test_normalize_response_text(self, transport):
|
||||
r = SimpleNamespace(
|
||||
content=[SimpleNamespace(type="text", text="Hello world")],
|
||||
stop_reason="end_turn",
|
||||
usage=SimpleNamespace(input_tokens=10, output_tokens=5),
|
||||
model="claude-sonnet-4-6",
|
||||
)
|
||||
nr = transport.normalize_response(r)
|
||||
assert isinstance(nr, NormalizedResponse)
|
||||
assert nr.content == "Hello world"
|
||||
assert nr.tool_calls is None or nr.tool_calls == []
|
||||
assert nr.finish_reason == "stop"
|
||||
|
||||
def test_normalize_response_tool_calls(self, transport):
|
||||
r = SimpleNamespace(
|
||||
content=[
|
||||
SimpleNamespace(type="tool_use", id="toolu_123", name="terminal", input={"command": "ls"}),
|
||||
],
|
||||
stop_reason="tool_use",
|
||||
usage=SimpleNamespace(input_tokens=10, output_tokens=20),
|
||||
model="claude-sonnet-4-6",
|
||||
)
|
||||
nr = transport.normalize_response(r)
|
||||
assert nr.finish_reason == "tool_calls"
|
||||
assert len(nr.tool_calls) == 1
|
||||
tc = nr.tool_calls[0]
|
||||
assert tc.name == "terminal"
|
||||
assert tc.id == "toolu_123"
|
||||
assert '"command"' in tc.arguments
|
||||
|
||||
def test_normalize_response_thinking(self, transport):
|
||||
r = SimpleNamespace(
|
||||
content=[
|
||||
SimpleNamespace(type="thinking", thinking="Let me think..."),
|
||||
SimpleNamespace(type="text", text="The answer is 42"),
|
||||
],
|
||||
stop_reason="end_turn",
|
||||
usage=SimpleNamespace(input_tokens=10, output_tokens=15),
|
||||
model="claude-sonnet-4-6",
|
||||
)
|
||||
nr = transport.normalize_response(r)
|
||||
assert nr.content == "The answer is 42"
|
||||
assert nr.reasoning == "Let me think..."
|
||||
|
||||
def test_build_kwargs_returns_dict(self, transport):
|
||||
messages = [{"role": "user", "content": "Hello"}]
|
||||
kw = transport.build_kwargs(
|
||||
model="claude-sonnet-4-6",
|
||||
messages=messages,
|
||||
max_tokens=1024,
|
||||
)
|
||||
assert isinstance(kw, dict)
|
||||
assert "model" in kw
|
||||
assert "max_tokens" in kw
|
||||
assert "messages" in kw
|
||||
|
||||
def test_convert_messages_extracts_system(self, transport):
|
||||
messages = [
|
||||
{"role": "system", "content": "You are helpful."},
|
||||
{"role": "user", "content": "Hi"},
|
||||
]
|
||||
system, msgs = transport.convert_messages(messages)
|
||||
assert system is not None
|
||||
assert len(msgs) >= 1
|
||||
@@ -1,130 +0,0 @@
|
||||
"""Tests for agent/transports/types.py — dataclass construction + helpers."""
|
||||
|
||||
import json
|
||||
|
||||
from agent.transports.types import (
|
||||
NormalizedResponse,
|
||||
ToolCall,
|
||||
Usage,
|
||||
build_tool_call,
|
||||
map_finish_reason,
|
||||
)
|
||||
|
||||
|
||||
class TestToolCall:
|
||||
def test_basic_construction(self):
|
||||
tc = ToolCall(id="call_abc", name="terminal", arguments='{"cmd": "ls"}')
|
||||
assert tc.id == "call_abc"
|
||||
assert tc.name == "terminal"
|
||||
assert tc.arguments == '{"cmd": "ls"}'
|
||||
assert tc.provider_data is None
|
||||
|
||||
def test_none_id(self):
|
||||
tc = ToolCall(id=None, name="read_file", arguments="{}")
|
||||
assert tc.id is None
|
||||
|
||||
def test_provider_data(self):
|
||||
tc = ToolCall(
|
||||
id="call_x",
|
||||
name="t",
|
||||
arguments="{}",
|
||||
provider_data={"call_id": "call_x", "response_item_id": "fc_x"},
|
||||
)
|
||||
assert tc.provider_data["call_id"] == "call_x"
|
||||
assert tc.provider_data["response_item_id"] == "fc_x"
|
||||
|
||||
|
||||
class TestUsage:
|
||||
def test_defaults(self):
|
||||
u = Usage()
|
||||
assert u.prompt_tokens == 0
|
||||
assert u.completion_tokens == 0
|
||||
assert u.total_tokens == 0
|
||||
assert u.cached_tokens == 0
|
||||
|
||||
def test_explicit(self):
|
||||
u = Usage(prompt_tokens=100, completion_tokens=50, total_tokens=150, cached_tokens=80)
|
||||
assert u.total_tokens == 150
|
||||
|
||||
|
||||
class TestNormalizedResponse:
|
||||
def test_text_only(self):
|
||||
r = NormalizedResponse(content="hello", tool_calls=None, finish_reason="stop")
|
||||
assert r.content == "hello"
|
||||
assert r.tool_calls is None
|
||||
assert r.finish_reason == "stop"
|
||||
assert r.reasoning is None
|
||||
assert r.usage is None
|
||||
assert r.provider_data is None
|
||||
|
||||
def test_with_tool_calls(self):
|
||||
tcs = [ToolCall(id="call_1", name="terminal", arguments='{"cmd":"pwd"}')]
|
||||
r = NormalizedResponse(content=None, tool_calls=tcs, finish_reason="tool_calls")
|
||||
assert r.finish_reason == "tool_calls"
|
||||
assert len(r.tool_calls) == 1
|
||||
assert r.tool_calls[0].name == "terminal"
|
||||
|
||||
def test_with_reasoning(self):
|
||||
r = NormalizedResponse(
|
||||
content="answer",
|
||||
tool_calls=None,
|
||||
finish_reason="stop",
|
||||
reasoning="I thought about it",
|
||||
)
|
||||
assert r.reasoning == "I thought about it"
|
||||
|
||||
def test_with_provider_data(self):
|
||||
r = NormalizedResponse(
|
||||
content=None,
|
||||
tool_calls=None,
|
||||
finish_reason="stop",
|
||||
provider_data={"reasoning_details": [{"type": "thinking", "thinking": "hmm"}]},
|
||||
)
|
||||
assert r.provider_data["reasoning_details"][0]["type"] == "thinking"
|
||||
|
||||
|
||||
class TestBuildToolCall:
|
||||
def test_dict_arguments_serialized(self):
|
||||
tc = build_tool_call(id="call_1", name="terminal", arguments={"cmd": "ls"})
|
||||
assert tc.arguments == json.dumps({"cmd": "ls"})
|
||||
assert tc.provider_data is None
|
||||
|
||||
def test_string_arguments_passthrough(self):
|
||||
tc = build_tool_call(id="call_2", name="read_file", arguments='{"path": "/tmp"}')
|
||||
assert tc.arguments == '{"path": "/tmp"}'
|
||||
|
||||
def test_provider_fields(self):
|
||||
tc = build_tool_call(
|
||||
id="call_3",
|
||||
name="terminal",
|
||||
arguments="{}",
|
||||
call_id="call_3",
|
||||
response_item_id="fc_3",
|
||||
)
|
||||
assert tc.provider_data == {"call_id": "call_3", "response_item_id": "fc_3"}
|
||||
|
||||
def test_none_id(self):
|
||||
tc = build_tool_call(id=None, name="t", arguments="{}")
|
||||
assert tc.id is None
|
||||
|
||||
|
||||
class TestMapFinishReason:
|
||||
ANTHROPIC_MAP = {
|
||||
"end_turn": "stop",
|
||||
"tool_use": "tool_calls",
|
||||
"max_tokens": "length",
|
||||
"stop_sequence": "stop",
|
||||
"refusal": "content_filter",
|
||||
}
|
||||
|
||||
def test_known_reason(self):
|
||||
assert map_finish_reason("end_turn", self.ANTHROPIC_MAP) == "stop"
|
||||
assert map_finish_reason("tool_use", self.ANTHROPIC_MAP) == "tool_calls"
|
||||
assert map_finish_reason("max_tokens", self.ANTHROPIC_MAP) == "length"
|
||||
assert map_finish_reason("refusal", self.ANTHROPIC_MAP) == "content_filter"
|
||||
|
||||
def test_unknown_reason_defaults_to_stop(self):
|
||||
assert map_finish_reason("something_new", self.ANTHROPIC_MAP) == "stop"
|
||||
|
||||
def test_none_reason(self):
|
||||
assert map_finish_reason(None, self.ANTHROPIC_MAP) == "stop"
|
||||
@@ -199,7 +199,7 @@ class TestMarkdown:
|
||||
class TestDataset:
|
||||
def test_sample_dataset_has_entries(self):
|
||||
dataset = generate_sample_dataset()
|
||||
assert len(dataset) >= 4
|
||||
assert len(dataset) >= 50
|
||||
|
||||
def test_sample_dataset_structure(self):
|
||||
dataset = generate_sample_dataset()
|
||||
@@ -216,6 +216,9 @@ class TestDataset:
|
||||
assert "screenshot" in categories
|
||||
assert "diagram" in categories
|
||||
assert "photo" in categories
|
||||
assert "chart" in categories
|
||||
assert "ocr" in categories
|
||||
assert "document" in categories
|
||||
|
||||
|
||||
class TestModels:
|
||||
|
||||
21
tests/test_vision_benchmark_artifacts.py
Normal file
21
tests/test_vision_benchmark_artifacts.py
Normal file
@@ -0,0 +1,21 @@
|
||||
import json
|
||||
from pathlib import Path
|
||||
|
||||
|
||||
DATASET = Path("benchmarks/test_images.json")
|
||||
REPORT = Path("metrics/vision-benchmark-smoke-2026-04-22.md")
|
||||
|
||||
|
||||
def test_benchmark_dataset_is_issue_sized_and_category_complete() -> None:
|
||||
items = json.loads(DATASET.read_text(encoding="utf-8"))
|
||||
assert len(items) >= 50
|
||||
categories = {item["category"] for item in items}
|
||||
assert {"screenshot", "diagram", "photo", "ocr", "chart", "document"}.issubset(categories)
|
||||
|
||||
|
||||
def test_metrics_report_exists_with_recommendation() -> None:
|
||||
assert REPORT.exists(), "missing benchmark report under metrics/"
|
||||
text = REPORT.read_text(encoding="utf-8")
|
||||
assert "Recommendation" in text
|
||||
assert "Gemma 4" in text
|
||||
assert "Gemini" in text
|
||||
Reference in New Issue
Block a user