mirror of
https://github.com/open-jarvis/OpenJarvis.git
synced 2026-07-30 19:02:16 +00:00
Jon Saad-FalconandClaude Opus 4.6
8d538cd1b0
Major changes across parallel sessions: - Add orchestrator SFT & GRPO training subpackage (learning/orchestrator/) with episode types, multi-objective reward, prompt registry, policy model, RL environment, and registered learning policies - Add structured THOUGHT/TOOL/INPUT/FINAL_ANSWER mode to OrchestratorAgent - Add 15 channel backends (Discord, Slack, Telegram, Email, Webhook, IRC, Matrix, Teams, WhatsApp, Signal, Mattermost, BlueBubbles, Feishu, Google Chat, Webchat) with channel tools and config - Add LiteLLM engine backend for unified LLM provider access - Add RLM agent and REPL tool - Remove ToolLearningPolicy — learning taxonomy now only targets Intelligence (LM weights/routing) and Agents (logic/ICL/tool strategies) - Rename SFTPolicy to SFTRouterPolicy (backward-compat alias kept) - Remove OpenClaw agent infrastructure (openclaw*.py, openclaw_bridge.py) - Fix async streaming tests (asyncio.run vs deprecated get_event_loop) - Fix server channel route tests (pytest.importorskip for optional fastapi) - Track uv.lock for reproducibility - Update CLAUDE.md and docs to reflect all changes 1676 tests pass, 37 skipped. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
128 lines
4.1 KiB
Python
128 lines
4.1 KiB
Python
"""Tests for orchestrator policy model."""
|
|
|
|
from __future__ import annotations
|
|
|
|
import pytest
|
|
|
|
from openjarvis.learning.orchestrator.policy_model import (
|
|
OrchestratorPolicyModel,
|
|
)
|
|
from openjarvis.learning.orchestrator.types import (
|
|
EpisodeState,
|
|
OrchestratorAction,
|
|
OrchestratorObservation,
|
|
)
|
|
|
|
|
|
class TestParseOutput:
|
|
"""Test _parse_output without loading a real model."""
|
|
|
|
def _model(self) -> OrchestratorPolicyModel:
|
|
return OrchestratorPolicyModel()
|
|
|
|
def test_valid_thought_tool_input(self):
|
|
m = self._model()
|
|
text = (
|
|
"THOUGHT: I need to calculate 2+2\n"
|
|
"TOOL: calculator\n"
|
|
"INPUT: 2+2"
|
|
)
|
|
po = m._parse_output(text, ["calculator", "think"])
|
|
assert po.thought == "I need to calculate 2+2"
|
|
assert po.tool_name == "calculator"
|
|
assert po.tool_input == "2+2"
|
|
assert po.is_final_answer is False
|
|
|
|
def test_final_answer(self):
|
|
m = self._model()
|
|
text = (
|
|
"THOUGHT: I have the result\n"
|
|
"FINAL_ANSWER: 42"
|
|
)
|
|
po = m._parse_output(text, ["calculator"])
|
|
assert po.is_final_answer is True
|
|
assert po.tool_input == "42"
|
|
|
|
def test_final_answer_with_space(self):
|
|
m = self._model()
|
|
text = "FINAL ANSWER: the result is 7"
|
|
po = m._parse_output(text, ["calculator"])
|
|
assert po.is_final_answer is True
|
|
|
|
def test_missing_fields_fallback(self):
|
|
m = self._model()
|
|
text = "just some random output"
|
|
po = m._parse_output(text, ["calculator", "think"])
|
|
# Should fallback to first available tool
|
|
assert po.tool_name == "calculator"
|
|
assert po.thought == "No thought provided"
|
|
|
|
def test_invalid_tool_name_fallback(self):
|
|
m = self._model()
|
|
text = "THOUGHT: reason\nTOOL: nonexistent_tool\nINPUT: hello"
|
|
po = m._parse_output(text, ["calculator", "think"])
|
|
assert po.tool_name == "calculator" # fallback
|
|
|
|
def test_case_insensitive_tool_match(self):
|
|
m = self._model()
|
|
text = "THOUGHT: reason\nTOOL: Calculator\nINPUT: 5+5"
|
|
po = m._parse_output(text, ["calculator", "think"])
|
|
assert po.tool_name == "calculator"
|
|
|
|
def test_empty_tools_list(self):
|
|
m = self._model()
|
|
text = "THOUGHT: reason\nTOOL: calc\nINPUT: 1"
|
|
po = m._parse_output(text, [])
|
|
assert po.tool_name == "unknown"
|
|
|
|
|
|
class TestBuildPrompt:
|
|
def test_includes_task(self):
|
|
m = OrchestratorPolicyModel()
|
|
state = EpisodeState(initial_prompt="What is 2+2?")
|
|
prompt = m._build_prompt(state, ["calculator"])
|
|
assert "What is 2+2?" in prompt
|
|
|
|
def test_includes_tools(self):
|
|
m = OrchestratorPolicyModel()
|
|
state = EpisodeState(initial_prompt="q")
|
|
prompt = m._build_prompt(state, ["calculator", "think"])
|
|
assert "calculator" in prompt
|
|
assert "think" in prompt
|
|
|
|
def test_includes_history(self):
|
|
m = OrchestratorPolicyModel()
|
|
state = EpisodeState(initial_prompt="q")
|
|
action = OrchestratorAction(
|
|
thought="use calc", tool_name="calculator", tool_input="2+2"
|
|
)
|
|
obs = OrchestratorObservation(content="4")
|
|
state.add_turn(action, obs)
|
|
prompt = m._build_prompt(state, ["calculator"])
|
|
assert "Turn 1:" in prompt
|
|
assert "use calc" in prompt
|
|
|
|
def test_format_instructions(self):
|
|
m = OrchestratorPolicyModel()
|
|
state = EpisodeState(initial_prompt="q")
|
|
prompt = m._build_prompt(state, ["calculator"])
|
|
assert "THOUGHT:" in prompt
|
|
assert "TOOL:" in prompt
|
|
assert "INPUT:" in prompt
|
|
|
|
|
|
class TestPredictActionRequiresModel:
|
|
def test_raises_without_model(self):
|
|
m = OrchestratorPolicyModel()
|
|
state = EpisodeState(initial_prompt="q")
|
|
with pytest.raises(RuntimeError, match="Cannot generate"):
|
|
m.predict_action(state, ["calculator"])
|
|
|
|
|
|
class TestRepr:
|
|
def test_repr(self):
|
|
m = OrchestratorPolicyModel()
|
|
r = repr(m)
|
|
assert "OrchestratorPolicyModel" in r
|
|
assert "None" in r
|