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>
83 lines
2.5 KiB
Python
83 lines
2.5 KiB
Python
"""Tests for SFT policy — learning from traces."""
|
|
|
|
from __future__ import annotations
|
|
|
|
from dataclasses import dataclass, field
|
|
from typing import Optional
|
|
|
|
from openjarvis.learning.sft_policy import SFTRouterPolicy
|
|
|
|
|
|
@dataclass
|
|
class _MockTrace:
|
|
query: str = ""
|
|
model: str = "model-a"
|
|
outcome: str = "success"
|
|
feedback: Optional[float] = 0.8
|
|
steps: list = field(default_factory=list)
|
|
total_latency_seconds: float = 1.0
|
|
|
|
|
|
class _MockTraceStore:
|
|
def __init__(self, traces):
|
|
self._traces = traces
|
|
|
|
def list_traces(self):
|
|
return self._traces
|
|
|
|
|
|
class TestSFTRouterPolicy:
|
|
def test_empty_traces(self):
|
|
policy = SFTRouterPolicy()
|
|
store = _MockTraceStore([])
|
|
result = policy.update(store)
|
|
assert result["updated"] is False
|
|
|
|
def test_learns_from_traces(self):
|
|
traces = [
|
|
_MockTrace(
|
|
query=f"def foo{i}(): pass",
|
|
model="code-model", outcome="success",
|
|
feedback=0.9,
|
|
)
|
|
for i in range(6)
|
|
]
|
|
policy = SFTRouterPolicy(min_samples=5)
|
|
store = _MockTraceStore(traces)
|
|
result = policy.update(store)
|
|
assert result["updated"] is True
|
|
assert "code" in result["policy_map"]
|
|
assert result["policy_map"]["code"] == "code-model"
|
|
|
|
def test_min_samples_threshold(self):
|
|
traces = [
|
|
_MockTrace(query="def foo(): pass", model="code-model", outcome="success")
|
|
for _ in range(3)
|
|
]
|
|
policy = SFTRouterPolicy(min_samples=5)
|
|
store = _MockTraceStore(traces)
|
|
result = policy.update(store)
|
|
assert result["updated"] is False
|
|
|
|
def test_classify_code(self):
|
|
assert SFTRouterPolicy._classify_query("def hello(): pass") == "code"
|
|
|
|
def test_classify_math(self):
|
|
assert SFTRouterPolicy._classify_query("solve the integral") == "math"
|
|
|
|
def test_classify_short(self):
|
|
assert SFTRouterPolicy._classify_query("hello world") == "short"
|
|
|
|
def test_classify_general(self):
|
|
query = "tell me about " + " ".join(["something"] * 20)
|
|
assert SFTRouterPolicy._classify_query(query) == "general"
|
|
|
|
def test_policy_map_property(self):
|
|
policy = SFTRouterPolicy()
|
|
assert policy.policy_map == {}
|
|
|
|
def test_is_intelligence_policy(self):
|
|
from openjarvis.learning._stubs import IntelligenceLearningPolicy
|
|
assert issubclass(SFTRouterPolicy, IntelligenceLearningPolicy)
|
|
assert SFTRouterPolicy.target == "intelligence"
|