Spaces:
Paused
Paused
File size: 6,425 Bytes
db4fa53 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 | from pathlib import Path
import json
from osint_env.training.config import load_self_play_config
def test_self_play_config_defaults_when_missing():
cfg = load_self_play_config("/tmp/does_not_exist_for_self_play_config.json")
assert cfg.rounds >= 1
assert cfg.pipeline_mode in {"legacy", "swarm_v2"}
assert cfg.model_topology in {"dual", "shared"}
assert cfg.phase_schedule in {"generator_answerer", "answerer_generator_answerer"}
assert cfg.tuning_mode in {"full", "lora"}
assert cfg.generator_phase.max_steps >= 1
assert cfg.answerer_phase.max_steps >= 1
assert cfg.generator_reward_weights.hardness > 0.0
assert cfg.swarm_v2.generator_swarm.shared_context is True
assert cfg.swarm_v2.validation.max_support_edges >= 1
assert cfg.wandb_enabled is False
assert cfg.wandb_project == "osint-self-play"
assert cfg.canonical_graph_mode == "generate"
def test_self_play_config_parses_overrides(tmp_path: Path):
cfg_path = tmp_path / "self_play.json"
cfg_path.write_text(
json.dumps(
{
"rounds": 5,
"output_dir": "artifacts/custom_self_play",
"dry_run": False,
"pipeline_mode": "swarm_v2",
"wandb_enabled": True,
"wandb_project": "osint-train-tests",
"wandb_entity": "example-team",
"wandb_run_name_prefix": "ci-self-play",
"canonical_graph_mode": "fixed",
"model_topology": "shared",
"phase_schedule": "answerer_generator_answerer",
"tuning_mode": "lora",
"shared_model_name_or_path": "/models/local-base",
"seed_tasks_per_round": 12,
"generated_tasks_per_round": 18,
"swarm_v2": {
"generator_swarm": {
"shared_context": True,
"max_agents": 5,
"max_breadth": 4,
"max_depth": 3,
"planner_rounds": 3,
"tools_per_agent": 2,
},
"answerer_swarm": {
"shared_context": True,
"max_agents": 4,
"max_breadth": 3,
"max_depth": 2,
"planner_rounds": 2,
"tools_per_agent": 2,
},
"validation": {
"max_support_edges": 6,
"max_path_hops": 3,
"max_context_nodes": 10,
"max_context_edges": 6,
"duplicate_similarity_threshold": 0.75,
},
"shared_context": {
"shared_by_default": True,
"max_nodes": 10,
"max_edges": 6,
"target_pressure": 0.9,
},
},
"generator_reward_weights": {
"validity": 0.2,
"hardness": 0.6,
"diversity": 0.1,
"consistency": 0.1,
},
"lora": {
"r": 32,
"alpha": 64,
"dropout": 0.1,
"target_modules": ["q_proj", "v_proj"],
"bias": "none",
"task_type": "CAUSAL_LM",
},
"generator_phase": {
"model_name_or_path": "Qwen/Qwen2.5-3B-Instruct",
"max_steps": 77,
"num_generations": 6,
"loss_type": "grpo",
"scale_rewards": "group",
"output_subdir": "gen_phase",
},
"answerer_phase": {
"model_name_or_path": "Qwen/Qwen2.5-1.5B-Instruct",
"max_steps": 55,
"num_generations": 5,
"output_subdir": "ans_phase",
},
}
),
encoding="utf-8",
)
cfg = load_self_play_config(cfg_path)
assert cfg.rounds == 5
assert cfg.output_dir == "artifacts/custom_self_play"
assert cfg.dry_run is False
assert cfg.pipeline_mode == "swarm_v2"
assert cfg.wandb_enabled is True
assert cfg.wandb_project == "osint-train-tests"
assert cfg.wandb_entity == "example-team"
assert cfg.wandb_run_name_prefix == "ci-self-play"
assert cfg.canonical_graph_mode == "fixed"
assert cfg.model_topology == "shared"
assert cfg.phase_schedule == "answerer_generator_answerer"
assert cfg.tuning_mode == "lora"
assert cfg.shared_model_name_or_path == "/models/local-base"
assert cfg.seed_tasks_per_round == 12
assert cfg.generated_tasks_per_round == 18
assert cfg.swarm_v2.generator_swarm.max_agents == 5
assert cfg.swarm_v2.answerer_swarm.max_agents == 4
assert cfg.swarm_v2.validation.max_support_edges == 6
assert cfg.swarm_v2.shared_context.target_pressure == 0.9
assert cfg.generator_reward_weights.hardness == 0.6
assert cfg.lora.r == 32
assert cfg.lora.alpha == 64
assert cfg.lora.target_modules == ["q_proj", "v_proj"]
assert cfg.generator_phase.model_name_or_path == "Qwen/Qwen2.5-3B-Instruct"
assert cfg.generator_phase.max_steps == 77
assert cfg.generator_phase.num_generations == 6
assert cfg.generator_phase.loss_type == "grpo"
assert cfg.generator_phase.scale_rewards == "group"
assert cfg.generator_phase.output_subdir == "gen_phase"
assert cfg.answerer_phase.model_name_or_path == "Qwen/Qwen2.5-1.5B-Instruct"
assert cfg.answerer_phase.max_steps == 55
assert cfg.answerer_phase.num_generations == 5
assert cfg.answerer_phase.output_subdir == "ans_phase"
def test_self_play_config_keeps_legacy_mode_when_not_set(tmp_path: Path):
cfg_path = tmp_path / "legacy_self_play.json"
cfg_path.write_text(
json.dumps(
{
"rounds": 2,
"max_support_edges": 11,
}
),
encoding="utf-8",
)
cfg = load_self_play_config(cfg_path)
assert cfg.pipeline_mode == "legacy"
assert cfg.max_support_edges == 11
assert cfg.swarm_v2.validation.max_support_edges == 11
|