mirrors/hermes-agent
1
0
Fork 0
mirror of https://github.com/NousResearch/hermes-agent.git synced 2026-04-25 00:51:20 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

More major refactor/tech debt removal!

Browse source
This commit is contained in:
teknium1 2026-02-21 20:22:33 -08:00
parent 6134939882
commit 08ff1c1aa8
22 changed files with 1394 additions and 2315 deletions
Download patch file Download diff file Expand all files Collapse all files

41
tools/rl_training_tool.py
View file

@ -1337,3 +1337,44 @@ def get_missing_keys() -> List[str]:
if not os.getenv("WANDB_API_KEY"):
missing.append("WANDB_API_KEY")
return missing
# ---------------------------------------------------------------------------
# Schemas + Registry
# ---------------------------------------------------------------------------
from tools.registry import registry
RL_LIST_ENVIRONMENTS_SCHEMA = {"name": "rl_list_environments", "description": "List all available RL environments. Returns environment names, paths, and descriptions. TIP: Read the file_path with file tools to understand how each environment works (verifiers, data loading, rewards).", "parameters": {"type": "object", "properties": {}, "required": []}}
RL_SELECT_ENVIRONMENT_SCHEMA = {"name": "rl_select_environment", "description": "Select an RL environment for training. Loads the environment's default configuration. After selecting, use rl_get_current_config() to see settings and rl_edit_config() to modify them.", "parameters": {"type": "object", "properties": {"name": {"type": "string", "description": "Name of the environment to select (from rl_list_environments)"}}, "required": ["name"]}}
RL_GET_CURRENT_CONFIG_SCHEMA = {"name": "rl_get_current_config", "description": "Get the current environment configuration. Returns only fields that can be modified: group_size, max_token_length, total_steps, steps_per_eval, use_wandb, wandb_name, max_num_workers.", "parameters": {"type": "object", "properties": {}, "required": []}}
RL_EDIT_CONFIG_SCHEMA = {"name": "rl_edit_config", "description": "Update a configuration field. Use rl_get_current_config() first to see all available fields for the selected environment. Each environment has different configurable options. Infrastructure settings (tokenizer, URLs, lora_rank, learning_rate) are locked.", "parameters": {"type": "object", "properties": {"field": {"type": "string", "description": "Name of the field to update (get available fields from rl_get_current_config)"}, "value": {"description": "New value for the field"}}, "required": ["field", "value"]}}
RL_START_TRAINING_SCHEMA = {"name": "rl_start_training", "description": "Start a new RL training run with the current environment and config. Most training parameters (lora_rank, learning_rate, etc.) are fixed. Use rl_edit_config() to set group_size, batch_size, wandb_project before starting. WARNING: Training takes hours.", "parameters": {"type": "object", "properties": {}, "required": []}}
RL_CHECK_STATUS_SCHEMA = {"name": "rl_check_status", "description": "Get status and metrics for a training run. RATE LIMITED: enforces 30-minute minimum between checks for the same run. Returns WandB metrics: step, state, reward_mean, loss, percent_correct.", "parameters": {"type": "object", "properties": {"run_id": {"type": "string", "description": "The run ID from rl_start_training()"}}, "required": ["run_id"]}}
RL_STOP_TRAINING_SCHEMA = {"name": "rl_stop_training", "description": "Stop a running training job. Use if metrics look bad, training is stagnant, or you want to try different settings.", "parameters": {"type": "object", "properties": {"run_id": {"type": "string", "description": "The run ID to stop"}}, "required": ["run_id"]}}
RL_GET_RESULTS_SCHEMA = {"name": "rl_get_results", "description": "Get final results and metrics for a completed training run. Returns final metrics and path to trained weights.", "parameters": {"type": "object", "properties": {"run_id": {"type": "string", "description": "The run ID to get results for"}}, "required": ["run_id"]}}
RL_LIST_RUNS_SCHEMA = {"name": "rl_list_runs", "description": "List all training runs (active and completed) with their status.", "parameters": {"type": "object", "properties": {}, "required": []}}
RL_TEST_INFERENCE_SCHEMA = {"name": "rl_test_inference", "description": "Quick inference test for any environment. Runs a few steps of inference + scoring using OpenRouter. Default: 3 steps x 16 completions = 48 rollouts per model, testing 3 models = 144 total. Tests environment loading, prompt construction, inference parsing, and verifier logic. Use BEFORE training to catch issues.", "parameters": {"type": "object", "properties": {"num_steps": {"type": "integer", "description": "Number of steps to run (default: 3, recommended max for testing)", "default": 3}, "group_size": {"type": "integer", "description": "Completions per step (default: 16, like training)", "default": 16}, "models": {"type": "array", "items": {"type": "string"}, "description": "Optional list of OpenRouter model IDs. Default: qwen/qwen3-8b, z-ai/glm-4.7-flash, minimax/minimax-m2.1"}}, "required": []}}
_rl_env = ["TINKER_API_KEY", "WANDB_API_KEY"]
registry.register(name="rl_list_environments", toolset="rl", schema=RL_LIST_ENVIRONMENTS_SCHEMA,
handler=lambda args, **kw: rl_list_environments(), check_fn=check_rl_api_keys, requires_env=_rl_env, is_async=True)
registry.register(name="rl_select_environment", toolset="rl", schema=RL_SELECT_ENVIRONMENT_SCHEMA,
handler=lambda args, **kw: rl_select_environment(name=args.get("name", "")), check_fn=check_rl_api_keys, requires_env=_rl_env, is_async=True)
registry.register(name="rl_get_current_config", toolset="rl", schema=RL_GET_CURRENT_CONFIG_SCHEMA,
handler=lambda args, **kw: rl_get_current_config(), check_fn=check_rl_api_keys, requires_env=_rl_env, is_async=True)
registry.register(name="rl_edit_config", toolset="rl", schema=RL_EDIT_CONFIG_SCHEMA,
handler=lambda args, **kw: rl_edit_config(field=args.get("field", ""), value=args.get("value")), check_fn=check_rl_api_keys, requires_env=_rl_env, is_async=True)
registry.register(name="rl_start_training", toolset="rl", schema=RL_START_TRAINING_SCHEMA,
handler=lambda args, **kw: rl_start_training(), check_fn=check_rl_api_keys, requires_env=_rl_env, is_async=True)
registry.register(name="rl_check_status", toolset="rl", schema=RL_CHECK_STATUS_SCHEMA,
handler=lambda args, **kw: rl_check_status(run_id=args.get("run_id", "")), check_fn=check_rl_api_keys, requires_env=_rl_env, is_async=True)
registry.register(name="rl_stop_training", toolset="rl", schema=RL_STOP_TRAINING_SCHEMA,
handler=lambda args, **kw: rl_stop_training(run_id=args.get("run_id", "")), check_fn=check_rl_api_keys, requires_env=_rl_env, is_async=True)
registry.register(name="rl_get_results", toolset="rl", schema=RL_GET_RESULTS_SCHEMA,
handler=lambda args, **kw: rl_get_results(run_id=args.get("run_id", "")), check_fn=check_rl_api_keys, requires_env=_rl_env, is_async=True)
registry.register(name="rl_list_runs", toolset="rl", schema=RL_LIST_RUNS_SCHEMA,
handler=lambda args, **kw: rl_list_runs(), check_fn=check_rl_api_keys, requires_env=_rl_env, is_async=True)
registry.register(name="rl_test_inference", toolset="rl", schema=RL_TEST_INFERENCE_SCHEMA,
handler=lambda args, **kw: rl_test_inference(num_steps=args.get("num_steps", 3), group_size=args.get("group_size", 16), models=args.get("models")),
check_fn=check_rl_api_keys, requires_env=_rl_env, is_async=True)