trinity.explorer.explorer module

trinity.explorer.explorer module#

The explorer module

class trinity.explorer.explorer.Explorer(config: Config)[源代码]#

基类：object

Responsible for exploring the taskset.

__init__(config: Config)[源代码]#
async setup_weight_sync_group(master_address: str, master_port: int, state_dict_meta: List = None)[源代码]#
async setup_model_level_weight_sync_group()[源代码]#

Setup process group for each model, only used in serve mode.

async prepare() → None[源代码]#

Preparation before running.

async get_weight(name: str) → Tensor[源代码]#

Get the weight of the loaded model (For checkpoint weights update).

async explore() → str[源代码]#
The timeline of the exploration process:
<--------------------------------- one period -------------------------------------> |
explorer | <---------------- step_1 --------------> | |
| <---------------- step_2 --------------> | |
... |
| <---------------- step_n ---------------> | |
| <---------------------- eval --------------------> | <-- sync --> |

|--------------------------------------------------------------------------------------|

trainer | <-- idle --> | <-- step_1 --> | <-- step_2 --> | ... | <-- step_n --> | <-- sync --> |

async explore_step() → bool[源代码]#
async need_sync() → bool[源代码]#
need_eval() → bool[源代码]#
async eval()[源代码]#

Evaluation on all evaluation data samples.

async benchmark() → bool[源代码]#

Benchmark the model checkpoints.

async save_checkpoint(sync_weight: bool = False) → None[源代码]#
async sync_weight() → None[源代码]#

Synchronize model weights.

async shutdown() → None[源代码]#
async is_alive() → bool[源代码]#

Check if the explorer is alive.

async serve() → None[源代码]#

Run the explorer in serving mode.

In serving mode, the explorer starts an OpenAI compatible server to handle requests. Agent applications can be deployed separately and interact with the explorer via the API.

import openai


client = openai.OpenAI(
    base_url=f"{explorer_server_url}/v1",
    api_key="EMPTY",
)
response = client.chat.completions.create(
    model=config.model.model_path,
    messages=[{"role": "user", "content": "Hello!"}]
)
classmethod get_actor(config: Config)[源代码]#

Get a Ray actor for the explorer.

目录