[TRACKING] feat: Integrate SwanLab for experiment tracking with online/offline mode and local dashboard support

README.md

@@ -45,7 +45,7 @@ verl is fast with:
   - Support model-based reward and function-based reward (verifiable reward)
 - flash-attention, [sequence packing](examples/ppo_trainer/run_qwen2-7b_seq_balance.sh), [long context](examples/ppo_trainer/run_deepseek7b_llm_sp2.sh) support via DeepSpeed Ulysses, [LoRA](examples/sft/gsm8k/run_qwen_05_peft.sh), [Liger-kernel](examples/sft/gsm8k/run_qwen_05_sp2_liger.sh)
 - scales up to 70B models and hundreds of GPUs
-- experiment tracking with wandb and mlflow
+- experiment tracking with wandb, swanlab and mlflow
 
 ## Upcoming Features
 - Reward model training

verl/utils/tracking.py

@@ -22,7 +22,7 @@
 
 
 class Tracking(object):
-    supported_backend = ['wandb', 'mlflow', 'console']
+    supported_backend = ["wandb", "mlflow", "swanlab", "console"]
 
     def __init__(self, project_name, experiment_name, default_backend: Union[str, List[str]] = 'console', config=None):
         if isinstance(default_backend, str):
@@ -47,6 +47,22 @@ def __init__(self, project_name, experiment_name, default_backend: Union[str, Li
             mlflow.log_params(_compute_mlflow_params_from_objects(config))
             self.logger['mlflow'] = _MlflowLoggingAdapter()
 
+        if "swanlab" in default_backend:
+            import swanlab
+            import os
+
+            SWANLAB_API_KEY = os.environ.get("SWANLAB_API_KEY", None)
+            SWANLAB_LOG_DIR = os.environ.get("SWANLAB_LOG_DIR", "swanlog")
+            SWANLAB_MODE = os.environ.get("SWANLAB_MODE", "cloud")
+            if SWANLAB_API_KEY:
+                swanlab.login(SWANLAB_API_KEY)  # NOTE: previous login information will be overwritten
+            swanlab.init(project=project_name,
+                         experiment_name=experiment_name,
+                         config=config,
+                         logdir=SWANLAB_LOG_DIR,
+                         mode=SWANLAB_MODE)
+            self.logger["swanlab"] = swanlab
+
         if 'console' in default_backend:
             from verl.utils.logger.aggregate_logger import LocalLogger
             self.console_logger = LocalLogger(print_to_console=True)
@@ -60,6 +76,8 @@ def log(self, data, step, backend=None):
     def __del__(self):
         if 'wandb' in self.logger:
             self.logger['wandb'].finish(exit_code=0)
+        if 'swanlab' in self.logger:
+            self.logger['swanlab'].finish()
 
 
 class _MlflowLoggingAdapter: