api: rename tracking logger to wandb logger type

docs/examples/config.rst

@@ -307,7 +307,7 @@ Trainer
      total_epochs: 30
      project_name: verl_examples
      experiment_name: gsm8k
-     logger: ['console', 'tracking']
+     logger: ['console', 'wandb']
      nnodes: 1
      n_gpus_per_node: 8
      save_freq: -1
@@ -319,8 +319,7 @@ Trainer
 - ``trainer.total_epochs``: Number of epochs in training.
 - ``trainer.project_name``: For wandb
 - ``trainer.experiment_name``: For wandb
-- ``trainer.logger``: Support console and tracking. For tracking, we
-  will initialize a wandb
+- ``trainer.logger``: Support console and wandb
 - ``trainer.nnodes``: Number of nodes used in the training.
 - ``trainer.n_gpus_per_node``: Number of GPUs per node.
 - ``trainer.save_freq``: The frequency (by iteration) to save checkpoint

docs/examples/gsm8k_example.rst

@@ -91,7 +91,7 @@ We also provide various training scripts for SFT on GSM8K dataset in `gsm8k sft
        trainer.project_name=gsm8k-sft \
        trainer.experiment_name=gsm8k-sft-deepseek-coder-6.7b-instruct \
        trainer.total_epochs=4 \
-       trainer.logger=['console','tracking']
+       trainer.logger=['console','wandb']
 
 Step 4: Perform PPO training with your model on GSM8K Dataset
 -------------------------------------------------------------
@@ -156,7 +156,7 @@ The script of run_deepseek7b_llm.sh
        critic.model.fsdp_config.optimizer_offload=False \
        algorithm.kl_ctrl.kl_coef=0.001 \
        trainer.critic_warmup=0 \
-       trainer.logger=['console','tracking'] \
+       trainer.logger=['console','wandb'] \
        trainer.project_name='verl_example_gsm8k' \
        trainer.experiment_name='deepseek_llm_7b_function_rm' \
        trainer.n_gpus_per_node=8 \

docs/start/quickstart.rst

@@ -97,7 +97,7 @@ We also provide various training scripts for SFT on GSM8K dataset in `gsm8k sft
        trainer.project_name=gsm8k-sft \
        trainer.experiment_name=gsm8k-sft-deepseek-coder-6.7b-instruct \
        trainer.total_epochs=4 \
-       trainer.logger=['console','tracking']
+       trainer.logger=['console','wandb']
 
 Step 4: Perform PPO training with your model on GSM8K Dataset
 -------------------------------------------------------------
@@ -163,7 +163,7 @@ The script of `run_deepseek7b_llm.sh`
        critic.model.fsdp_config.optimizer_offload=False \
        algorithm.kl_ctrl.kl_coef=0.001 \
        trainer.critic_warmup=0 \
-       trainer.logger=['console','tracking'] \
+       trainer.logger=['console','wandb'] \
        trainer.project_name='verl_example_gsm8k' \
        trainer.experiment_name='deepseek_llm_7b_function_rm' \
        trainer.n_gpus_per_node=8 \

examples/ppo_trainer/run_deepseek7b_llm.sh

@@ -29,7 +29,7 @@ python3 -m verl.trainer.main_ppo \
     critic.model.fsdp_config.optimizer_offload=False \
     algorithm.kl_ctrl.kl_coef=0.001 \
     trainer.critic_warmup=0 \
-    trainer.logger=['console','tracking'] \
+    trainer.logger=['console','wandb'] \
     trainer.project_name='verl_example_gsm8k' \
     trainer.experiment_name='deepseek_llm_7b_function_rm' \
     trainer.n_gpus_per_node=8 \

examples/ppo_trainer/run_deepseek_full_hh_rlhf.sh

@@ -31,7 +31,7 @@ python3 -m verl.trainer.main_ppo --config-path=./config --config-name='ppo_megat
     reward_model.param_offload=False \
     algorithm.kl_ctrl.kl_coef=0.001 \
     trainer.critic_warmup=0 \
-    trainer.logger=['console','tracking'] \
+    trainer.logger=['console','wandb'] \
     trainer.project_name='verl_megatron_full_hh_rlhf_examples' \
     trainer.experiment_name='deepseek_llm_7b_model_rm' \
     trainer.n_gpus_per_node=8 \

examples/ppo_trainer/run_deepseek_math_gsm8k_megatron.sh

@@ -30,7 +30,7 @@ python3 -m verl.trainer.main_ppo --config-path=./config --config-name='ppo_megat
     critic.ppo_micro_batch_size=32 \
     algorithm.kl_ctrl.kl_coef=0.001 \
     trainer.critic_warmup=0 \
-    trainer.logger=['console','tracking'] \
+    trainer.logger=['console','wandb'] \
     trainer.project_name='verl_megatron_math_gsm8k_examples' \
     trainer.experiment_name='deepseek_llm_7b_function_rm' \
     trainer.n_gpus_per_node=8 \

examples/ppo_trainer/run_deepseek_megatron.sh

@@ -22,7 +22,7 @@ python3 -m verl.trainer.main_ppo --config-path=./config --config-name='ppo_megat
     critic.ppo_micro_batch_size=64 \
     algorithm.kl_ctrl.kl_coef=0.001 \
     trainer.critic_warmup=0 \
-    trainer.logger=['console','tracking'] \
+    trainer.logger=['console','wandb'] \
     trainer.project_name='verl_megatron_gsm8k_examples' \
     trainer.experiment_name='deepseek_llm_7b_function_rm' \
     trainer.n_gpus_per_node=8 \

examples/ppo_trainer/run_gemma.sh

@@ -29,7 +29,7 @@ python3 -m verl.trainer.main_ppo \
     critic.model.fsdp_config.optimizer_offload=False \
     algorithm.kl_ctrl.kl_coef=0.001 \
     trainer.critic_warmup=0 \
-    trainer.logger=['console','tracking'] \
+    trainer.logger=['console','wandb'] \
     trainer.project_name='verl_example' \
     trainer.experiment_name='gemma2b_function_rm' \
     trainer.n_gpus_per_node=2 \

examples/ppo_trainer/run_qwen2-7b.sh

@@ -37,7 +37,7 @@ python3 -m verl.trainer.main_ppo \
     critic.model.fsdp_config.optimizer_offload=False \
     algorithm.kl_ctrl.kl_coef=0.001 \
     trainer.critic_warmup=0 \
-    trainer.logger=['console','tracking'] \
+    trainer.logger=['console','wandb'] \
     trainer.project_name='verl_example' \
     trainer.experiment_name='Qwen2-7B-Instruct_function_rm' \
     trainer.n_gpus_per_node=8 \

examples/ppo_trainer/run_qwen2-7b_rm.sh

@@ -44,7 +44,7 @@ python3 -m verl.trainer.main_ppo \
     reward_model.micro_batch_size=16 \
     algorithm.kl_ctrl.kl_coef=0.001 \
     trainer.critic_warmup=0 \
-    trainer.logger=['console','tracking'] \
+    trainer.logger=['console','wandb'] \
     trainer.project_name='verl_example' \
     trainer.experiment_name='Qwen2-7B-Instruct_hybrid_rm' \
     trainer.n_gpus_per_node=8 \

examples/ppo_trainer/run_qwen2.5-32b.sh

@@ -38,7 +38,7 @@ python3 -m verl.trainer.main_ppo \
     critic.model.fsdp_config.optimizer_offload=False \
     algorithm.kl_ctrl.kl_coef=0.0001 \
     trainer.critic_warmup=0 \
-    trainer.logger=['console','tracking'] \
+    trainer.logger=['console','wandb'] \
     trainer.project_name='verl_example' \
     trainer.experiment_name='Qwen2.5-32B-Instruct_function_rm' \
     trainer.n_gpus_per_node=8 \

examples/sft/gsm8k/run_deepseek_6b7.sh

@@ -16,4 +16,4 @@ torchrun --standalone --nnodes=1 --nproc_per_node=$nproc_per_node \
     trainer.project_name=gsm8k-sft \
     trainer.experiment_name=gsm8k-sft-deepseek-coder-6.7b-instruct \
     trainer.total_epochs=4 \
-    trainer.logger=['console','tracking']
+    trainer.logger=['console','wandb']

examples/sft/gsm8k/run_gemma_2b.sh

@@ -2,9 +2,16 @@
 
 set -x
 
-hdfs_path=hdfs://user/verl/experiments/gsm8k/gemma-2b-it/ # replace to your own hdfs/local path
+if [ "$#" -lt 2 ]; then
+    echo "Usage: run_gemma_2b.sh <nproc_per_node> <save_path> [other_configs...]"
+    exit 1
+fi
 
 nproc_per_node=$1
+hdfs_path=$2
+
+# Shift the arguments so $@ refers to the rest
+shift 2
 
 torchrun --standalone --nnodes=1 --nproc_per_node=$nproc_per_node \
      -m verl.trainer.fsdp_sft_trainer \
@@ -18,4 +25,4 @@ torchrun --standalone --nnodes=1 --nproc_per_node=$nproc_per_node \
     trainer.project_name=gsm8k-sft \
     trainer.experiment_name=gsm8k-sft-gemma-2b-it \
     trainer.total_epochs=3 \
-    trainer.logger=['console','tracking']
+    trainer.logger=['console','wandb'] $@

examples/sft/gsm8k/run_gemma_7b.sh

@@ -16,4 +16,4 @@ torchrun --standalone --nnodes=1 --nproc_per_node=$nproc_per_node \
     trainer.project_name=gsm8k-sft \
     trainer.experiment_name=gsm8k-sft-gemma-1.1-7b-it \
     trainer.total_epochs=4 \
-    trainer.logger=['console','tracking']
+    trainer.logger=['console','wandb']

examples/split_placement/config/ppo_trainer_split.yaml

@@ -121,7 +121,7 @@ trainer:
   total_epochs: 30
   project_name: verl_examples
   experiment_name: gsm8k
-  logger: ['console', 'tracking']
+  logger: ['console', 'wandb']
   nnodes: 1
   n_gpus_per_node: 8
   save_freq: -1

examples/split_placement/run_deepseek7b_llm.sh

@@ -29,7 +29,7 @@ python3 main_ppo_split.py \
     critic.model.fsdp_config.optimizer_offload=False \
     algorithm.kl_ctrl.kl_coef=0.001 \
     trainer.critic_warmup=0 \
-    trainer.logger=['console','tracking'] \
+    trainer.logger=['console','wandb'] \
     trainer.project_name='verl_example_gsm8k' \
     trainer.experiment_name='deepseek_llm_7b_function_rm' \
     trainer.n_gpus_per_node=8 \

verl/trainer/config/ppo_megatron_trainer.yaml

@@ -135,7 +135,7 @@ trainer:
   total_epochs: 30
   project_name: verl_examples
   experiment_name: gsm8k
-  logger: ['console', 'tracking']
+  logger: ['console', 'wandb']
   nnodes: 1
   n_gpus_per_node: 8
   save_freq: -1

verl/trainer/config/ppo_trainer.yaml

@@ -121,7 +121,7 @@ trainer:
   total_epochs: 30
   project_name: verl_examples
   experiment_name: gsm8k
-  logger: ['console', 'tracking']
+  logger: ['console', 'wandb']
   nnodes: 1
   n_gpus_per_node: 8
   save_freq: -1

verl/utils/tracking.py

@@ -19,20 +19,24 @@
 
 
 class Tracking(object):
-    supported_backend = ['tracking', 'console']
+    supported_backend = ['wandb', 'console']
 
     def __init__(self, project_name, experiment_name, default_backend: Union[str, List[str]] = 'console', config=None):
         if isinstance(default_backend, str):
             default_backend = [default_backend]
         for backend in default_backend:
-            assert backend in self.supported_backend, f'{backend} is not supported'
+            if backend == 'tracking':
+                import warnings
+                warnings.warn("`tracking` logger is deprecated. use `wandb` instead.", DeprecationWarning)
+            else:
+                assert backend in self.supported_backend, f'{backend} is not supported'
 
         self.logger = {}
 
-        if 'tracking' in default_backend:
+        if 'tracking' in default_backend or 'wandb' in default_backend:
             import wandb
             wandb.init(project=project_name, name=experiment_name, config=config)
-            self.logger['tracking'] = wandb
+            self.logger['wandb'] = wandb
 
         if 'console' in default_backend:
             from verl.utils.logger.aggregate_logger import LocalLogger