| mode: both |
| project: Trinity-RFT-context-length-exp |
| group: length-test |
| name: length-test |
| checkpoint_root_dir: ${oc.env:TRINITY_CHECKPOINT_ROOT_DIR,./checkpoints/length-test} |
| continue_from_checkpoint: false |
| algorithm: |
| algorithm_type: grpo |
| repeat_times: ${oc.env:REPEAT_TIMES,8} |
| advantage_fn: grpo |
| sample_strategy: default |
| policy_loss_fn: ppo |
| kl_penalty_fn: none |
| kl_loss_fn: k2 |
| entropy_loss_fn: default |
| optimizer: |
| lr: 1.0e-05 |
| lr_warmup_steps_ratio: 0.0 |
| lr_scheduler_type: constant |
| data_processor: {} |
| model: |
| model_path: ${oc.env:MODEL_PATH,Qwen/Qwen3-0.6B} |
| max_prompt_tokens: ${oc.env:PROMPT_LEN,2048} |
| max_model_len: ${oc.env:MAX_MODEL_LEN,4096} |
| rope_scaling: ${oc.decode:${oc.env:ROPE_SCALING,null}} |
| cluster: |
| node_num: 1 |
| gpu_per_node: ${oc.env:GPU_NUM,8} |
| buffer: |
| batch_size: 1 |
| total_steps: 2 |
| explorer_input: |
| taskset: |
| name: taskset |
| storage_type: file |
| path: openai/gsm8k |
| split: train |
| subset_name: main |
| format: |
| prompt_key: question |
| response_key: answer |
| rollout_args: |
| temperature: 1.0 |
| logprobs: 0 |
| workflow_args: |
| prompt_len: ${model.max_prompt_tokens} |
| max_model_len: ${model.max_model_len} |
| eval_tasksets: [] |
| default_workflow_type: scripts.context_length_test.workflow.synthetic_exp_workflow.SyntheticExpWorkflow |
| default_reward_fn_type: math_reward |
| trainer_input: |
| experience_buffer: |
| name: experience_buffer |
| storage_type: queue |
| replay_buffer: |
| enable: false |
| priority_fn: linear_decay |
| reuse_cooldown_time: null |
| priority_fn_args: |
| decay: 2.0 |
| explorer: |
| runner_per_model: 8 |
| rollout_model: |
| engine_num: ${oc.env:ENGINE_NUM,1} |
| tensor_parallel_size: 1 |
| enforce_eager: true |
| enable_prefix_caching: false |
| enable_chunked_prefill: false |
| gpu_memory_utilization: 0.9 |
| dtype: bfloat16 |
| seed: 42 |
| enable_thinking: false |
| enable_history: false |
| enable_openai_api: false |
| enable_auto_tool_choice: false |
| tool_call_parser: null |
| reasoning_parser: null |
| auxiliary_models: [] |
| eval_interval: 1000 |
| trainer: |
| trainer_type: verl |
| trainer_strategy: ${oc.env:TRAINER_STRATEGY,fsdp} |
| save_interval: 100 |
| enable_preview: true |
| grad_clip: 1.0 |
| ulysses_sequence_parallel_size: ${oc.env:SP_NUM,1} |
| save_hf_checkpoint: ${oc.env:SAVE_HF_CHECKPOINT,last} |
| trainer_config: |
| actor_rollout_ref: |
| actor: |
| entropy_from_logits_with_chunking: ${oc.env:ENTROPY_SAVING,false} |
| entropy_checkpointing: ${oc.env:ENTROPY_SAVING,false} |
| fsdp_config: |
| param_offload: ${oc.env:OFFLOAD,false} |
| optimizer_offload: ${oc.env:OFFLOAD,false} |
| offload_policy: ${oc.env:OFFLOAD,false} |
| ref: |
| entropy_from_logits_with_chunking: ${oc.env:ENTROPY_SAVING,false} |
| entropy_checkpointing: ${oc.env:ENTROPY_SAVING,false} |
| fsdp_config: |
| param_offload: ${oc.env:OFFLOAD,false} |
| optimizer_offload: ${oc.env:OFFLOAD,false} |
| offload_policy: ${oc.env:OFFLOAD,false} |
| monitor: |
| monitor_type: tensorboard |
| synchronizer: |
| sync_method: nccl |
| sync_style: fixed |
| sync_interval: 1 |
| sync_timeout: 1200 |
| log: |
| level: INFO |
|
|
