Upload train_r2egym_14B_agent.yaml

train_r2egym_14B_agent.yaml

@@ -0,0 +1,41 @@
+### model
+model_name_or_path: Qwen/Qwen2.5-Coder-14B-instruct
+trust_remote_code: true
+
+### method
+stage: sft
+do_train: true
+finetuning_type: full
+deepspeed: examples/deepspeed/ds_z3_offload_config.json
+
+### dataset
+dataset: R2E-Gym/R2EGym-SFT-Trajectories
+template: qwen
+cutoff_len: 13000
+max_samples: 100000
+overwrite_cache: true
+preprocessing_num_workers: 16
+
+### output
+output_dir: saves/R2EGym-14B-Agent
+logging_steps: 10
+save_steps: 10000
+plot_loss: true
+overwrite_output_dir: false
+
+### train
+flash_attn: fa2
+enable_liger_kernel: true
+use_unsloth_gc: true
+per_device_train_batch_size: 1
+gradient_accumulation_steps: 1
+learning_rate: 1.0e-5
+num_train_epochs: 2.0
+lr_scheduler_type: cosine
+warmup_ratio: 0.1
+bf16: true
+ddp_timeout: 180000000
+
+### wandb
+report_to: none
+run_name: R2EGym-14B-Agent