| { | |
| "dataset.debug": false, | |
| "dataset.git_diff": "", | |
| "dataset.git_sha1": "unknown", | |
| "dataset.manual_sample_ids": [], | |
| "dataset.output_dir": "output", | |
| "dataset.path": "/mnt/wd_ssd/beagle_train_data/datasets/ds_EXAONE-3.5-2.4B-Instruct", | |
| "dataset.run_name": "temp_run", | |
| "dataset.seed": 42, | |
| "dataset_generation.batch_size": 1, | |
| "dataset_generation.debug": false, | |
| "dataset_generation.debug_target": null, | |
| "dataset_generation.ds_prefix": "ds_", | |
| "dataset_generation.git_diff": "", | |
| "dataset_generation.git_sha1": "unknown", | |
| "dataset_generation.max_length": 4096, | |
| "dataset_generation.output_dir": "output", | |
| "dataset_generation.run_name": "temp_run", | |
| "dataset_generation.save_every": 1000, | |
| "dataset_generation.seed": 42, | |
| "inference.debug": false, | |
| "inference.disable_timer": true, | |
| "inference.draft_tree_shape": "mc_sim_7b_65", | |
| "inference.git_diff": "", | |
| "inference.git_sha1": "unknown", | |
| "inference.max_new_tokens": 512, | |
| "inference.mode": "speculative", | |
| "inference.output_dir": "output", | |
| "inference.run_name": "temp_run", | |
| "inference.seed": 42, | |
| "modeling.add_noise": true, | |
| "modeling.attention_offset": "random.randrange(10)", | |
| "modeling.attention_wind": "2", | |
| "modeling.ckpt_path": null, | |
| "modeling.debug": false, | |
| "modeling.decoder_key_remap": {}, | |
| "modeling.dtype": "torch.bfloat16", | |
| "modeling.frozen_targets": [], | |
| "modeling.git_diff": "", | |
| "modeling.git_sha1": "unknown", | |
| "modeling.layer_path": "transformer.h", | |
| "modeling.lmhead_path": "lm_head", | |
| "modeling.model_path": "beagle/models/exaone", | |
| "modeling.norm_path": "transformer.ln_f", | |
| "modeling.output_dir": "output", | |
| "modeling.reuse_layer": -1, | |
| "modeling.rotary_path": "transformer.rotary", | |
| "modeling.run_name": "temp_run", | |
| "modeling.seed": 42, | |
| "modeling.tokenizer_path": "LGAI-EXAONE/EXAONE-3.5-2.4B-Instruct", | |
| "modeling.use_fc_eagle": false, | |
| "modeling.use_lower_layers": 5, | |
| "modeling.use_state_distill": false, | |
| "training.adam_beta2": 0.95, | |
| "training.bf16": true, | |
| "training.ddp_find_unused_parameters": false, | |
| "training.debug": false, | |
| "training.filter_out_shorts": true, | |
| "training.git_diff": "diff --git a/beagle/test.py b/beagle/test.py\nindex d15596f..36ede0e 100644\n--- a/beagle/test.py\n+++ b/beagle/test.py\n@@ -1023,6 +1023,7 @@ def test_beagle(max_new_tokens=512,\n #LOWER_LAYERS = 5\n decoder_path = 'w32zhong/wild-blaze-41500'\n #decoder_path = 'w32zhong/happy-yogurt-28000'\n+ #decoder_path = 'w32zhong/ancient-haze-finished'\n LOWER_LAYERS = 1\n \n if simulate_training:", | |
| "training.git_sha1": "696157960d1844b21c1411e0c1f33ca776c85c67", | |
| "training.gradient_accumulation_steps": 4, | |
| "training.learning_rate": 3e-05, | |
| "training.logging_steps": 1, | |
| "training.lr_scheduler_type": "constant_with_warmup", | |
| "training.max_grad_norm": 0.5, | |
| "training.max_length": 2048, | |
| "training.max_steps": -1, | |
| "training.num_train_epochs": 10, | |
| "training.output_dir": "output", | |
| "training.overwrite_output_dir": true, | |
| "training.per_device_train_batch_size": 2, | |
| "training.project": "debug", | |
| "training.report_to": "wandb", | |
| "training.resume_from_checkpoint": true, | |
| "training.resume_wandb_runid": "wcmht3d3", | |
| "training.run_name": "jumping-voice-26", | |
| "training.save_steps": 500, | |
| "training.save_total_limit": 2, | |
| "training.seed": 42, | |
| "training.warmup_steps": 50 | |
| } |