aklein4's picture
Upload checkpoint at step 950
0ed6663 verified
Raw History Blame Contribute Delete
3.62 kB
{
"git_commit": "a34fdf8bdceb4b644a9106819b9d6f7d6d0020e0",
"wandb_run_id": "fnn7p0ul",
"config": {
"seed": 42,
"notes": null,
"trainer": {
"run_dir": "local_data/runs/baseline-fixed",
"checkpoint": {
"load_state": false,
"save_optimizer": false,
"interval": 50
},
"wandb": {
"project": "oloop-instruct",
"enabled": true
},
"huggingface": {
"private": false,
"enabled": true
},
"global_batch_size": 128,
"max_steps": 1000000,
"num_logit_iterations": 4,
"aux_loss_weight": 0.1,
"max_grad_norm": 1.0,
"debug": false,
"groups": {
"slow": {
"_target_": "optax.adamw",
"lr": 3e-05,
"b1": 0.9,
"b2": 0.95,
"eps": 1e-06,
"weight_decay": 0.0,
"warmup_steps": 100,
"stable_steps": 899900,
"decay_steps": 100000,
"min_lr_ratio": 0.0
}
},
"_target_": "trainers.horizon_lm_trainer.HorizonLMTrainer"
},
"data": {
"skip_batches": 0,
"dataset": {
"path": "aklein4/300k-horizons",
"split": "train",
"streaming": true
},
"collator": {
"_target_": "collators.horizon.HorizonCollator",
"tokenizer_url": "meta-llama/Llama-3.2-1B-Instruct",
"max_length": 1024,
"cluster_length": 64
}
},
"distributed": {
"initialize": "auto"
},
"mesh": {
"fsdp_axis_size": -1
},
"name": "baseline-fixed",
"initialization": "meta-llama/Llama-3.2-1B-Instruct",
"initialization_step": null,
"model": {
"sharding": {
"default": [],
"rules": [
{
"pattern": "^embed_tokens/embedding$",
"spec": [
"fsdp",
null
]
},
{
"pattern": "layers/mlp/(gate_proj|up_proj|down_proj)",
"spec": [
null,
null,
"fsdp"
]
}
]
},
"_target_": "models.llama.LlamaForCausalLM",
"vocab_size": 128256,
"pad_token_id": -100,
"hidden_size": 2048,
"intermediate_size": 8192,
"num_hidden_layers": 16,
"num_attention_heads": 32,
"num_key_value_heads": 8,
"head_dim": 64,
"max_position_embeddings": 131072,
"rope": {
"theta": 500000.0,
"factor": 32.0,
"high_freq_factor": 4.0,
"low_freq_factor": 1.0,
"original_context_len": 8192
},
"rms_norm_eps": 1e-05,
"compute_dtype": "bfloat16",
"gradient_checkpointing": true,
"attention_backend": "auto"
}
},
"step": 950,
"checkpoint_items": [
"params"
]
}