BananaMind-2.1-MiniCoder / checkpoint_metadata.json
Banaxi-Tech's picture
Save 15% checkpoint at 4,500,488,192 tokens
bb2934d verified
Raw
History Blame Contribute Delete
2.37 kB
{
"variant": "minicoder",
"parameters": 24949999,
"transformer_parameters": 19950029,
"ngram_parameters": 4999970,
"architecture": {
"variant": "minicoder",
"vocab_size": 8192,
"hidden_size": 384,
"intermediate_size": 456,
"num_hidden_layers": 13,
"num_attention_heads": 12,
"num_key_value_heads": 1,
"head_dim": 32,
"ngram_buckets": 51699,
"ngram_rank_per_hash": 48,
"loop_schedule": [
0,
1,
2,
3,
4,
5,
6,
7,
8,
4,
5,
6,
7,
8,
9,
10,
11,
12
],
"ngram_injection_schedule_positions": [
4,
9
],
"refresh_kernel_size": 9,
"max_position_embeddings": 4096,
"rope_theta": 100000.0,
"rms_norm_eps": 1e-05,
"initializer_range": 0.02,
"use_cache": false,
"transformers_version": "5.16.1",
"architectures": null,
"output_hidden_states": false,
"return_dict": true,
"dtype": null,
"chunk_size_feed_forward": 0,
"is_encoder_decoder": false,
"id2label": {
"0": "LABEL_0",
"1": "LABEL_1"
},
"label2id": {
"LABEL_0": 0,
"LABEL_1": 1
},
"problem_type": null,
"_name_or_path": "",
"tie_word_embeddings": true,
"bos_token_id": 1,
"eos_token_id": 2,
"pad_token_id": 0,
"unk_token_id": 3,
"model_type": "bananamind21_coder",
"output_attentions": false
},
"training_percent": 15,
"step": 8584,
"total_steps": 57221,
"tokens_seen": 4500488192,
"target_tokens": 30000000000,
"tokens_per_full_step": 524288,
"final_step_supervised_tokens": 240640,
"world_size": 4,
"local_batch": 32,
"global_batch_sequences": 128,
"gpu_name": "NVIDIA H200",
"architecture_revision": "330e6d42645239a75d24104a69cd4283b9be3539",
"dataset_revisions": {
"stack_v3": "df4b205fbba4cc1c2fd1f205b10d66f730798bb9",
"fineweb_edu": "87f09149ef4734204d70ed1d046ddc9ca3f2b8f9"
},
"target_source_shares": {
"stack_v3": 0.75,
"fineweb_edu": 0.25
},
"target_source_tokens": {
"stack_v3": 22500000000,
"fineweb_edu": 7500000000
},
"actual_source_tokens": {
"stack_v3": 3375366144,
"fineweb_edu": 1125122048
},
"muon_peak_lr": 0.02,
"adamw_peak_lr": 0.002,
"ngram_peak_lr": 0.001,
"elapsed_seconds_this_job": 4361.163550853729
}