Solstice-1 / provenance /pretrained_base_metadata.json
wzebrowski's picture
Prepare Solstice-1 175M private release package
7bcd331 verified
Raw History Blame Contribute Delete
7.5 kB
{
"best_val_loss": 2.8356336172670127,
"ema": 2.458311012190913,
"initialization": "normal-0.02-residual-scaled-v1",
"lr_scheduler": {
"bad_evals": 7,
"best": 2.914223026111722,
"cap": 9.830400000000001e-05,
"config_fingerprint": "ba037a7d13990f03806dfb7dab1a1107b9fccd2373da86870deebf3457f2caa1",
"cooldown_remaining": 0,
"evaluations": 100,
"format": "sml-v2-plateau-v1",
"last_loss": 2.9307962145656345,
"last_step": 119019,
"last_tokens": 15600058368,
"reductions": 5
},
"recipe": {
"batches": [
7,
3,
2,
2,
2
],
"betas": [
0.9,
0.95
],
"cache_gib_by_mac": [
24,
24,
8,
8,
8
],
"code_data_transition": {
"format": "sml-prose-python-sql-phase-v1",
"manifest": "42ba571cb16dccbf79e65781e82a21881c78c244bb16fac46c40a6e0f123b1e2",
"parent_contract": {
"code": "b7ff0ae2c1f095bd36300267f5dc2fd60d58be71d828a1b85a796ddf176e942e",
"data": "4412630e242b60857c2ad4a15afb494338ce9686247e6a9e073867d378f53237",
"recipe": "4f6def65dbd6e1f236be703c2ab977928a55f6bb144ef33688a6e3b0488aacb3",
"tokenizer": "1194292c6d906ac19e6f01c6ad2b42825143fad426a22aa8856d903776701ecb"
},
"parent_replica_hash": "051bad7730581c758ce7f469b0214d56cf2e8a9b2df6791c1066b1290881361f",
"parent_step": 280771,
"start_tokens": 36801216512,
"validation_fingerprint": "2e9444ddc7de3cf4685911d967c1ad605d9b8cfd9244f7039621e828bf088ffa"
},
"data_mode": "hf_stream",
"decay_start_fraction": 0.8,
"eps": 1e-08,
"eval_batch_size": 2,
"eval_batches_per_source": 32,
"eval_every_tokens": 100000000,
"grad_accum": 4,
"grad_clip": 1.0,
"gradual_decay_transition": {
"decay_end_tokens": 100000000000,
"floor_lr": 3e-05,
"format": "sml-v2-open-ended-gradual-v1",
"parent_contract": {
"code": "1bc7d03c972319212ea775d4667fc0d3f5c0d2d738f3bfff4967c0093913df00",
"data": "4412630e242b60857c2ad4a15afb494338ce9686247e6a9e073867d378f53237",
"recipe": "19905de3f7b0d4309d2a3341ad393ebb7934e8ef53fe7822afc97f5197acdfff",
"tokenizer": "1194292c6d906ac19e6f01c6ad2b42825143fad426a22aa8856d903776701ecb"
},
"parent_replica_hash": "f6238b6fd8a3447dd4cf669f83ab0705bc352e8b5b58bc5b46533246f97fd5f4",
"parent_step": 129579,
"shape": "linear",
"start_lr": 9.830400000000001e-05,
"start_tokens": 16984178688,
"validation_fingerprint": "2e9444ddc7de3cf4685911d967c1ad605d9b8cfd9244f7039621e828bf088ffa"
},
"keep_checkpoints": 3,
"log_every": 10,
"min_lr_ratio": 0.1,
"model": {
"attention_impl": "fast",
"bias": false,
"ce_impl": "metal",
"d_model": 576,
"ffn_impl": "packed-metal",
"loss_dtype": "float32",
"max_seq_len": 2048,
"mlp_multiple_of": 256,
"mlp_ratio": 4.0,
"n_heads": 9,
"n_kv_heads": 3,
"n_layers": 44,
"qk_norm": true,
"vocab_size": 32000
},
"name": "opensml-deep44-174m-fresh-prose-stream-20b-20261022",
"peak_lr": 0.0003,
"plateau_scheduler": {
"cooldown": 0,
"factor": 0.8,
"format": "sml-v2-plateau-v1",
"initial_lr": 0.0003,
"min_delta": 0.0,
"min_lr": 3e-05,
"patience": 10
},
"playground_publish_best": false,
"reserve_gib": 16,
"sample_every_tokens": 100000000,
"sample_prompts_file": "sample_prompts.json",
"save_every_tokens": 100000000,
"scheduler_transition": {
"baseline_best": 3.1120742715895178,
"format": "sml-v2-scheduler-transition-v1",
"parent_contract": {
"code": "680fdd3a84c6cae314e9d9038ca00e1f2712284e255d5d8d66599984f57ebb6c",
"data": "2f863c538203e9f286b549c7f56598fb9d9b51cef81af77f9627ba7a405e5ec7",
"recipe": "89d8b84a2fe6ff50ef274ccb70b01e4d5394546ea8371876dccf032f817a2f14",
"tokenizer": "1194292c6d906ac19e6f01c6ad2b42825143fad426a22aa8856d903776701ecb"
},
"parent_replica_hash": "52bfaaef0938ea869a6ab16890e6d311df863e56156eafcab58a7e2a694d9ec4",
"parent_step": 43145,
"start_tokens": 5655101440
},
"seed": 7337,
"stable_cooldown_transition": {
"duration_tokens": 2000027648,
"final_lr": 1e-05,
"format": "sml-v2-stable-cooldown-v1",
"parent_contract": {
"code": "ae36174cc1f5b20f6bf7cf4bc56c7c5f59136dd22104d3dff775d3628600271f",
"data": "42ba571cb16dccbf79e65781e82a21881c78c244bb16fac46c40a6e0f123b1e2",
"recipe": "d25763cd3f2a452c0cbcb4adec3503644d5bbdccea1deab56412a4843c3d8ec4",
"tokenizer": "1194292c6d906ac19e6f01c6ad2b42825143fad426a22aa8856d903776701ecb"
},
"parent_replica_hash": "b6d6568e89215fb219747c3426536ac21e9bcebb9a97bc264e3481d091e37641",
"parent_step": 305186,
"shape": "1-sqrt",
"start_lr": 7.93658732685024e-05,
"start_tokens": 40001339392,
"validation_fingerprint": "2e9444ddc7de3cf4685911d967c1ad605d9b8cfd9244f7039621e828bf088ffa"
},
"stable_schedule_transition": {
"format": "sml-v2-open-ended-stable-v1",
"parent_contract": {
"code": "680fdd3a84c6cae314e9d9038ca00e1f2712284e255d5d8d66599984f57ebb6c",
"data": "2f863c538203e9f286b549c7f56598fb9d9b51cef81af77f9627ba7a405e5ec7",
"recipe": "99caa9bb0f8182f591c4b466b09a6c9aad4b2ff27ac9314bc44fe1ca821cd9ca",
"tokenizer": "1194292c6d906ac19e6f01c6ad2b42825143fad426a22aa8856d903776701ecb"
},
"parent_replica_hash": "b86edb7b0448bc58c6f8e1266ba194eebe4442fc860d3ba888901671ef04663e",
"parent_scheduler": {
"bad_evals": 7,
"best": 2.914223026111722,
"cap": 9.830400000000001e-05,
"config_fingerprint": "ba037a7d13990f03806dfb7dab1a1107b9fccd2373da86870deebf3457f2caa1",
"cooldown_remaining": 0,
"evaluations": 100,
"format": "sml-v2-plateau-v1",
"last_loss": 2.9307962145656345,
"last_step": 119019,
"last_tokens": 15600058368,
"reductions": 5
},
"parent_step": 119722,
"parent_stream": {
"dedup_window": 100000,
"prefetch_batches": 4,
"read_attempts": 5,
"shuffle_buffer": 256,
"tokenizer_corpus_policy": "reuse-frozen"
},
"stable_lr": 9.830400000000001e-05,
"start_lr": 9.830400000000001e-05,
"start_tokens": 15692201984,
"validation_fingerprint": "2e9444ddc7de3cf4685911d967c1ad605d9b8cfd9244f7039621e828bf088ffa",
"warmup_tokens": 0
},
"stream": {
"dedup_window": 100000,
"prefetch_batches": 4,
"read_attempts": 5,
"shuffle_buffer": 256,
"source_policy": "all-available-v1",
"tokenizer_corpus_policy": "reuse-frozen"
},
"target_tokens": 20000000000,
"warmup_fraction": 0.01,
"weight_decay": 0.1
},
"replica_hash": "df01e3e57d4ff8045fff3eb346be964e64cb9f45597aa7988eeb3b962613f3e0",
"step": 320445,
"tokens": 42001367040,
"v2_contract": {
"code": "06f0ae28bc1536e2cf32de5ca05d05f3ab59b165f07034b1f6a449742f940e33",
"data": "42ba571cb16dccbf79e65781e82a21881c78c244bb16fac46c40a6e0f123b1e2",
"recipe": "4fdbfa8c03e80dd2911def5694d87ae3f91b4c79178960e006bbc25c46b13730",
"tokenizer": "1194292c6d906ac19e6f01c6ad2b42825143fad426a22aa8856d903776701ecb"
},
"validation_fingerprint": "2e9444ddc7de3cf4685911d967c1ad605d9b8cfd9244f7039621e828bf088ffa"
}