Auto upload 2026-08-09T04:40:09.292092
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- .gitattributes +1 -0
- __pycache__/exp.cpython-311.pyc +0 -0
- out/mlp-s10-94L_run/checkpoint-100/config.json +35 -0
- out/mlp-s10-94L_run/checkpoint-100/model.safetensors +3 -0
- out/mlp-s10-94L_run/checkpoint-100/optimizer.pt +3 -0
- out/mlp-s10-94L_run/checkpoint-100/rng_state.pth +3 -0
- out/mlp-s10-94L_run/checkpoint-100/scheduler.pt +3 -0
- out/mlp-s10-94L_run/checkpoint-100/tokenizer.json +0 -0
- out/mlp-s10-94L_run/checkpoint-100/tokenizer_config.json +13 -0
- out/mlp-s10-94L_run/checkpoint-100/trainer_state.json +85 -0
- out/mlp-s10-94L_run/checkpoint-100/training_args.bin +3 -0
- out/mlp-s10-94L_run/checkpoint-1000/config.json +35 -0
- out/mlp-s10-94L_run/checkpoint-1000/model.safetensors +3 -0
- out/mlp-s10-94L_run/checkpoint-1000/optimizer.pt +3 -0
- out/mlp-s10-94L_run/checkpoint-1000/rng_state.pth +3 -0
- out/mlp-s10-94L_run/checkpoint-1000/scheduler.pt +3 -0
- out/mlp-s10-94L_run/checkpoint-1000/tokenizer.json +0 -0
- out/mlp-s10-94L_run/checkpoint-1000/tokenizer_config.json +13 -0
- out/mlp-s10-94L_run/checkpoint-1000/trainer_state.json +544 -0
- out/mlp-s10-94L_run/checkpoint-1000/training_args.bin +3 -0
- out/mlp-s10-94L_run/checkpoint-200/config.json +35 -0
- out/mlp-s10-94L_run/checkpoint-200/model.safetensors +3 -0
- out/mlp-s10-94L_run/checkpoint-200/optimizer.pt +3 -0
- out/mlp-s10-94L_run/checkpoint-200/rng_state.pth +3 -0
- out/mlp-s10-94L_run/checkpoint-200/scheduler.pt +3 -0
- out/mlp-s10-94L_run/checkpoint-200/tokenizer.json +0 -0
- out/mlp-s10-94L_run/checkpoint-200/tokenizer_config.json +13 -0
- out/mlp-s10-94L_run/checkpoint-200/trainer_state.json +136 -0
- out/mlp-s10-94L_run/checkpoint-200/training_args.bin +3 -0
- out/mlp-s10-94L_run/checkpoint-300/config.json +35 -0
- out/mlp-s10-94L_run/checkpoint-300/model.safetensors +3 -0
- out/mlp-s10-94L_run/checkpoint-300/optimizer.pt +3 -0
- out/mlp-s10-94L_run/checkpoint-300/rng_state.pth +3 -0
- out/mlp-s10-94L_run/checkpoint-300/scheduler.pt +3 -0
- out/mlp-s10-94L_run/checkpoint-300/tokenizer.json +0 -0
- out/mlp-s10-94L_run/checkpoint-300/tokenizer_config.json +13 -0
- out/mlp-s10-94L_run/checkpoint-300/trainer_state.json +187 -0
- out/mlp-s10-94L_run/checkpoint-300/training_args.bin +3 -0
- out/mlp-s10-94L_run/checkpoint-400/config.json +35 -0
- out/mlp-s10-94L_run/checkpoint-400/model.safetensors +3 -0
- out/mlp-s10-94L_run/checkpoint-400/optimizer.pt +3 -0
- out/mlp-s10-94L_run/checkpoint-400/rng_state.pth +3 -0
- out/mlp-s10-94L_run/checkpoint-400/scheduler.pt +3 -0
- out/mlp-s10-94L_run/checkpoint-400/tokenizer.json +0 -0
- out/mlp-s10-94L_run/checkpoint-400/tokenizer_config.json +13 -0
- out/mlp-s10-94L_run/checkpoint-400/trainer_state.json +238 -0
- out/mlp-s10-94L_run/checkpoint-400/training_args.bin +3 -0
- out/mlp-s10-94L_run/checkpoint-500/config.json +35 -0
- out/mlp-s10-94L_run/checkpoint-500/model.safetensors +3 -0
- out/mlp-s10-94L_run/checkpoint-500/optimizer.pt +3 -0
.gitattributes
CHANGED
|
@@ -82,3 +82,4 @@ wandb/run-20260808_233557-lhpx30l9/run-lhpx30l9.wandb filter=lfs diff=lfs merge=
|
|
| 82 |
wandb/run-20260808_235917-9b17953b/run-9b17953b.wandb filter=lfs diff=lfs merge=lfs -text
|
| 83 |
wandb/run-20260809_002232-gduul5qe/run-gduul5qe.wandb filter=lfs diff=lfs merge=lfs -text
|
| 84 |
wandb/run-20260809_004653-7146bnsd/run-7146bnsd.wandb filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
| 82 |
wandb/run-20260808_235917-9b17953b/run-9b17953b.wandb filter=lfs diff=lfs merge=lfs -text
|
| 83 |
wandb/run-20260809_002232-gduul5qe/run-gduul5qe.wandb filter=lfs diff=lfs merge=lfs -text
|
| 84 |
wandb/run-20260809_004653-7146bnsd/run-7146bnsd.wandb filter=lfs diff=lfs merge=lfs -text
|
| 85 |
+
wandb/run-20260809_035819-cvzjg5ej/run-cvzjg5ej.wandb filter=lfs diff=lfs merge=lfs -text
|
__pycache__/exp.cpython-311.pyc
CHANGED
|
Binary files a/__pycache__/exp.cpython-311.pyc and b/__pycache__/exp.cpython-311.pyc differ
|
|
|
out/mlp-s10-94L_run/checkpoint-100/config.json
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "s10",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"TinyLlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 1,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eos_token_id": 2,
|
| 11 |
+
"head_dim": 32,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 128,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 256,
|
| 16 |
+
"max_position_embeddings": 512,
|
| 17 |
+
"mlp_bias": false,
|
| 18 |
+
"mlp_type": "mlp",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 94,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"pretraining_tp": 1,
|
| 25 |
+
"rms_norm_eps": 1e-06,
|
| 26 |
+
"rope_parameters": {
|
| 27 |
+
"rope_theta": 10000.0,
|
| 28 |
+
"rope_type": "default"
|
| 29 |
+
},
|
| 30 |
+
"tie_word_embeddings": true,
|
| 31 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 32 |
+
"transformers_version": "5.15.0.dev0",
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"vocab_size": 4096
|
| 35 |
+
}
|
out/mlp-s10-94L_run/checkpoint-100/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2be26509aa8682be1562cdefdc4fb702fa74f487c6ebeb27a02060b133a0c94a
|
| 3 |
+
size 31981960
|
out/mlp-s10-94L_run/checkpoint-100/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:de4ee11bbfdf5bd8a135bcd6b03e23684bb89f9318b0733ff65ec4481b31abe2
|
| 3 |
+
size 64439848
|
out/mlp-s10-94L_run/checkpoint-100/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3cf9097d4513154245c48236b6ec5137b7ee2a21c9f58f2cba798ea275c6026f
|
| 3 |
+
size 14244
|
out/mlp-s10-94L_run/checkpoint-100/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:906745ab5f61e73e8c1ba850ef3b839e2766eba889b49c843836266988895165
|
| 3 |
+
size 1064
|
out/mlp-s10-94L_run/checkpoint-100/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
out/mlp-s10-94L_run/checkpoint-100/tokenizer_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|endoftext|>",
|
| 5 |
+
"eos_token": "<|endoftext|>",
|
| 6 |
+
"errors": "replace",
|
| 7 |
+
"is_local": false,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 1024,
|
| 10 |
+
"pad_token": "<|endoftext|>",
|
| 11 |
+
"tokenizer_class": "GPT2Tokenizer",
|
| 12 |
+
"unk_token": "<|endoftext|>"
|
| 13 |
+
}
|
out/mlp-s10-94L_run/checkpoint-100/trainer_state.json
ADDED
|
@@ -0,0 +1,85 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.05392289026691831,
|
| 6 |
+
"eval_steps": 50,
|
| 7 |
+
"global_step": 100,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.010784578053383662,
|
| 14 |
+
"grad_norm": 3.75,
|
| 15 |
+
"learning_rate": 0.001,
|
| 16 |
+
"loss": 28.570291137695314,
|
| 17 |
+
"step": 20
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.021569156106767323,
|
| 21 |
+
"grad_norm": 0.263671875,
|
| 22 |
+
"learning_rate": 0.001,
|
| 23 |
+
"loss": 24.279835510253907,
|
| 24 |
+
"step": 40
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.026961445133459154,
|
| 28 |
+
"eval_loss": 5.908648490905762,
|
| 29 |
+
"eval_runtime": 16.3836,
|
| 30 |
+
"eval_samples_per_second": 581.495,
|
| 31 |
+
"eval_steps_per_second": 4.578,
|
| 32 |
+
"step": 50
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 0.032353734160150985,
|
| 36 |
+
"grad_norm": 2.03125,
|
| 37 |
+
"learning_rate": 0.001,
|
| 38 |
+
"loss": 23.61173858642578,
|
| 39 |
+
"step": 60
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.043138312213534646,
|
| 43 |
+
"grad_norm": 3.9375,
|
| 44 |
+
"learning_rate": 0.001,
|
| 45 |
+
"loss": 22.80414733886719,
|
| 46 |
+
"step": 80
|
| 47 |
+
},
|
| 48 |
+
{
|
| 49 |
+
"epoch": 0.05392289026691831,
|
| 50 |
+
"grad_norm": 4.5,
|
| 51 |
+
"learning_rate": 0.001,
|
| 52 |
+
"loss": 22.137115478515625,
|
| 53 |
+
"step": 100
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"epoch": 0.05392289026691831,
|
| 57 |
+
"eval_loss": 5.443320274353027,
|
| 58 |
+
"eval_runtime": 16.3492,
|
| 59 |
+
"eval_samples_per_second": 582.72,
|
| 60 |
+
"eval_steps_per_second": 4.587,
|
| 61 |
+
"step": 100
|
| 62 |
+
}
|
| 63 |
+
],
|
| 64 |
+
"logging_steps": 20,
|
| 65 |
+
"max_steps": 1500,
|
| 66 |
+
"num_input_tokens_seen": 0,
|
| 67 |
+
"num_train_epochs": 1,
|
| 68 |
+
"save_steps": 100,
|
| 69 |
+
"stateful_callbacks": {
|
| 70 |
+
"TrainerControl": {
|
| 71 |
+
"args": {
|
| 72 |
+
"should_epoch_stop": false,
|
| 73 |
+
"should_evaluate": false,
|
| 74 |
+
"should_log": false,
|
| 75 |
+
"should_save": true,
|
| 76 |
+
"should_training_stop": false
|
| 77 |
+
},
|
| 78 |
+
"attributes": {}
|
| 79 |
+
}
|
| 80 |
+
},
|
| 81 |
+
"total_flos": 2426166627532800.0,
|
| 82 |
+
"train_batch_size": 128,
|
| 83 |
+
"trial_name": null,
|
| 84 |
+
"trial_params": null
|
| 85 |
+
}
|
out/mlp-s10-94L_run/checkpoint-100/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:873bf272a464f8d4f32a1497a17e4dd88aaf0d116e87d1e95dbff747318696f0
|
| 3 |
+
size 4920
|
out/mlp-s10-94L_run/checkpoint-1000/config.json
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "s10",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"TinyLlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 1,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eos_token_id": 2,
|
| 11 |
+
"head_dim": 32,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 128,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 256,
|
| 16 |
+
"max_position_embeddings": 512,
|
| 17 |
+
"mlp_bias": false,
|
| 18 |
+
"mlp_type": "mlp",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 94,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"pretraining_tp": 1,
|
| 25 |
+
"rms_norm_eps": 1e-06,
|
| 26 |
+
"rope_parameters": {
|
| 27 |
+
"rope_theta": 10000.0,
|
| 28 |
+
"rope_type": "default"
|
| 29 |
+
},
|
| 30 |
+
"tie_word_embeddings": true,
|
| 31 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 32 |
+
"transformers_version": "5.15.0.dev0",
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"vocab_size": 4096
|
| 35 |
+
}
|
out/mlp-s10-94L_run/checkpoint-1000/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e024a0695e460a7767d8870e3aab9f1e1ac1c23d1574c25ef36b65f711261df1
|
| 3 |
+
size 31981960
|
out/mlp-s10-94L_run/checkpoint-1000/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f69330f380b6a898c34ed87209a757d3f977d7f4f9141dddf96d61b05965096e
|
| 3 |
+
size 64439848
|
out/mlp-s10-94L_run/checkpoint-1000/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:645b71843375baf4cc10bc75a2f0f04e91b8e3f8e8929f518a87022898b3bc20
|
| 3 |
+
size 14244
|
out/mlp-s10-94L_run/checkpoint-1000/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0fc3ca8af69cc9d003a05a139e7997960ab393fe31ae1d898b29038573704b4e
|
| 3 |
+
size 1064
|
out/mlp-s10-94L_run/checkpoint-1000/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
out/mlp-s10-94L_run/checkpoint-1000/tokenizer_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|endoftext|>",
|
| 5 |
+
"eos_token": "<|endoftext|>",
|
| 6 |
+
"errors": "replace",
|
| 7 |
+
"is_local": false,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 1024,
|
| 10 |
+
"pad_token": "<|endoftext|>",
|
| 11 |
+
"tokenizer_class": "GPT2Tokenizer",
|
| 12 |
+
"unk_token": "<|endoftext|>"
|
| 13 |
+
}
|
out/mlp-s10-94L_run/checkpoint-1000/trainer_state.json
ADDED
|
@@ -0,0 +1,544 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.5392289026691831,
|
| 6 |
+
"eval_steps": 50,
|
| 7 |
+
"global_step": 1000,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.010784578053383662,
|
| 14 |
+
"grad_norm": 3.75,
|
| 15 |
+
"learning_rate": 0.001,
|
| 16 |
+
"loss": 28.570291137695314,
|
| 17 |
+
"step": 20
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.021569156106767323,
|
| 21 |
+
"grad_norm": 0.263671875,
|
| 22 |
+
"learning_rate": 0.001,
|
| 23 |
+
"loss": 24.279835510253907,
|
| 24 |
+
"step": 40
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.026961445133459154,
|
| 28 |
+
"eval_loss": 5.908648490905762,
|
| 29 |
+
"eval_runtime": 16.3836,
|
| 30 |
+
"eval_samples_per_second": 581.495,
|
| 31 |
+
"eval_steps_per_second": 4.578,
|
| 32 |
+
"step": 50
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 0.032353734160150985,
|
| 36 |
+
"grad_norm": 2.03125,
|
| 37 |
+
"learning_rate": 0.001,
|
| 38 |
+
"loss": 23.61173858642578,
|
| 39 |
+
"step": 60
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.043138312213534646,
|
| 43 |
+
"grad_norm": 3.9375,
|
| 44 |
+
"learning_rate": 0.001,
|
| 45 |
+
"loss": 22.80414733886719,
|
| 46 |
+
"step": 80
|
| 47 |
+
},
|
| 48 |
+
{
|
| 49 |
+
"epoch": 0.05392289026691831,
|
| 50 |
+
"grad_norm": 4.5,
|
| 51 |
+
"learning_rate": 0.001,
|
| 52 |
+
"loss": 22.137115478515625,
|
| 53 |
+
"step": 100
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"epoch": 0.05392289026691831,
|
| 57 |
+
"eval_loss": 5.443320274353027,
|
| 58 |
+
"eval_runtime": 16.3492,
|
| 59 |
+
"eval_samples_per_second": 582.72,
|
| 60 |
+
"eval_steps_per_second": 4.587,
|
| 61 |
+
"step": 100
|
| 62 |
+
},
|
| 63 |
+
{
|
| 64 |
+
"epoch": 0.06470746832030197,
|
| 65 |
+
"grad_norm": 9.0625,
|
| 66 |
+
"learning_rate": 0.001,
|
| 67 |
+
"loss": 21.51625213623047,
|
| 68 |
+
"step": 120
|
| 69 |
+
},
|
| 70 |
+
{
|
| 71 |
+
"epoch": 0.07549204637368563,
|
| 72 |
+
"grad_norm": 8.0625,
|
| 73 |
+
"learning_rate": 0.001,
|
| 74 |
+
"loss": 20.9159912109375,
|
| 75 |
+
"step": 140
|
| 76 |
+
},
|
| 77 |
+
{
|
| 78 |
+
"epoch": 0.08088433540037746,
|
| 79 |
+
"eval_loss": 5.031351566314697,
|
| 80 |
+
"eval_runtime": 16.3105,
|
| 81 |
+
"eval_samples_per_second": 584.103,
|
| 82 |
+
"eval_steps_per_second": 4.598,
|
| 83 |
+
"step": 150
|
| 84 |
+
},
|
| 85 |
+
{
|
| 86 |
+
"epoch": 0.08627662442706929,
|
| 87 |
+
"grad_norm": 3.828125,
|
| 88 |
+
"learning_rate": 0.001,
|
| 89 |
+
"loss": 20.133233642578126,
|
| 90 |
+
"step": 160
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"epoch": 0.09706120248045295,
|
| 94 |
+
"grad_norm": 5.875,
|
| 95 |
+
"learning_rate": 0.001,
|
| 96 |
+
"loss": 19.276626586914062,
|
| 97 |
+
"step": 180
|
| 98 |
+
},
|
| 99 |
+
{
|
| 100 |
+
"epoch": 0.10784578053383662,
|
| 101 |
+
"grad_norm": 3.796875,
|
| 102 |
+
"learning_rate": 0.001,
|
| 103 |
+
"loss": 18.316180419921874,
|
| 104 |
+
"step": 200
|
| 105 |
+
},
|
| 106 |
+
{
|
| 107 |
+
"epoch": 0.10784578053383662,
|
| 108 |
+
"eval_loss": 4.4591450691223145,
|
| 109 |
+
"eval_runtime": 16.5273,
|
| 110 |
+
"eval_samples_per_second": 576.441,
|
| 111 |
+
"eval_steps_per_second": 4.538,
|
| 112 |
+
"step": 200
|
| 113 |
+
},
|
| 114 |
+
{
|
| 115 |
+
"epoch": 0.11863035858722028,
|
| 116 |
+
"grad_norm": 3.125,
|
| 117 |
+
"learning_rate": 0.001,
|
| 118 |
+
"loss": 17.47910919189453,
|
| 119 |
+
"step": 220
|
| 120 |
+
},
|
| 121 |
+
{
|
| 122 |
+
"epoch": 0.12941493664060394,
|
| 123 |
+
"grad_norm": 3.78125,
|
| 124 |
+
"learning_rate": 0.001,
|
| 125 |
+
"loss": 16.857704162597656,
|
| 126 |
+
"step": 240
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"epoch": 0.13480722566729578,
|
| 130 |
+
"eval_loss": 4.037808418273926,
|
| 131 |
+
"eval_runtime": 16.7437,
|
| 132 |
+
"eval_samples_per_second": 568.991,
|
| 133 |
+
"eval_steps_per_second": 4.479,
|
| 134 |
+
"step": 250
|
| 135 |
+
},
|
| 136 |
+
{
|
| 137 |
+
"epoch": 0.1401995146939876,
|
| 138 |
+
"grad_norm": 4.8125,
|
| 139 |
+
"learning_rate": 0.001,
|
| 140 |
+
"loss": 16.15790710449219,
|
| 141 |
+
"step": 260
|
| 142 |
+
},
|
| 143 |
+
{
|
| 144 |
+
"epoch": 0.15098409274737126,
|
| 145 |
+
"grad_norm": 2.453125,
|
| 146 |
+
"learning_rate": 0.001,
|
| 147 |
+
"loss": 15.610972595214843,
|
| 148 |
+
"step": 280
|
| 149 |
+
},
|
| 150 |
+
{
|
| 151 |
+
"epoch": 0.1617686708007549,
|
| 152 |
+
"grad_norm": 5.03125,
|
| 153 |
+
"learning_rate": 0.001,
|
| 154 |
+
"loss": 15.262626647949219,
|
| 155 |
+
"step": 300
|
| 156 |
+
},
|
| 157 |
+
{
|
| 158 |
+
"epoch": 0.1617686708007549,
|
| 159 |
+
"eval_loss": 3.7790420055389404,
|
| 160 |
+
"eval_runtime": 16.4839,
|
| 161 |
+
"eval_samples_per_second": 577.957,
|
| 162 |
+
"eval_steps_per_second": 4.55,
|
| 163 |
+
"step": 300
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.17255324885413859,
|
| 167 |
+
"grad_norm": 2.53125,
|
| 168 |
+
"learning_rate": 0.001,
|
| 169 |
+
"loss": 14.862835693359376,
|
| 170 |
+
"step": 320
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.18333782690752223,
|
| 174 |
+
"grad_norm": 1.8828125,
|
| 175 |
+
"learning_rate": 0.001,
|
| 176 |
+
"loss": 14.4195556640625,
|
| 177 |
+
"step": 340
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.18873011593421407,
|
| 181 |
+
"eval_loss": 3.508929491043091,
|
| 182 |
+
"eval_runtime": 16.5096,
|
| 183 |
+
"eval_samples_per_second": 577.059,
|
| 184 |
+
"eval_steps_per_second": 4.543,
|
| 185 |
+
"step": 350
|
| 186 |
+
},
|
| 187 |
+
{
|
| 188 |
+
"epoch": 0.1941224049609059,
|
| 189 |
+
"grad_norm": 2.21875,
|
| 190 |
+
"learning_rate": 0.001,
|
| 191 |
+
"loss": 14.034010314941407,
|
| 192 |
+
"step": 360
|
| 193 |
+
},
|
| 194 |
+
{
|
| 195 |
+
"epoch": 0.20490698301428956,
|
| 196 |
+
"grad_norm": 1.6171875,
|
| 197 |
+
"learning_rate": 0.001,
|
| 198 |
+
"loss": 13.656217956542969,
|
| 199 |
+
"step": 380
|
| 200 |
+
},
|
| 201 |
+
{
|
| 202 |
+
"epoch": 0.21569156106767323,
|
| 203 |
+
"grad_norm": 1.0859375,
|
| 204 |
+
"learning_rate": 0.001,
|
| 205 |
+
"loss": 13.353656005859374,
|
| 206 |
+
"step": 400
|
| 207 |
+
},
|
| 208 |
+
{
|
| 209 |
+
"epoch": 0.21569156106767323,
|
| 210 |
+
"eval_loss": 3.3230206966400146,
|
| 211 |
+
"eval_runtime": 16.5021,
|
| 212 |
+
"eval_samples_per_second": 577.322,
|
| 213 |
+
"eval_steps_per_second": 4.545,
|
| 214 |
+
"step": 400
|
| 215 |
+
},
|
| 216 |
+
{
|
| 217 |
+
"epoch": 0.22647613912105688,
|
| 218 |
+
"grad_norm": 2.65625,
|
| 219 |
+
"learning_rate": 0.001,
|
| 220 |
+
"loss": 13.050607299804687,
|
| 221 |
+
"step": 420
|
| 222 |
+
},
|
| 223 |
+
{
|
| 224 |
+
"epoch": 0.23726071717444056,
|
| 225 |
+
"grad_norm": 2.046875,
|
| 226 |
+
"learning_rate": 0.001,
|
| 227 |
+
"loss": 12.783986663818359,
|
| 228 |
+
"step": 440
|
| 229 |
+
},
|
| 230 |
+
{
|
| 231 |
+
"epoch": 0.2426530062011324,
|
| 232 |
+
"eval_loss": 3.1205618381500244,
|
| 233 |
+
"eval_runtime": 16.4401,
|
| 234 |
+
"eval_samples_per_second": 579.497,
|
| 235 |
+
"eval_steps_per_second": 4.562,
|
| 236 |
+
"step": 450
|
| 237 |
+
},
|
| 238 |
+
{
|
| 239 |
+
"epoch": 0.2480452952278242,
|
| 240 |
+
"grad_norm": 2.34375,
|
| 241 |
+
"learning_rate": 0.001,
|
| 242 |
+
"loss": 12.455777740478515,
|
| 243 |
+
"step": 460
|
| 244 |
+
},
|
| 245 |
+
{
|
| 246 |
+
"epoch": 0.2588298732812079,
|
| 247 |
+
"grad_norm": 1.734375,
|
| 248 |
+
"learning_rate": 0.001,
|
| 249 |
+
"loss": 12.180888366699218,
|
| 250 |
+
"step": 480
|
| 251 |
+
},
|
| 252 |
+
{
|
| 253 |
+
"epoch": 0.26961445133459155,
|
| 254 |
+
"grad_norm": 1.65625,
|
| 255 |
+
"learning_rate": 0.001,
|
| 256 |
+
"loss": 11.921077728271484,
|
| 257 |
+
"step": 500
|
| 258 |
+
},
|
| 259 |
+
{
|
| 260 |
+
"epoch": 0.26961445133459155,
|
| 261 |
+
"eval_loss": 2.950833797454834,
|
| 262 |
+
"eval_runtime": 16.4185,
|
| 263 |
+
"eval_samples_per_second": 580.259,
|
| 264 |
+
"eval_steps_per_second": 4.568,
|
| 265 |
+
"step": 500
|
| 266 |
+
},
|
| 267 |
+
{
|
| 268 |
+
"epoch": 0.2803990293879752,
|
| 269 |
+
"grad_norm": 1.5234375,
|
| 270 |
+
"learning_rate": 0.001,
|
| 271 |
+
"loss": 11.677184295654296,
|
| 272 |
+
"step": 520
|
| 273 |
+
},
|
| 274 |
+
{
|
| 275 |
+
"epoch": 0.29118360744135885,
|
| 276 |
+
"grad_norm": 2.296875,
|
| 277 |
+
"learning_rate": 0.001,
|
| 278 |
+
"loss": 11.456565856933594,
|
| 279 |
+
"step": 540
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"epoch": 0.2965758964680507,
|
| 283 |
+
"eval_loss": 2.813769817352295,
|
| 284 |
+
"eval_runtime": 16.3503,
|
| 285 |
+
"eval_samples_per_second": 582.681,
|
| 286 |
+
"eval_steps_per_second": 4.587,
|
| 287 |
+
"step": 550
|
| 288 |
+
},
|
| 289 |
+
{
|
| 290 |
+
"epoch": 0.3019681854947425,
|
| 291 |
+
"grad_norm": 1.84375,
|
| 292 |
+
"learning_rate": 0.001,
|
| 293 |
+
"loss": 11.254852294921875,
|
| 294 |
+
"step": 560
|
| 295 |
+
},
|
| 296 |
+
{
|
| 297 |
+
"epoch": 0.3127527635481262,
|
| 298 |
+
"grad_norm": 1.9609375,
|
| 299 |
+
"learning_rate": 0.001,
|
| 300 |
+
"loss": 11.02415771484375,
|
| 301 |
+
"step": 580
|
| 302 |
+
},
|
| 303 |
+
{
|
| 304 |
+
"epoch": 0.3235373416015098,
|
| 305 |
+
"grad_norm": 2.109375,
|
| 306 |
+
"learning_rate": 0.001,
|
| 307 |
+
"loss": 10.82885513305664,
|
| 308 |
+
"step": 600
|
| 309 |
+
},
|
| 310 |
+
{
|
| 311 |
+
"epoch": 0.3235373416015098,
|
| 312 |
+
"eval_loss": 2.6941640377044678,
|
| 313 |
+
"eval_runtime": 16.5123,
|
| 314 |
+
"eval_samples_per_second": 576.964,
|
| 315 |
+
"eval_steps_per_second": 4.542,
|
| 316 |
+
"step": 600
|
| 317 |
+
},
|
| 318 |
+
{
|
| 319 |
+
"epoch": 0.3343219196548935,
|
| 320 |
+
"grad_norm": 1.2578125,
|
| 321 |
+
"learning_rate": 0.001,
|
| 322 |
+
"loss": 10.672529602050782,
|
| 323 |
+
"step": 620
|
| 324 |
+
},
|
| 325 |
+
{
|
| 326 |
+
"epoch": 0.34510649770827717,
|
| 327 |
+
"grad_norm": 1.9375,
|
| 328 |
+
"learning_rate": 0.001,
|
| 329 |
+
"loss": 10.486093139648437,
|
| 330 |
+
"step": 640
|
| 331 |
+
},
|
| 332 |
+
{
|
| 333 |
+
"epoch": 0.350498786734969,
|
| 334 |
+
"eval_loss": 2.5829620361328125,
|
| 335 |
+
"eval_runtime": 16.3847,
|
| 336 |
+
"eval_samples_per_second": 581.459,
|
| 337 |
+
"eval_steps_per_second": 4.577,
|
| 338 |
+
"step": 650
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.35589107576166085,
|
| 342 |
+
"grad_norm": 1.4765625,
|
| 343 |
+
"learning_rate": 0.001,
|
| 344 |
+
"loss": 10.307514953613282,
|
| 345 |
+
"step": 660
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.36667565381504447,
|
| 349 |
+
"grad_norm": 1.609375,
|
| 350 |
+
"learning_rate": 0.001,
|
| 351 |
+
"loss": 10.156155395507813,
|
| 352 |
+
"step": 680
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.37746023186842814,
|
| 356 |
+
"grad_norm": 2.078125,
|
| 357 |
+
"learning_rate": 0.001,
|
| 358 |
+
"loss": 10.004931640625,
|
| 359 |
+
"step": 700
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.37746023186842814,
|
| 363 |
+
"eval_loss": 2.4850780963897705,
|
| 364 |
+
"eval_runtime": 16.6426,
|
| 365 |
+
"eval_samples_per_second": 572.445,
|
| 366 |
+
"eval_steps_per_second": 4.506,
|
| 367 |
+
"step": 700
|
| 368 |
+
},
|
| 369 |
+
{
|
| 370 |
+
"epoch": 0.3882448099218118,
|
| 371 |
+
"grad_norm": 1.8203125,
|
| 372 |
+
"learning_rate": 0.001,
|
| 373 |
+
"loss": 9.830353546142579,
|
| 374 |
+
"step": 720
|
| 375 |
+
},
|
| 376 |
+
{
|
| 377 |
+
"epoch": 0.3990293879751955,
|
| 378 |
+
"grad_norm": 1.671875,
|
| 379 |
+
"learning_rate": 0.001,
|
| 380 |
+
"loss": 9.726395416259766,
|
| 381 |
+
"step": 740
|
| 382 |
+
},
|
| 383 |
+
{
|
| 384 |
+
"epoch": 0.4044216770018873,
|
| 385 |
+
"eval_loss": 2.3961310386657715,
|
| 386 |
+
"eval_runtime": 16.484,
|
| 387 |
+
"eval_samples_per_second": 577.953,
|
| 388 |
+
"eval_steps_per_second": 4.55,
|
| 389 |
+
"step": 750
|
| 390 |
+
},
|
| 391 |
+
{
|
| 392 |
+
"epoch": 0.4098139660285791,
|
| 393 |
+
"grad_norm": 1.5,
|
| 394 |
+
"learning_rate": 0.001,
|
| 395 |
+
"loss": 9.585889434814453,
|
| 396 |
+
"step": 760
|
| 397 |
+
},
|
| 398 |
+
{
|
| 399 |
+
"epoch": 0.4205985440819628,
|
| 400 |
+
"grad_norm": 1.4921875,
|
| 401 |
+
"learning_rate": 0.001,
|
| 402 |
+
"loss": 9.474940490722656,
|
| 403 |
+
"step": 780
|
| 404 |
+
},
|
| 405 |
+
{
|
| 406 |
+
"epoch": 0.43138312213534646,
|
| 407 |
+
"grad_norm": 1.7890625,
|
| 408 |
+
"learning_rate": 0.001,
|
| 409 |
+
"loss": 9.303884124755859,
|
| 410 |
+
"step": 800
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 0.43138312213534646,
|
| 414 |
+
"eval_loss": 2.3192710876464844,
|
| 415 |
+
"eval_runtime": 16.362,
|
| 416 |
+
"eval_samples_per_second": 582.263,
|
| 417 |
+
"eval_steps_per_second": 4.584,
|
| 418 |
+
"step": 800
|
| 419 |
+
},
|
| 420 |
+
{
|
| 421 |
+
"epoch": 0.44216770018873014,
|
| 422 |
+
"grad_norm": 1.96875,
|
| 423 |
+
"learning_rate": 0.001,
|
| 424 |
+
"loss": 9.214578247070312,
|
| 425 |
+
"step": 820
|
| 426 |
+
},
|
| 427 |
+
{
|
| 428 |
+
"epoch": 0.45295227824211376,
|
| 429 |
+
"grad_norm": 1.265625,
|
| 430 |
+
"learning_rate": 0.001,
|
| 431 |
+
"loss": 9.110013580322265,
|
| 432 |
+
"step": 840
|
| 433 |
+
},
|
| 434 |
+
{
|
| 435 |
+
"epoch": 0.4583445672688056,
|
| 436 |
+
"eval_loss": 2.255786895751953,
|
| 437 |
+
"eval_runtime": 16.6344,
|
| 438 |
+
"eval_samples_per_second": 572.728,
|
| 439 |
+
"eval_steps_per_second": 4.509,
|
| 440 |
+
"step": 850
|
| 441 |
+
},
|
| 442 |
+
{
|
| 443 |
+
"epoch": 0.46373685629549743,
|
| 444 |
+
"grad_norm": 1.53125,
|
| 445 |
+
"learning_rate": 0.001,
|
| 446 |
+
"loss": 8.989684295654296,
|
| 447 |
+
"step": 860
|
| 448 |
+
},
|
| 449 |
+
{
|
| 450 |
+
"epoch": 0.4745214343488811,
|
| 451 |
+
"grad_norm": 1.46875,
|
| 452 |
+
"learning_rate": 0.001,
|
| 453 |
+
"loss": 8.903449249267577,
|
| 454 |
+
"step": 880
|
| 455 |
+
},
|
| 456 |
+
{
|
| 457 |
+
"epoch": 0.4853060124022648,
|
| 458 |
+
"grad_norm": 1.5625,
|
| 459 |
+
"learning_rate": 0.001,
|
| 460 |
+
"loss": 8.798023223876953,
|
| 461 |
+
"step": 900
|
| 462 |
+
},
|
| 463 |
+
{
|
| 464 |
+
"epoch": 0.4853060124022648,
|
| 465 |
+
"eval_loss": 2.1990966796875,
|
| 466 |
+
"eval_runtime": 16.4377,
|
| 467 |
+
"eval_samples_per_second": 579.581,
|
| 468 |
+
"eval_steps_per_second": 4.563,
|
| 469 |
+
"step": 900
|
| 470 |
+
},
|
| 471 |
+
{
|
| 472 |
+
"epoch": 0.4960905904556484,
|
| 473 |
+
"grad_norm": 1.3515625,
|
| 474 |
+
"learning_rate": 0.001,
|
| 475 |
+
"loss": 8.711866760253907,
|
| 476 |
+
"step": 920
|
| 477 |
+
},
|
| 478 |
+
{
|
| 479 |
+
"epoch": 0.5068751685090321,
|
| 480 |
+
"grad_norm": 1.375,
|
| 481 |
+
"learning_rate": 0.001,
|
| 482 |
+
"loss": 8.65683822631836,
|
| 483 |
+
"step": 940
|
| 484 |
+
},
|
| 485 |
+
{
|
| 486 |
+
"epoch": 0.5122674575357239,
|
| 487 |
+
"eval_loss": 2.15165638923645,
|
| 488 |
+
"eval_runtime": 16.515,
|
| 489 |
+
"eval_samples_per_second": 576.869,
|
| 490 |
+
"eval_steps_per_second": 4.541,
|
| 491 |
+
"step": 950
|
| 492 |
+
},
|
| 493 |
+
{
|
| 494 |
+
"epoch": 0.5176597465624158,
|
| 495 |
+
"grad_norm": 1.546875,
|
| 496 |
+
"learning_rate": 0.001,
|
| 497 |
+
"loss": 8.596742248535156,
|
| 498 |
+
"step": 960
|
| 499 |
+
},
|
| 500 |
+
{
|
| 501 |
+
"epoch": 0.5284443246157994,
|
| 502 |
+
"grad_norm": 1.3828125,
|
| 503 |
+
"learning_rate": 0.001,
|
| 504 |
+
"loss": 8.528240966796876,
|
| 505 |
+
"step": 980
|
| 506 |
+
},
|
| 507 |
+
{
|
| 508 |
+
"epoch": 0.5392289026691831,
|
| 509 |
+
"grad_norm": 1.390625,
|
| 510 |
+
"learning_rate": 0.001,
|
| 511 |
+
"loss": 8.455123138427734,
|
| 512 |
+
"step": 1000
|
| 513 |
+
},
|
| 514 |
+
{
|
| 515 |
+
"epoch": 0.5392289026691831,
|
| 516 |
+
"eval_loss": 2.1049983501434326,
|
| 517 |
+
"eval_runtime": 16.4683,
|
| 518 |
+
"eval_samples_per_second": 578.506,
|
| 519 |
+
"eval_steps_per_second": 4.554,
|
| 520 |
+
"step": 1000
|
| 521 |
+
}
|
| 522 |
+
],
|
| 523 |
+
"logging_steps": 20,
|
| 524 |
+
"max_steps": 1500,
|
| 525 |
+
"num_input_tokens_seen": 0,
|
| 526 |
+
"num_train_epochs": 1,
|
| 527 |
+
"save_steps": 100,
|
| 528 |
+
"stateful_callbacks": {
|
| 529 |
+
"TrainerControl": {
|
| 530 |
+
"args": {
|
| 531 |
+
"should_epoch_stop": false,
|
| 532 |
+
"should_evaluate": false,
|
| 533 |
+
"should_log": false,
|
| 534 |
+
"should_save": true,
|
| 535 |
+
"should_training_stop": false
|
| 536 |
+
},
|
| 537 |
+
"attributes": {}
|
| 538 |
+
}
|
| 539 |
+
},
|
| 540 |
+
"total_flos": 2.4261666275328e+16,
|
| 541 |
+
"train_batch_size": 128,
|
| 542 |
+
"trial_name": null,
|
| 543 |
+
"trial_params": null
|
| 544 |
+
}
|
out/mlp-s10-94L_run/checkpoint-1000/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:873bf272a464f8d4f32a1497a17e4dd88aaf0d116e87d1e95dbff747318696f0
|
| 3 |
+
size 4920
|
out/mlp-s10-94L_run/checkpoint-200/config.json
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "s10",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"TinyLlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 1,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eos_token_id": 2,
|
| 11 |
+
"head_dim": 32,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 128,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 256,
|
| 16 |
+
"max_position_embeddings": 512,
|
| 17 |
+
"mlp_bias": false,
|
| 18 |
+
"mlp_type": "mlp",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 94,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"pretraining_tp": 1,
|
| 25 |
+
"rms_norm_eps": 1e-06,
|
| 26 |
+
"rope_parameters": {
|
| 27 |
+
"rope_theta": 10000.0,
|
| 28 |
+
"rope_type": "default"
|
| 29 |
+
},
|
| 30 |
+
"tie_word_embeddings": true,
|
| 31 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 32 |
+
"transformers_version": "5.15.0.dev0",
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"vocab_size": 4096
|
| 35 |
+
}
|
out/mlp-s10-94L_run/checkpoint-200/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:515aaf4623cc4bbdf339385647192c1001d74d42f1127e98e0e13e95ebf8a0c6
|
| 3 |
+
size 31981960
|
out/mlp-s10-94L_run/checkpoint-200/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:65b74feea8eabc2a019f74d207e54f41128d62e299b263be00a4f00a214f044d
|
| 3 |
+
size 64439848
|
out/mlp-s10-94L_run/checkpoint-200/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ecefbb3f17bb76b6655eb0157c98b5287c17fa4b4c72a6b9068b0823ce9fd18d
|
| 3 |
+
size 14244
|
out/mlp-s10-94L_run/checkpoint-200/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2d3b5a970fc92af42664e84ba686059eaaacfaa668691c0cf6d2afddda677461
|
| 3 |
+
size 1064
|
out/mlp-s10-94L_run/checkpoint-200/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
out/mlp-s10-94L_run/checkpoint-200/tokenizer_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|endoftext|>",
|
| 5 |
+
"eos_token": "<|endoftext|>",
|
| 6 |
+
"errors": "replace",
|
| 7 |
+
"is_local": false,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 1024,
|
| 10 |
+
"pad_token": "<|endoftext|>",
|
| 11 |
+
"tokenizer_class": "GPT2Tokenizer",
|
| 12 |
+
"unk_token": "<|endoftext|>"
|
| 13 |
+
}
|
out/mlp-s10-94L_run/checkpoint-200/trainer_state.json
ADDED
|
@@ -0,0 +1,136 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.10784578053383662,
|
| 6 |
+
"eval_steps": 50,
|
| 7 |
+
"global_step": 200,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.010784578053383662,
|
| 14 |
+
"grad_norm": 3.75,
|
| 15 |
+
"learning_rate": 0.001,
|
| 16 |
+
"loss": 28.570291137695314,
|
| 17 |
+
"step": 20
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.021569156106767323,
|
| 21 |
+
"grad_norm": 0.263671875,
|
| 22 |
+
"learning_rate": 0.001,
|
| 23 |
+
"loss": 24.279835510253907,
|
| 24 |
+
"step": 40
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.026961445133459154,
|
| 28 |
+
"eval_loss": 5.908648490905762,
|
| 29 |
+
"eval_runtime": 16.3836,
|
| 30 |
+
"eval_samples_per_second": 581.495,
|
| 31 |
+
"eval_steps_per_second": 4.578,
|
| 32 |
+
"step": 50
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 0.032353734160150985,
|
| 36 |
+
"grad_norm": 2.03125,
|
| 37 |
+
"learning_rate": 0.001,
|
| 38 |
+
"loss": 23.61173858642578,
|
| 39 |
+
"step": 60
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.043138312213534646,
|
| 43 |
+
"grad_norm": 3.9375,
|
| 44 |
+
"learning_rate": 0.001,
|
| 45 |
+
"loss": 22.80414733886719,
|
| 46 |
+
"step": 80
|
| 47 |
+
},
|
| 48 |
+
{
|
| 49 |
+
"epoch": 0.05392289026691831,
|
| 50 |
+
"grad_norm": 4.5,
|
| 51 |
+
"learning_rate": 0.001,
|
| 52 |
+
"loss": 22.137115478515625,
|
| 53 |
+
"step": 100
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"epoch": 0.05392289026691831,
|
| 57 |
+
"eval_loss": 5.443320274353027,
|
| 58 |
+
"eval_runtime": 16.3492,
|
| 59 |
+
"eval_samples_per_second": 582.72,
|
| 60 |
+
"eval_steps_per_second": 4.587,
|
| 61 |
+
"step": 100
|
| 62 |
+
},
|
| 63 |
+
{
|
| 64 |
+
"epoch": 0.06470746832030197,
|
| 65 |
+
"grad_norm": 9.0625,
|
| 66 |
+
"learning_rate": 0.001,
|
| 67 |
+
"loss": 21.51625213623047,
|
| 68 |
+
"step": 120
|
| 69 |
+
},
|
| 70 |
+
{
|
| 71 |
+
"epoch": 0.07549204637368563,
|
| 72 |
+
"grad_norm": 8.0625,
|
| 73 |
+
"learning_rate": 0.001,
|
| 74 |
+
"loss": 20.9159912109375,
|
| 75 |
+
"step": 140
|
| 76 |
+
},
|
| 77 |
+
{
|
| 78 |
+
"epoch": 0.08088433540037746,
|
| 79 |
+
"eval_loss": 5.031351566314697,
|
| 80 |
+
"eval_runtime": 16.3105,
|
| 81 |
+
"eval_samples_per_second": 584.103,
|
| 82 |
+
"eval_steps_per_second": 4.598,
|
| 83 |
+
"step": 150
|
| 84 |
+
},
|
| 85 |
+
{
|
| 86 |
+
"epoch": 0.08627662442706929,
|
| 87 |
+
"grad_norm": 3.828125,
|
| 88 |
+
"learning_rate": 0.001,
|
| 89 |
+
"loss": 20.133233642578126,
|
| 90 |
+
"step": 160
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"epoch": 0.09706120248045295,
|
| 94 |
+
"grad_norm": 5.875,
|
| 95 |
+
"learning_rate": 0.001,
|
| 96 |
+
"loss": 19.276626586914062,
|
| 97 |
+
"step": 180
|
| 98 |
+
},
|
| 99 |
+
{
|
| 100 |
+
"epoch": 0.10784578053383662,
|
| 101 |
+
"grad_norm": 3.796875,
|
| 102 |
+
"learning_rate": 0.001,
|
| 103 |
+
"loss": 18.316180419921874,
|
| 104 |
+
"step": 200
|
| 105 |
+
},
|
| 106 |
+
{
|
| 107 |
+
"epoch": 0.10784578053383662,
|
| 108 |
+
"eval_loss": 4.4591450691223145,
|
| 109 |
+
"eval_runtime": 16.5273,
|
| 110 |
+
"eval_samples_per_second": 576.441,
|
| 111 |
+
"eval_steps_per_second": 4.538,
|
| 112 |
+
"step": 200
|
| 113 |
+
}
|
| 114 |
+
],
|
| 115 |
+
"logging_steps": 20,
|
| 116 |
+
"max_steps": 1500,
|
| 117 |
+
"num_input_tokens_seen": 0,
|
| 118 |
+
"num_train_epochs": 1,
|
| 119 |
+
"save_steps": 100,
|
| 120 |
+
"stateful_callbacks": {
|
| 121 |
+
"TrainerControl": {
|
| 122 |
+
"args": {
|
| 123 |
+
"should_epoch_stop": false,
|
| 124 |
+
"should_evaluate": false,
|
| 125 |
+
"should_log": false,
|
| 126 |
+
"should_save": true,
|
| 127 |
+
"should_training_stop": false
|
| 128 |
+
},
|
| 129 |
+
"attributes": {}
|
| 130 |
+
}
|
| 131 |
+
},
|
| 132 |
+
"total_flos": 4852333255065600.0,
|
| 133 |
+
"train_batch_size": 128,
|
| 134 |
+
"trial_name": null,
|
| 135 |
+
"trial_params": null
|
| 136 |
+
}
|
out/mlp-s10-94L_run/checkpoint-200/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:873bf272a464f8d4f32a1497a17e4dd88aaf0d116e87d1e95dbff747318696f0
|
| 3 |
+
size 4920
|
out/mlp-s10-94L_run/checkpoint-300/config.json
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "s10",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"TinyLlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 1,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eos_token_id": 2,
|
| 11 |
+
"head_dim": 32,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 128,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 256,
|
| 16 |
+
"max_position_embeddings": 512,
|
| 17 |
+
"mlp_bias": false,
|
| 18 |
+
"mlp_type": "mlp",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 94,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"pretraining_tp": 1,
|
| 25 |
+
"rms_norm_eps": 1e-06,
|
| 26 |
+
"rope_parameters": {
|
| 27 |
+
"rope_theta": 10000.0,
|
| 28 |
+
"rope_type": "default"
|
| 29 |
+
},
|
| 30 |
+
"tie_word_embeddings": true,
|
| 31 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 32 |
+
"transformers_version": "5.15.0.dev0",
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"vocab_size": 4096
|
| 35 |
+
}
|
out/mlp-s10-94L_run/checkpoint-300/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:cbd6e8decd8bec1226de38af592cce777770c132c904845b0d7a5cb6382009a2
|
| 3 |
+
size 31981960
|
out/mlp-s10-94L_run/checkpoint-300/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:07997fea4a1b6f79f73da03ce515391657690ed21b96dd52031f331daf8109aa
|
| 3 |
+
size 64439848
|
out/mlp-s10-94L_run/checkpoint-300/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:068fbd993087219c15b8c0baa13fc39644a4dcdfe92d8be3fa6434deece90371
|
| 3 |
+
size 14244
|
out/mlp-s10-94L_run/checkpoint-300/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:76cd21f32221b90fb8288fd2428595ec8b5039fe2425086d67cacb5c8d814c9e
|
| 3 |
+
size 1064
|
out/mlp-s10-94L_run/checkpoint-300/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
out/mlp-s10-94L_run/checkpoint-300/tokenizer_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|endoftext|>",
|
| 5 |
+
"eos_token": "<|endoftext|>",
|
| 6 |
+
"errors": "replace",
|
| 7 |
+
"is_local": false,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 1024,
|
| 10 |
+
"pad_token": "<|endoftext|>",
|
| 11 |
+
"tokenizer_class": "GPT2Tokenizer",
|
| 12 |
+
"unk_token": "<|endoftext|>"
|
| 13 |
+
}
|
out/mlp-s10-94L_run/checkpoint-300/trainer_state.json
ADDED
|
@@ -0,0 +1,187 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.1617686708007549,
|
| 6 |
+
"eval_steps": 50,
|
| 7 |
+
"global_step": 300,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.010784578053383662,
|
| 14 |
+
"grad_norm": 3.75,
|
| 15 |
+
"learning_rate": 0.001,
|
| 16 |
+
"loss": 28.570291137695314,
|
| 17 |
+
"step": 20
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.021569156106767323,
|
| 21 |
+
"grad_norm": 0.263671875,
|
| 22 |
+
"learning_rate": 0.001,
|
| 23 |
+
"loss": 24.279835510253907,
|
| 24 |
+
"step": 40
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.026961445133459154,
|
| 28 |
+
"eval_loss": 5.908648490905762,
|
| 29 |
+
"eval_runtime": 16.3836,
|
| 30 |
+
"eval_samples_per_second": 581.495,
|
| 31 |
+
"eval_steps_per_second": 4.578,
|
| 32 |
+
"step": 50
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 0.032353734160150985,
|
| 36 |
+
"grad_norm": 2.03125,
|
| 37 |
+
"learning_rate": 0.001,
|
| 38 |
+
"loss": 23.61173858642578,
|
| 39 |
+
"step": 60
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.043138312213534646,
|
| 43 |
+
"grad_norm": 3.9375,
|
| 44 |
+
"learning_rate": 0.001,
|
| 45 |
+
"loss": 22.80414733886719,
|
| 46 |
+
"step": 80
|
| 47 |
+
},
|
| 48 |
+
{
|
| 49 |
+
"epoch": 0.05392289026691831,
|
| 50 |
+
"grad_norm": 4.5,
|
| 51 |
+
"learning_rate": 0.001,
|
| 52 |
+
"loss": 22.137115478515625,
|
| 53 |
+
"step": 100
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"epoch": 0.05392289026691831,
|
| 57 |
+
"eval_loss": 5.443320274353027,
|
| 58 |
+
"eval_runtime": 16.3492,
|
| 59 |
+
"eval_samples_per_second": 582.72,
|
| 60 |
+
"eval_steps_per_second": 4.587,
|
| 61 |
+
"step": 100
|
| 62 |
+
},
|
| 63 |
+
{
|
| 64 |
+
"epoch": 0.06470746832030197,
|
| 65 |
+
"grad_norm": 9.0625,
|
| 66 |
+
"learning_rate": 0.001,
|
| 67 |
+
"loss": 21.51625213623047,
|
| 68 |
+
"step": 120
|
| 69 |
+
},
|
| 70 |
+
{
|
| 71 |
+
"epoch": 0.07549204637368563,
|
| 72 |
+
"grad_norm": 8.0625,
|
| 73 |
+
"learning_rate": 0.001,
|
| 74 |
+
"loss": 20.9159912109375,
|
| 75 |
+
"step": 140
|
| 76 |
+
},
|
| 77 |
+
{
|
| 78 |
+
"epoch": 0.08088433540037746,
|
| 79 |
+
"eval_loss": 5.031351566314697,
|
| 80 |
+
"eval_runtime": 16.3105,
|
| 81 |
+
"eval_samples_per_second": 584.103,
|
| 82 |
+
"eval_steps_per_second": 4.598,
|
| 83 |
+
"step": 150
|
| 84 |
+
},
|
| 85 |
+
{
|
| 86 |
+
"epoch": 0.08627662442706929,
|
| 87 |
+
"grad_norm": 3.828125,
|
| 88 |
+
"learning_rate": 0.001,
|
| 89 |
+
"loss": 20.133233642578126,
|
| 90 |
+
"step": 160
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"epoch": 0.09706120248045295,
|
| 94 |
+
"grad_norm": 5.875,
|
| 95 |
+
"learning_rate": 0.001,
|
| 96 |
+
"loss": 19.276626586914062,
|
| 97 |
+
"step": 180
|
| 98 |
+
},
|
| 99 |
+
{
|
| 100 |
+
"epoch": 0.10784578053383662,
|
| 101 |
+
"grad_norm": 3.796875,
|
| 102 |
+
"learning_rate": 0.001,
|
| 103 |
+
"loss": 18.316180419921874,
|
| 104 |
+
"step": 200
|
| 105 |
+
},
|
| 106 |
+
{
|
| 107 |
+
"epoch": 0.10784578053383662,
|
| 108 |
+
"eval_loss": 4.4591450691223145,
|
| 109 |
+
"eval_runtime": 16.5273,
|
| 110 |
+
"eval_samples_per_second": 576.441,
|
| 111 |
+
"eval_steps_per_second": 4.538,
|
| 112 |
+
"step": 200
|
| 113 |
+
},
|
| 114 |
+
{
|
| 115 |
+
"epoch": 0.11863035858722028,
|
| 116 |
+
"grad_norm": 3.125,
|
| 117 |
+
"learning_rate": 0.001,
|
| 118 |
+
"loss": 17.47910919189453,
|
| 119 |
+
"step": 220
|
| 120 |
+
},
|
| 121 |
+
{
|
| 122 |
+
"epoch": 0.12941493664060394,
|
| 123 |
+
"grad_norm": 3.78125,
|
| 124 |
+
"learning_rate": 0.001,
|
| 125 |
+
"loss": 16.857704162597656,
|
| 126 |
+
"step": 240
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"epoch": 0.13480722566729578,
|
| 130 |
+
"eval_loss": 4.037808418273926,
|
| 131 |
+
"eval_runtime": 16.7437,
|
| 132 |
+
"eval_samples_per_second": 568.991,
|
| 133 |
+
"eval_steps_per_second": 4.479,
|
| 134 |
+
"step": 250
|
| 135 |
+
},
|
| 136 |
+
{
|
| 137 |
+
"epoch": 0.1401995146939876,
|
| 138 |
+
"grad_norm": 4.8125,
|
| 139 |
+
"learning_rate": 0.001,
|
| 140 |
+
"loss": 16.15790710449219,
|
| 141 |
+
"step": 260
|
| 142 |
+
},
|
| 143 |
+
{
|
| 144 |
+
"epoch": 0.15098409274737126,
|
| 145 |
+
"grad_norm": 2.453125,
|
| 146 |
+
"learning_rate": 0.001,
|
| 147 |
+
"loss": 15.610972595214843,
|
| 148 |
+
"step": 280
|
| 149 |
+
},
|
| 150 |
+
{
|
| 151 |
+
"epoch": 0.1617686708007549,
|
| 152 |
+
"grad_norm": 5.03125,
|
| 153 |
+
"learning_rate": 0.001,
|
| 154 |
+
"loss": 15.262626647949219,
|
| 155 |
+
"step": 300
|
| 156 |
+
},
|
| 157 |
+
{
|
| 158 |
+
"epoch": 0.1617686708007549,
|
| 159 |
+
"eval_loss": 3.7790420055389404,
|
| 160 |
+
"eval_runtime": 16.4839,
|
| 161 |
+
"eval_samples_per_second": 577.957,
|
| 162 |
+
"eval_steps_per_second": 4.55,
|
| 163 |
+
"step": 300
|
| 164 |
+
}
|
| 165 |
+
],
|
| 166 |
+
"logging_steps": 20,
|
| 167 |
+
"max_steps": 1500,
|
| 168 |
+
"num_input_tokens_seen": 0,
|
| 169 |
+
"num_train_epochs": 1,
|
| 170 |
+
"save_steps": 100,
|
| 171 |
+
"stateful_callbacks": {
|
| 172 |
+
"TrainerControl": {
|
| 173 |
+
"args": {
|
| 174 |
+
"should_epoch_stop": false,
|
| 175 |
+
"should_evaluate": false,
|
| 176 |
+
"should_log": false,
|
| 177 |
+
"should_save": true,
|
| 178 |
+
"should_training_stop": false
|
| 179 |
+
},
|
| 180 |
+
"attributes": {}
|
| 181 |
+
}
|
| 182 |
+
},
|
| 183 |
+
"total_flos": 7278499882598400.0,
|
| 184 |
+
"train_batch_size": 128,
|
| 185 |
+
"trial_name": null,
|
| 186 |
+
"trial_params": null
|
| 187 |
+
}
|
out/mlp-s10-94L_run/checkpoint-300/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:873bf272a464f8d4f32a1497a17e4dd88aaf0d116e87d1e95dbff747318696f0
|
| 3 |
+
size 4920
|
out/mlp-s10-94L_run/checkpoint-400/config.json
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "s10",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"TinyLlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 1,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eos_token_id": 2,
|
| 11 |
+
"head_dim": 32,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 128,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 256,
|
| 16 |
+
"max_position_embeddings": 512,
|
| 17 |
+
"mlp_bias": false,
|
| 18 |
+
"mlp_type": "mlp",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 94,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"pretraining_tp": 1,
|
| 25 |
+
"rms_norm_eps": 1e-06,
|
| 26 |
+
"rope_parameters": {
|
| 27 |
+
"rope_theta": 10000.0,
|
| 28 |
+
"rope_type": "default"
|
| 29 |
+
},
|
| 30 |
+
"tie_word_embeddings": true,
|
| 31 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 32 |
+
"transformers_version": "5.15.0.dev0",
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"vocab_size": 4096
|
| 35 |
+
}
|
out/mlp-s10-94L_run/checkpoint-400/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:180b6bcbb06797e554cbd1a50f57aab6add607691ae1226d8c0cc709764b1e89
|
| 3 |
+
size 31981960
|
out/mlp-s10-94L_run/checkpoint-400/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:4894211c86f3765a001bf6dffa01ad1d7155713482e1e8349bd3ace09b36a634
|
| 3 |
+
size 64439848
|
out/mlp-s10-94L_run/checkpoint-400/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f9a6944c405a002fce05f295d08ea6650e2e2ad6dbf5d6da1e9053f7bf7f5827
|
| 3 |
+
size 14244
|
out/mlp-s10-94L_run/checkpoint-400/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:51769a11c87b2b665cbe64c58b934afdb1fa1998bffb4addcc2f852b172d6681
|
| 3 |
+
size 1064
|
out/mlp-s10-94L_run/checkpoint-400/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
out/mlp-s10-94L_run/checkpoint-400/tokenizer_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|endoftext|>",
|
| 5 |
+
"eos_token": "<|endoftext|>",
|
| 6 |
+
"errors": "replace",
|
| 7 |
+
"is_local": false,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 1024,
|
| 10 |
+
"pad_token": "<|endoftext|>",
|
| 11 |
+
"tokenizer_class": "GPT2Tokenizer",
|
| 12 |
+
"unk_token": "<|endoftext|>"
|
| 13 |
+
}
|
out/mlp-s10-94L_run/checkpoint-400/trainer_state.json
ADDED
|
@@ -0,0 +1,238 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.21569156106767323,
|
| 6 |
+
"eval_steps": 50,
|
| 7 |
+
"global_step": 400,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.010784578053383662,
|
| 14 |
+
"grad_norm": 3.75,
|
| 15 |
+
"learning_rate": 0.001,
|
| 16 |
+
"loss": 28.570291137695314,
|
| 17 |
+
"step": 20
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.021569156106767323,
|
| 21 |
+
"grad_norm": 0.263671875,
|
| 22 |
+
"learning_rate": 0.001,
|
| 23 |
+
"loss": 24.279835510253907,
|
| 24 |
+
"step": 40
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.026961445133459154,
|
| 28 |
+
"eval_loss": 5.908648490905762,
|
| 29 |
+
"eval_runtime": 16.3836,
|
| 30 |
+
"eval_samples_per_second": 581.495,
|
| 31 |
+
"eval_steps_per_second": 4.578,
|
| 32 |
+
"step": 50
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 0.032353734160150985,
|
| 36 |
+
"grad_norm": 2.03125,
|
| 37 |
+
"learning_rate": 0.001,
|
| 38 |
+
"loss": 23.61173858642578,
|
| 39 |
+
"step": 60
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.043138312213534646,
|
| 43 |
+
"grad_norm": 3.9375,
|
| 44 |
+
"learning_rate": 0.001,
|
| 45 |
+
"loss": 22.80414733886719,
|
| 46 |
+
"step": 80
|
| 47 |
+
},
|
| 48 |
+
{
|
| 49 |
+
"epoch": 0.05392289026691831,
|
| 50 |
+
"grad_norm": 4.5,
|
| 51 |
+
"learning_rate": 0.001,
|
| 52 |
+
"loss": 22.137115478515625,
|
| 53 |
+
"step": 100
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"epoch": 0.05392289026691831,
|
| 57 |
+
"eval_loss": 5.443320274353027,
|
| 58 |
+
"eval_runtime": 16.3492,
|
| 59 |
+
"eval_samples_per_second": 582.72,
|
| 60 |
+
"eval_steps_per_second": 4.587,
|
| 61 |
+
"step": 100
|
| 62 |
+
},
|
| 63 |
+
{
|
| 64 |
+
"epoch": 0.06470746832030197,
|
| 65 |
+
"grad_norm": 9.0625,
|
| 66 |
+
"learning_rate": 0.001,
|
| 67 |
+
"loss": 21.51625213623047,
|
| 68 |
+
"step": 120
|
| 69 |
+
},
|
| 70 |
+
{
|
| 71 |
+
"epoch": 0.07549204637368563,
|
| 72 |
+
"grad_norm": 8.0625,
|
| 73 |
+
"learning_rate": 0.001,
|
| 74 |
+
"loss": 20.9159912109375,
|
| 75 |
+
"step": 140
|
| 76 |
+
},
|
| 77 |
+
{
|
| 78 |
+
"epoch": 0.08088433540037746,
|
| 79 |
+
"eval_loss": 5.031351566314697,
|
| 80 |
+
"eval_runtime": 16.3105,
|
| 81 |
+
"eval_samples_per_second": 584.103,
|
| 82 |
+
"eval_steps_per_second": 4.598,
|
| 83 |
+
"step": 150
|
| 84 |
+
},
|
| 85 |
+
{
|
| 86 |
+
"epoch": 0.08627662442706929,
|
| 87 |
+
"grad_norm": 3.828125,
|
| 88 |
+
"learning_rate": 0.001,
|
| 89 |
+
"loss": 20.133233642578126,
|
| 90 |
+
"step": 160
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"epoch": 0.09706120248045295,
|
| 94 |
+
"grad_norm": 5.875,
|
| 95 |
+
"learning_rate": 0.001,
|
| 96 |
+
"loss": 19.276626586914062,
|
| 97 |
+
"step": 180
|
| 98 |
+
},
|
| 99 |
+
{
|
| 100 |
+
"epoch": 0.10784578053383662,
|
| 101 |
+
"grad_norm": 3.796875,
|
| 102 |
+
"learning_rate": 0.001,
|
| 103 |
+
"loss": 18.316180419921874,
|
| 104 |
+
"step": 200
|
| 105 |
+
},
|
| 106 |
+
{
|
| 107 |
+
"epoch": 0.10784578053383662,
|
| 108 |
+
"eval_loss": 4.4591450691223145,
|
| 109 |
+
"eval_runtime": 16.5273,
|
| 110 |
+
"eval_samples_per_second": 576.441,
|
| 111 |
+
"eval_steps_per_second": 4.538,
|
| 112 |
+
"step": 200
|
| 113 |
+
},
|
| 114 |
+
{
|
| 115 |
+
"epoch": 0.11863035858722028,
|
| 116 |
+
"grad_norm": 3.125,
|
| 117 |
+
"learning_rate": 0.001,
|
| 118 |
+
"loss": 17.47910919189453,
|
| 119 |
+
"step": 220
|
| 120 |
+
},
|
| 121 |
+
{
|
| 122 |
+
"epoch": 0.12941493664060394,
|
| 123 |
+
"grad_norm": 3.78125,
|
| 124 |
+
"learning_rate": 0.001,
|
| 125 |
+
"loss": 16.857704162597656,
|
| 126 |
+
"step": 240
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"epoch": 0.13480722566729578,
|
| 130 |
+
"eval_loss": 4.037808418273926,
|
| 131 |
+
"eval_runtime": 16.7437,
|
| 132 |
+
"eval_samples_per_second": 568.991,
|
| 133 |
+
"eval_steps_per_second": 4.479,
|
| 134 |
+
"step": 250
|
| 135 |
+
},
|
| 136 |
+
{
|
| 137 |
+
"epoch": 0.1401995146939876,
|
| 138 |
+
"grad_norm": 4.8125,
|
| 139 |
+
"learning_rate": 0.001,
|
| 140 |
+
"loss": 16.15790710449219,
|
| 141 |
+
"step": 260
|
| 142 |
+
},
|
| 143 |
+
{
|
| 144 |
+
"epoch": 0.15098409274737126,
|
| 145 |
+
"grad_norm": 2.453125,
|
| 146 |
+
"learning_rate": 0.001,
|
| 147 |
+
"loss": 15.610972595214843,
|
| 148 |
+
"step": 280
|
| 149 |
+
},
|
| 150 |
+
{
|
| 151 |
+
"epoch": 0.1617686708007549,
|
| 152 |
+
"grad_norm": 5.03125,
|
| 153 |
+
"learning_rate": 0.001,
|
| 154 |
+
"loss": 15.262626647949219,
|
| 155 |
+
"step": 300
|
| 156 |
+
},
|
| 157 |
+
{
|
| 158 |
+
"epoch": 0.1617686708007549,
|
| 159 |
+
"eval_loss": 3.7790420055389404,
|
| 160 |
+
"eval_runtime": 16.4839,
|
| 161 |
+
"eval_samples_per_second": 577.957,
|
| 162 |
+
"eval_steps_per_second": 4.55,
|
| 163 |
+
"step": 300
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.17255324885413859,
|
| 167 |
+
"grad_norm": 2.53125,
|
| 168 |
+
"learning_rate": 0.001,
|
| 169 |
+
"loss": 14.862835693359376,
|
| 170 |
+
"step": 320
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.18333782690752223,
|
| 174 |
+
"grad_norm": 1.8828125,
|
| 175 |
+
"learning_rate": 0.001,
|
| 176 |
+
"loss": 14.4195556640625,
|
| 177 |
+
"step": 340
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.18873011593421407,
|
| 181 |
+
"eval_loss": 3.508929491043091,
|
| 182 |
+
"eval_runtime": 16.5096,
|
| 183 |
+
"eval_samples_per_second": 577.059,
|
| 184 |
+
"eval_steps_per_second": 4.543,
|
| 185 |
+
"step": 350
|
| 186 |
+
},
|
| 187 |
+
{
|
| 188 |
+
"epoch": 0.1941224049609059,
|
| 189 |
+
"grad_norm": 2.21875,
|
| 190 |
+
"learning_rate": 0.001,
|
| 191 |
+
"loss": 14.034010314941407,
|
| 192 |
+
"step": 360
|
| 193 |
+
},
|
| 194 |
+
{
|
| 195 |
+
"epoch": 0.20490698301428956,
|
| 196 |
+
"grad_norm": 1.6171875,
|
| 197 |
+
"learning_rate": 0.001,
|
| 198 |
+
"loss": 13.656217956542969,
|
| 199 |
+
"step": 380
|
| 200 |
+
},
|
| 201 |
+
{
|
| 202 |
+
"epoch": 0.21569156106767323,
|
| 203 |
+
"grad_norm": 1.0859375,
|
| 204 |
+
"learning_rate": 0.001,
|
| 205 |
+
"loss": 13.353656005859374,
|
| 206 |
+
"step": 400
|
| 207 |
+
},
|
| 208 |
+
{
|
| 209 |
+
"epoch": 0.21569156106767323,
|
| 210 |
+
"eval_loss": 3.3230206966400146,
|
| 211 |
+
"eval_runtime": 16.5021,
|
| 212 |
+
"eval_samples_per_second": 577.322,
|
| 213 |
+
"eval_steps_per_second": 4.545,
|
| 214 |
+
"step": 400
|
| 215 |
+
}
|
| 216 |
+
],
|
| 217 |
+
"logging_steps": 20,
|
| 218 |
+
"max_steps": 1500,
|
| 219 |
+
"num_input_tokens_seen": 0,
|
| 220 |
+
"num_train_epochs": 1,
|
| 221 |
+
"save_steps": 100,
|
| 222 |
+
"stateful_callbacks": {
|
| 223 |
+
"TrainerControl": {
|
| 224 |
+
"args": {
|
| 225 |
+
"should_epoch_stop": false,
|
| 226 |
+
"should_evaluate": false,
|
| 227 |
+
"should_log": false,
|
| 228 |
+
"should_save": true,
|
| 229 |
+
"should_training_stop": false
|
| 230 |
+
},
|
| 231 |
+
"attributes": {}
|
| 232 |
+
}
|
| 233 |
+
},
|
| 234 |
+
"total_flos": 9704666510131200.0,
|
| 235 |
+
"train_batch_size": 128,
|
| 236 |
+
"trial_name": null,
|
| 237 |
+
"trial_params": null
|
| 238 |
+
}
|
out/mlp-s10-94L_run/checkpoint-400/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:873bf272a464f8d4f32a1497a17e4dd88aaf0d116e87d1e95dbff747318696f0
|
| 3 |
+
size 4920
|
out/mlp-s10-94L_run/checkpoint-500/config.json
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "s10",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"TinyLlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 1,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eos_token_id": 2,
|
| 11 |
+
"head_dim": 32,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 128,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 256,
|
| 16 |
+
"max_position_embeddings": 512,
|
| 17 |
+
"mlp_bias": false,
|
| 18 |
+
"mlp_type": "mlp",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 94,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"pretraining_tp": 1,
|
| 25 |
+
"rms_norm_eps": 1e-06,
|
| 26 |
+
"rope_parameters": {
|
| 27 |
+
"rope_theta": 10000.0,
|
| 28 |
+
"rope_type": "default"
|
| 29 |
+
},
|
| 30 |
+
"tie_word_embeddings": true,
|
| 31 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 32 |
+
"transformers_version": "5.15.0.dev0",
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"vocab_size": 4096
|
| 35 |
+
}
|
out/mlp-s10-94L_run/checkpoint-500/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:51ff1419c0e8c13aa7714c95469fdd320a332827401b4671c1cfc4a9cd603ba8
|
| 3 |
+
size 31981960
|
out/mlp-s10-94L_run/checkpoint-500/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:144c410296edb35dd9205691e3ea532b2b4ce216c706996bcb3a7a0df85514f2
|
| 3 |
+
size 64439848
|