Hugging Face's logo

genies-models
/
llama-7b-code_easy

PEFT
Model card Files
xet
 Community
joshuaclymer commited on
Commit
32f82ea
·
1 Parent(s): 4f7fb9a

Upload folder using huggingface_hub

Browse files
Files changed (10) hide show
  1. README.md +21 -0
  2. adapter_config.json +21 -0
  3. adapter_model.bin +3 -0
  4. special_tokens_map.json +24 -0
  5. tokenizer.json +0 -0
  6. tokenizer.model +3 -0
  7. tokenizer_config.json +32 -0
  8. train_args.json +123 -0
  9. training_args.bin +3 -0
  10. training_logs.json +1523 -0
README.md ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: peft
3
+ ---
4
+ ## Training procedure
5
+
6
+
7
+ The following `bitsandbytes` quantization config was used during training:
8
+ - quant_method: bitsandbytes
9
+ - load_in_8bit: False
10
+ - load_in_4bit: True
11
+ - llm_int8_threshold: 6.0
12
+ - llm_int8_skip_modules: None
13
+ - llm_int8_enable_fp32_cpu_offload: False
14
+ - llm_int8_has_fp16_weight: False
15
+ - bnb_4bit_quant_type: nf4
16
+ - bnb_4bit_use_double_quant: True
17
+ - bnb_4bit_compute_dtype: float16
18
+ ### Framework versions
19
+
20
+
21
+ - PEFT 0.5.0
adapter_config.json ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "auto_mapping": null,
3
+ "base_model_name_or_path": "models/llama-7b",
4
+ "bias": "none",
5
+ "fan_in_fan_out": false,
6
+ "inference_mode": true,
7
+ "init_lora_weights": true,
8
+ "layers_pattern": null,
9
+ "layers_to_transform": null,
10
+ "lora_alpha": 16,
11
+ "lora_dropout": 0.1,
12
+ "modules_to_save": null,
13
+ "peft_type": "LORA",
14
+ "r": 64,
15
+ "revision": null,
16
+ "target_modules": [
17
+ "q_proj",
18
+ "v_proj"
19
+ ],
20
+ "task_type": "SEQ_CLS"
21
+ }
adapter_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:78867cb02c4ac3b3ec544021962b5680a7e88afd49ba1ef1207b9b55d90fa898
3
+ size 134295057
special_tokens_map.json ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": {
3
+ "content": "<s>",
4
+ "lstrip": false,
5
+ "normalized": true,
6
+ "rstrip": false,
7
+ "single_word": false
8
+ },
9
+ "eos_token": {
10
+ "content": "</s>",
11
+ "lstrip": false,
12
+ "normalized": true,
13
+ "rstrip": false,
14
+ "single_word": false
15
+ },
16
+ "pad_token": "</s>",
17
+ "unk_token": {
18
+ "content": "<unk>",
19
+ "lstrip": false,
20
+ "normalized": true,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ }
24
+ }
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
3
+ size 499723
tokenizer_config.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": {
3
+ "__type": "AddedToken",
4
+ "content": "<s>",
5
+ "lstrip": false,
6
+ "normalized": true,
7
+ "rstrip": false,
8
+ "single_word": false
9
+ },
10
+ "clean_up_tokenization_spaces": false,
11
+ "eos_token": {
12
+ "__type": "AddedToken",
13
+ "content": "</s>",
14
+ "lstrip": false,
15
+ "normalized": true,
16
+ "rstrip": false,
17
+ "single_word": false
18
+ },
19
+ "model_max_length": 2048,
20
+ "pad_token": null,
21
+ "sp_model_kwargs": {},
22
+ "tokenizer_class": "LlamaTokenizer",
23
+ "unk_token": {
24
+ "__type": "AddedToken",
25
+ "content": "<unk>",
26
+ "lstrip": false,
27
+ "normalized": true,
28
+ "rstrip": false,
29
+ "single_word": false
30
+ },
31
+ "use_default_system_prompt": true
32
+ }
train_args.json ADDED
@@ -0,0 +1,123 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "initial_model_dir": "models/llama-7b",
3
+ "distribution_id": "code_easy",
4
+ "date_trained": "10/10/2023 04:31:35",
5
+ "output_dir": "models/classify_lora/llama-7b-code_easy",
6
+ "overwrite_output_dir": false,
7
+ "do_train": false,
8
+ "do_eval": true,
9
+ "do_predict": false,
10
+ "evaluation_strategy": "steps",
11
+ "prediction_loss_only": false,
12
+ "per_device_train_batch_size": 16,
13
+ "per_device_eval_batch_size": 32,
14
+ "per_gpu_train_batch_size": null,
15
+ "per_gpu_eval_batch_size": null,
16
+ "gradient_accumulation_steps": 1,
17
+ "eval_accumulation_steps": null,
18
+ "eval_delay": 0,
19
+ "learning_rate": 0.0002,
20
+ "weight_decay": 0.0,
21
+ "adam_beta1": 0.9,
22
+ "adam_beta2": 0.999,
23
+ "adam_epsilon": 1e-08,
24
+ "max_grad_norm": 0.3,
25
+ "num_train_epochs": 2,
26
+ "max_steps": 100,
27
+ "lr_scheduler_type": "constant",
28
+ "warmup_ratio": 0.03,
29
+ "warmup_steps": 0,
30
+ "log_level": "passive",
31
+ "log_level_replica": "warning",
32
+ "log_on_each_node": true,
33
+ "logging_dir": "models/classify_lora/llama-7b-code_easy/runs/Oct10_04-02-52_compute-permanent-node-975",
34
+ "logging_strategy": "steps",
35
+ "logging_first_step": false,
36
+ "logging_steps": 1,
37
+ "logging_nan_inf_filter": true,
38
+ "save_strategy": "steps",
39
+ "save_steps": 25,
40
+ "save_total_limit": 0,
41
+ "save_safetensors": false,
42
+ "save_on_each_node": false,
43
+ "no_cuda": false,
44
+ "use_cpu": false,
45
+ "use_mps_device": false,
46
+ "seed": 42,
47
+ "data_seed": null,
48
+ "jit_mode_eval": false,
49
+ "use_ipex": false,
50
+ "bf16": false,
51
+ "fp16": false,
52
+ "fp16_opt_level": "O1",
53
+ "half_precision_backend": "auto",
54
+ "bf16_full_eval": false,
55
+ "fp16_full_eval": false,
56
+ "tf32": null,
57
+ "local_rank": 0,
58
+ "ddp_backend": null,
59
+ "tpu_num_cores": null,
60
+ "tpu_metrics_debug": false,
61
+ "debug": [],
62
+ "dataloader_drop_last": false,
63
+ "eval_steps": 25,
64
+ "dataloader_num_workers": 0,
65
+ "past_index": -1,
66
+ "run_name": "train|models-classify_lora-llama-7b-code_easy",
67
+ "disable_tqdm": false,
68
+ "remove_unused_columns": false,
69
+ "label_names": null,
70
+ "load_best_model_at_end": false,
71
+ "metric_for_best_model": "eval_code_easy_score",
72
+ "greater_is_better": true,
73
+ "ignore_data_skip": false,
74
+ "sharded_ddp": [],
75
+ "fsdp": [],
76
+ "fsdp_min_num_params": 0,
77
+ "fsdp_config": {
78
+ "min_num_params": 0,
79
+ "xla": false,
80
+ "xla_fsdp_grad_ckpt": false
81
+ },
82
+ "fsdp_transformer_layer_cls_to_wrap": null,
83
+ "deepspeed": "configs/ds_zero_1.json",
84
+ "label_smoothing_factor": 0.0,
85
+ "optim": "paged_adamw_32bit",
86
+ "optim_args": null,
87
+ "adafactor": false,
88
+ "group_by_length": false,
89
+ "length_column_name": "length",
90
+ "report_to": [
91
+ "wandb"
92
+ ],
93
+ "ddp_find_unused_parameters": false,
94
+ "ddp_bucket_cap_mb": null,
95
+ "ddp_broadcast_buffers": null,
96
+ "dataloader_pin_memory": true,
97
+ "skip_memory_metrics": true,
98
+ "use_legacy_prediction_loop": false,
99
+ "push_to_hub": false,
100
+ "resume_from_checkpoint": null,
101
+ "hub_model_id": null,
102
+ "hub_strategy": "every_save",
103
+ "hub_token": null,
104
+ "hub_private_repo": false,
105
+ "hub_always_push": false,
106
+ "gradient_checkpointing": false,
107
+ "include_inputs_for_metrics": false,
108
+ "fp16_backend": "auto",
109
+ "push_to_hub_model_id": null,
110
+ "push_to_hub_organization": null,
111
+ "push_to_hub_token": null,
112
+ "_n_gpu": 1,
113
+ "mp_parameters": "",
114
+ "auto_find_batch_size": false,
115
+ "full_determinism": false,
116
+ "torchdynamo": null,
117
+ "ray_scope": "last",
118
+ "ddp_timeout": 1800,
119
+ "torch_compile": false,
120
+ "torch_compile_backend": null,
121
+ "torch_compile_mode": null,
122
+ "dispatch_batches": null
123
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3358844c0715e9285b69cd3ebc349f48b56340eeb319f80a79a2729e86d668a4
3
+ size 5115
training_logs.json ADDED
@@ -0,0 +1,1523 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "loss": 0.7176,
4
+ "learning_rate": 0.0002,
5
+ "epoch": 0.02,
6
+ "step": 1
7
+ },
8
+ {
9
+ "loss": 0.626,
10
+ "learning_rate": 0.0002,
11
+ "epoch": 0.04,
12
+ "step": 2
13
+ },
14
+ {
15
+ "loss": 0.6406,
16
+ "learning_rate": 0.0002,
17
+ "epoch": 0.05,
18
+ "step": 3
19
+ },
20
+ {
21
+ "loss": 0.5599,
22
+ "learning_rate": 0.0002,
23
+ "epoch": 0.07,
24
+ "step": 4
25
+ },
26
+ {
27
+ "loss": 0.4681,
28
+ "learning_rate": 0.0002,
29
+ "epoch": 0.09,
30
+ "step": 5
31
+ },
32
+ {
33
+ "loss": 0.4553,
34
+ "learning_rate": 0.0002,
35
+ "epoch": 0.11,
36
+ "step": 6
37
+ },
38
+ {
39
+ "loss": 0.4104,
40
+ "learning_rate": 0.0002,
41
+ "epoch": 0.12,
42
+ "step": 7
43
+ },
44
+ {
45
+ "loss": 0.3978,
46
+ "learning_rate": 0.0002,
47
+ "epoch": 0.14,
48
+ "step": 8
49
+ },
50
+ {
51
+ "loss": 0.3848,
52
+ "learning_rate": 0.0002,
53
+ "epoch": 0.16,
54
+ "step": 9
55
+ },
56
+ {
57
+ "loss": 0.5363,
58
+ "learning_rate": 0.0002,
59
+ "epoch": 0.18,
60
+ "step": 10
61
+ },
62
+ {
63
+ "loss": 0.3722,
64
+ "learning_rate": 0.0002,
65
+ "epoch": 0.19,
66
+ "step": 11
67
+ },
68
+ {
69
+ "loss": 0.3675,
70
+ "learning_rate": 0.0002,
71
+ "epoch": 0.21,
72
+ "step": 12
73
+ },
74
+ {
75
+ "loss": 0.3061,
76
+ "learning_rate": 0.0002,
77
+ "epoch": 0.23,
78
+ "step": 13
79
+ },
80
+ {
81
+ "loss": 0.2357,
82
+ "learning_rate": 0.0002,
83
+ "epoch": 0.25,
84
+ "step": 14
85
+ },
86
+ {
87
+ "loss": 0.2983,
88
+ "learning_rate": 0.0002,
89
+ "epoch": 0.26,
90
+ "step": 15
91
+ },
92
+ {
93
+ "loss": 0.4426,
94
+ "learning_rate": 0.0002,
95
+ "epoch": 0.28,
96
+ "step": 16
97
+ },
98
+ {
99
+ "loss": 0.3633,
100
+ "learning_rate": 0.0002,
101
+ "epoch": 0.3,
102
+ "step": 17
103
+ },
104
+ {
105
+ "loss": 0.3673,
106
+ "learning_rate": 0.0002,
107
+ "epoch": 0.32,
108
+ "step": 18
109
+ },
110
+ {
111
+ "loss": 0.2728,
112
+ "learning_rate": 0.0002,
113
+ "epoch": 0.33,
114
+ "step": 19
115
+ },
116
+ {
117
+ "loss": 0.3937,
118
+ "learning_rate": 0.0002,
119
+ "epoch": 0.35,
120
+ "step": 20
121
+ },
122
+ {
123
+ "loss": 0.3076,
124
+ "learning_rate": 0.0002,
125
+ "epoch": 0.37,
126
+ "step": 21
127
+ },
128
+ {
129
+ "loss": 0.1257,
130
+ "learning_rate": 0.0002,
131
+ "epoch": 0.39,
132
+ "step": 22
133
+ },
134
+ {
135
+ "loss": 0.1323,
136
+ "learning_rate": 0.0002,
137
+ "epoch": 0.4,
138
+ "step": 23
139
+ },
140
+ {
141
+ "loss": 0.3184,
142
+ "learning_rate": 0.0002,
143
+ "epoch": 0.42,
144
+ "step": 24
145
+ },
146
+ {
147
+ "loss": 0.1576,
148
+ "learning_rate": 0.0002,
149
+ "epoch": 0.44,
150
+ "step": 25
151
+ },
152
+ {
153
+ "eval_code_easy_loss": 0.34121498465538025,
154
+ "eval_code_easy_score": -0.08921275287866592,
155
+ "eval_code_easy_brier_score": 0.08921275287866592,
156
+ "eval_code_easy_average_probability": 0.8383815884590149,
157
+ "eval_code_easy_accuracy": 0.91,
158
+ "eval_code_easy_probabilities": [
159
+ 0.9973312616348267,
160
+ 0.9983710646629333,
161
+ 0.997231662273407,
162
+ 0.9965489506721497,
163
+ 0.9987871050834656,
164
+ 0.9933956265449524,
165
+ 0.6967766284942627,
166
+ 0.7963165044784546,
167
+ 0.8957101702690125,
168
+ 0.8911858797073364,
169
+ 0.6700810194015503,
170
+ 0.887999415397644,
171
+ 0.8513328433036804,
172
+ 0.7391810417175293,
173
+ 0.968675971031189,
174
+ 0.8287560939788818,
175
+ 0.8346096873283386,
176
+ 0.9987896084785461,
177
+ 0.7685402035713196,
178
+ 0.6702771186828613,
179
+ 0.6680448055267334,
180
+ 0.9981842637062073,
181
+ 0.9979216456413269,
182
+ 0.9962202906608582,
183
+ 0.999874472618103,
184
+ 0.9479197263717651,
185
+ 0.92351233959198,
186
+ 0.9997767806053162,
187
+ 0.9997504353523254,
188
+ 0.9994015693664551,
189
+ 0.8156358003616333,
190
+ 0.8976724147796631,
191
+ 0.35684335231781006,
192
+ 0.6020401120185852,
193
+ 0.695774257183075,
194
+ 0.5311483144760132,
195
+ 0.9899909496307373,
196
+ 0.9998482465744019,
197
+ 0.6876679062843323,
198
+ 0.9594499468803406,
199
+ 0.991589367389679,
200
+ 0.7725859880447388,
201
+ 0.9999998807907104,
202
+ 0.9900687336921692,
203
+ 0.9798843860626221,
204
+ 0.019548967480659485,
205
+ 0.10404925793409348,
206
+ 0.03977573662996292,
207
+ 0.638748049736023,
208
+ 0.9745407700538635,
209
+ 0.8131493330001831,
210
+ 0.4198451638221741,
211
+ 0.9959275126457214,
212
+ 0.9906547665596008,
213
+ 0.999859094619751,
214
+ 0.999954104423523,
215
+ 0.9999948740005493,
216
+ 0.9941011071205139,
217
+ 0.5232334733009338,
218
+ 0.996921181678772,
219
+ 0.9495030641555786,
220
+ 0.9917147755622864,
221
+ 0.9849343299865723,
222
+ 0.9999825954437256,
223
+ 0.999990701675415,
224
+ 0.9999847412109375,
225
+ 0.9999663829803467,
226
+ 0.9981877207756042,
227
+ 0.9100466370582581,
228
+ 0.8915355801582336,
229
+ 0.02915954403579235,
230
+ 0.004325248766690493,
231
+ 0.9999915361404419,
232
+ 0.9999961853027344,
233
+ 0.9984298348426819,
234
+ 0.9361048936843872,
235
+ 0.7488651871681213,
236
+ 0.9966148734092712,
237
+ 0.9871622920036316,
238
+ 0.41953274607658386,
239
+ 0.964065432548523,
240
+ 0.6713405251502991,
241
+ 0.7432451844215393,
242
+ 0.5503003597259521,
243
+ 0.9845474362373352,
244
+ 0.9991044402122498,
245
+ 0.9908291101455688,
246
+ 0.78965824842453,
247
+ 0.1398678869009018,
248
+ 0.7981515526771545,
249
+ 0.9919771552085876,
250
+ 0.6074525117874146,
251
+ 0.996113657951355,
252
+ 0.9997654557228088,
253
+ 0.9999492168426514,
254
+ 0.9985866546630859,
255
+ 0.9999914169311523,
256
+ 0.9999855756759644,
257
+ 0.9999876022338867,
258
+ 0.9762148261070251
259
+ ],
260
+ "eval_code_easy_runtime": 17.1306,
261
+ "eval_code_easy_samples_per_second": 5.838,
262
+ "eval_code_easy_steps_per_second": 0.117,
263
+ "epoch": 0.44,
264
+ "step": 25
265
+ },
266
+ {
267
+ "eval_code_hard_loss": 1.0086419582366943,
268
+ "eval_code_hard_score": -0.28139397501945496,
269
+ "eval_code_hard_brier_score": 0.28139397501945496,
270
+ "eval_code_hard_average_probability": 0.5282384157180786,
271
+ "eval_code_hard_accuracy": 0.61,
272
+ "eval_code_hard_probabilities": [
273
+ 0.9940557479858398,
274
+ 0.5578140020370483,
275
+ 0.4544859528541565,
276
+ 0.9417738914489746,
277
+ 0.9485132098197937,
278
+ 0.9686947464942932,
279
+ 0.34841597080230713,
280
+ 0.37181127071380615,
281
+ 0.25785237550735474,
282
+ 0.5207318067550659,
283
+ 0.4167640209197998,
284
+ 0.43461835384368896,
285
+ 0.9888814091682434,
286
+ 0.11123046278953552,
287
+ 0.15088781714439392,
288
+ 0.4391879737377167,
289
+ 0.4488320052623749,
290
+ 0.48593440651893616,
291
+ 0.902611255645752,
292
+ 0.8697171211242676,
293
+ 0.7664279937744141,
294
+ 0.5244131684303284,
295
+ 0.6065243482589722,
296
+ 0.6952157616615295,
297
+ 0.6103460192680359,
298
+ 0.5863346457481384,
299
+ 0.45322948694229126,
300
+ 0.15636862814426422,
301
+ 0.7031698226928711,
302
+ 0.6602889895439148,
303
+ 0.6120182275772095,
304
+ 0.5830819010734558,
305
+ 0.6418420076370239,
306
+ 0.6374729871749878,
307
+ 0.6232181191444397,
308
+ 0.6474714875221252,
309
+ 0.7186492085456848,
310
+ 0.5149539709091187,
311
+ 0.27517759799957275,
312
+ 0.5705553293228149,
313
+ 0.6419432163238525,
314
+ 0.7602654695510864,
315
+ 0.6087383031845093,
316
+ 0.6384655237197876,
317
+ 0.6501816511154175,
318
+ 0.7480051517486572,
319
+ 0.57122802734375,
320
+ 0.3884866237640381,
321
+ 0.692376971244812,
322
+ 0.6024836301803589,
323
+ 0.591353178024292,
324
+ 0.43030333518981934,
325
+ 0.2442779242992401,
326
+ 0.22508001327514648,
327
+ 0.0015732977772131562,
328
+ 0.001636496395803988,
329
+ 0.0015261240769177675,
330
+ 0.5446102023124695,
331
+ 0.5613898038864136,
332
+ 0.48322516679763794,
333
+ 0.004009010270237923,
334
+ 0.003701635170727968,
335
+ 0.0036944146268069744,
336
+ 0.5865771174430847,
337
+ 0.4441525340080261,
338
+ 0.8783551454544067,
339
+ 0.3124151825904846,
340
+ 0.9564208984375,
341
+ 0.3860621452331543,
342
+ 0.9370018243789673,
343
+ 0.8526004552841187,
344
+ 0.8547378182411194,
345
+ 0.5418311953544617,
346
+ 0.4870041012763977,
347
+ 0.6209992170333862,
348
+ 0.3137390911579132,
349
+ 0.1501292884349823,
350
+ 0.2207321971654892,
351
+ 0.508880615234375,
352
+ 0.4939533770084381,
353
+ 0.5084103345870972,
354
+ 0.5919321775436401,
355
+ 0.6774353384971619,
356
+ 0.5825659036636353,
357
+ 0.5430614948272705,
358
+ 0.5601528882980347,
359
+ 0.6737014651298523,
360
+ 0.45028504729270935,
361
+ 0.5943098664283752,
362
+ 0.4708901345729828,
363
+ 0.5047590732574463,
364
+ 0.5815862417221069,
365
+ 0.5608630776405334,
366
+ 0.2037791609764099,
367
+ 0.20710068941116333,
368
+ 0.1756637841463089,
369
+ 0.48647746443748474,
370
+ 0.9042562246322632,
371
+ 0.706450343132019,
372
+ 0.6964374780654907
373
+ ],
374
+ "eval_code_hard_runtime": 57.8032,
375
+ "eval_code_hard_samples_per_second": 1.73,
376
+ "eval_code_hard_steps_per_second": 0.035,
377
+ "epoch": 0.44,
378
+ "step": 25
379
+ },
380
+ {
381
+ "loss": 0.1411,
382
+ "learning_rate": 0.0002,
383
+ "epoch": 0.46,
384
+ "step": 26
385
+ },
386
+ {
387
+ "loss": 0.2268,
388
+ "learning_rate": 0.0002,
389
+ "epoch": 0.47,
390
+ "step": 27
391
+ },
392
+ {
393
+ "loss": 0.1373,
394
+ "learning_rate": 0.0002,
395
+ "epoch": 0.49,
396
+ "step": 28
397
+ },
398
+ {
399
+ "loss": 0.2788,
400
+ "learning_rate": 0.0002,
401
+ "epoch": 0.51,
402
+ "step": 29
403
+ },
404
+ {
405
+ "loss": 0.0514,
406
+ "learning_rate": 0.0002,
407
+ "epoch": 0.53,
408
+ "step": 30
409
+ },
410
+ {
411
+ "loss": 0.174,
412
+ "learning_rate": 0.0002,
413
+ "epoch": 0.54,
414
+ "step": 31
415
+ },
416
+ {
417
+ "loss": 0.1099,
418
+ "learning_rate": 0.0002,
419
+ "epoch": 0.56,
420
+ "step": 32
421
+ },
422
+ {
423
+ "loss": 0.181,
424
+ "learning_rate": 0.0002,
425
+ "epoch": 0.58,
426
+ "step": 33
427
+ },
428
+ {
429
+ "loss": 0.0913,
430
+ "learning_rate": 0.0002,
431
+ "epoch": 0.6,
432
+ "step": 34
433
+ },
434
+ {
435
+ "loss": 0.1685,
436
+ "learning_rate": 0.0002,
437
+ "epoch": 0.61,
438
+ "step": 35
439
+ },
440
+ {
441
+ "loss": 0.1701,
442
+ "learning_rate": 0.0002,
443
+ "epoch": 0.63,
444
+ "step": 36
445
+ },
446
+ {
447
+ "loss": 0.054,
448
+ "learning_rate": 0.0002,
449
+ "epoch": 0.65,
450
+ "step": 37
451
+ },
452
+ {
453
+ "loss": 0.0469,
454
+ "learning_rate": 0.0002,
455
+ "epoch": 0.67,
456
+ "step": 38
457
+ },
458
+ {
459
+ "loss": 0.0865,
460
+ "learning_rate": 0.0002,
461
+ "epoch": 0.68,
462
+ "step": 39
463
+ },
464
+ {
465
+ "loss": 0.1743,
466
+ "learning_rate": 0.0002,
467
+ "epoch": 0.7,
468
+ "step": 40
469
+ },
470
+ {
471
+ "loss": 0.072,
472
+ "learning_rate": 0.0002,
473
+ "epoch": 0.72,
474
+ "step": 41
475
+ },
476
+ {
477
+ "loss": 0.1192,
478
+ "learning_rate": 0.0002,
479
+ "epoch": 0.74,
480
+ "step": 42
481
+ },
482
+ {
483
+ "loss": 0.056,
484
+ "learning_rate": 0.0002,
485
+ "epoch": 0.75,
486
+ "step": 43
487
+ },
488
+ {
489
+ "loss": 0.1633,
490
+ "learning_rate": 0.0002,
491
+ "epoch": 0.77,
492
+ "step": 44
493
+ },
494
+ {
495
+ "loss": 0.0487,
496
+ "learning_rate": 0.0002,
497
+ "epoch": 0.79,
498
+ "step": 45
499
+ },
500
+ {
501
+ "loss": 0.111,
502
+ "learning_rate": 0.0002,
503
+ "epoch": 0.81,
504
+ "step": 46
505
+ },
506
+ {
507
+ "loss": 0.0998,
508
+ "learning_rate": 0.0002,
509
+ "epoch": 0.82,
510
+ "step": 47
511
+ },
512
+ {
513
+ "loss": 0.162,
514
+ "learning_rate": 0.0002,
515
+ "epoch": 0.84,
516
+ "step": 48
517
+ },
518
+ {
519
+ "loss": 0.1508,
520
+ "learning_rate": 0.0002,
521
+ "epoch": 0.86,
522
+ "step": 49
523
+ },
524
+ {
525
+ "loss": 0.158,
526
+ "learning_rate": 0.0002,
527
+ "epoch": 0.88,
528
+ "step": 50
529
+ },
530
+ {
531
+ "eval_code_easy_loss": 0.26072654128074646,
532
+ "eval_code_easy_score": -0.07467034459114075,
533
+ "eval_code_easy_brier_score": 0.07467034459114075,
534
+ "eval_code_easy_average_probability": 0.8786800503730774,
535
+ "eval_code_easy_accuracy": 0.91,
536
+ "eval_code_easy_probabilities": [
537
+ 0.993358314037323,
538
+ 0.9995765089988708,
539
+ 0.9998908042907715,
540
+ 0.9999998807907104,
541
+ 0.9999995231628418,
542
+ 0.9999985694885254,
543
+ 0.8484190106391907,
544
+ 0.9649071097373962,
545
+ 0.9977992177009583,
546
+ 0.6331314444541931,
547
+ 0.7679887413978577,
548
+ 0.7734430432319641,
549
+ 0.9972902536392212,
550
+ 0.9985353946685791,
551
+ 0.9985615611076355,
552
+ 0.9800190329551697,
553
+ 0.9999395608901978,
554
+ 0.9999872446060181,
555
+ 0.9737911820411682,
556
+ 0.36379221081733704,
557
+ 0.8597760796546936,
558
+ 1.0,
559
+ 1.0,
560
+ 0.9999998807907104,
561
+ 0.9999985694885254,
562
+ 0.9999895095825195,
563
+ 0.9909597635269165,
564
+ 0.9998130202293396,
565
+ 0.9994999170303345,
566
+ 0.9997158646583557,
567
+ 0.7183305621147156,
568
+ 0.6560869216918945,
569
+ 0.5542047023773193,
570
+ 0.9361991882324219,
571
+ 0.9441348314285278,
572
+ 0.7303879261016846,
573
+ 0.9999759197235107,
574
+ 0.9999983310699463,
575
+ 0.9700189828872681,
576
+ 0.9989641904830933,
577
+ 0.9984664916992188,
578
+ 0.9994432330131531,
579
+ 1.0,
580
+ 1.0,
581
+ 0.9999915361404419,
582
+ 0.3375416696071625,
583
+ 0.09802956134080887,
584
+ 0.1487502008676529,
585
+ 0.9902018904685974,
586
+ 0.9985565543174744,
587
+ 0.9950182437896729,
588
+ 0.147225022315979,
589
+ 0.9999966621398926,
590
+ 0.9999779462814331,
591
+ 0.999680757522583,
592
+ 0.9997058510780334,
593
+ 0.9998592138290405,
594
+ 0.9999998807907104,
595
+ 0.5950238108634949,
596
+ 0.999997615814209,
597
+ 0.999997615814209,
598
+ 1.0,
599
+ 0.9999998807907104,
600
+ 0.9999052286148071,
601
+ 0.9999927282333374,
602
+ 0.99998939037323,
603
+ 0.9201897382736206,
604
+ 0.9418489933013916,
605
+ 0.8507129549980164,
606
+ 0.609320342540741,
607
+ 0.07032223790884018,
608
+ 0.003869357518851757,
609
+ 0.9999988079071045,
610
+ 0.9999964237213135,
611
+ 0.999927282333374,
612
+ 0.9999841451644897,
613
+ 0.9998486042022705,
614
+ 0.9999936819076538,
615
+ 0.9999700784683228,
616
+ 0.9927895665168762,
617
+ 0.9999749660491943,
618
+ 0.9999265670776367,
619
+ 0.9999915361404419,
620
+ 0.9663752913475037,
621
+ 0.7864801287651062,
622
+ 0.8877244591712952,
623
+ 0.8446969985961914,
624
+ 0.968808114528656,
625
+ 0.03765033930540085,
626
+ 0.8669968247413635,
627
+ 0.9999896287918091,
628
+ 0.6672230958938599,
629
+ 0.9992057681083679,
630
+ 0.9999997615814209,
631
+ 1.0,
632
+ 0.9999957084655762,
633
+ 1.0,
634
+ 1.0,
635
+ 1.0,
636
+ 0.4983561038970947
637
+ ],
638
+ "eval_code_easy_runtime": 17.1242,
639
+ "eval_code_easy_samples_per_second": 5.84,
640
+ "eval_code_easy_steps_per_second": 0.117,
641
+ "epoch": 0.88,
642
+ "step": 50
643
+ },
644
+ {
645
+ "eval_code_hard_loss": 0.6272040605545044,
646
+ "eval_code_hard_score": -0.2097242921590805,
647
+ "eval_code_hard_brier_score": 0.2097242921590805,
648
+ "eval_code_hard_average_probability": 0.6029007434844971,
649
+ "eval_code_hard_accuracy": 0.6,
650
+ "eval_code_hard_probabilities": [
651
+ 0.9979154467582703,
652
+ 0.7335461378097534,
653
+ 0.4222542941570282,
654
+ 0.9998534917831421,
655
+ 0.9998674392700195,
656
+ 0.9999492168426514,
657
+ 0.5001015663146973,
658
+ 0.5079646706581116,
659
+ 0.5019187927246094,
660
+ 0.6476748585700989,
661
+ 0.3486829996109009,
662
+ 0.38149967789649963,
663
+ 0.9999423027038574,
664
+ 0.9999504089355469,
665
+ 0.9999474287033081,
666
+ 0.4916134774684906,
667
+ 0.5060228705406189,
668
+ 0.49220144748687744,
669
+ 0.9999898672103882,
670
+ 0.9999915361404419,
671
+ 0.9999905824661255,
672
+ 0.8047722578048706,
673
+ 0.8523899912834167,
674
+ 0.8705809116363525,
675
+ 0.3922067880630493,
676
+ 0.4324615001678467,
677
+ 0.3545004725456238,
678
+ 0.15673233568668365,
679
+ 0.6471846699714661,
680
+ 0.6424362063407898,
681
+ 0.510160505771637,
682
+ 0.5128241181373596,
683
+ 0.46828266978263855,
684
+ 0.43302497267723083,
685
+ 0.5793876051902771,
686
+ 0.5684614181518555,
687
+ 0.33457034826278687,
688
+ 0.3091184198856354,
689
+ 0.0008105101296678185,
690
+ 0.37077412009239197,
691
+ 0.4145500063896179,
692
+ 0.4761010408401489,
693
+ 0.7447249293327332,
694
+ 0.438394695520401,
695
+ 0.7413656115531921,
696
+ 0.4960051476955414,
697
+ 0.5138289928436279,
698
+ 0.5173673033714294,
699
+ 0.5924621820449829,
700
+ 0.7587030529975891,
701
+ 0.7857908606529236,
702
+ 0.3681909143924713,
703
+ 0.3770434856414795,
704
+ 0.6424773931503296,
705
+ 0.9851202368736267,
706
+ 0.9859110116958618,
707
+ 0.9855308532714844,
708
+ 0.5370909571647644,
709
+ 0.6129180192947388,
710
+ 0.5251724123954773,
711
+ 0.9999006986618042,
712
+ 0.9998983144760132,
713
+ 0.9998980760574341,
714
+ 0.8960296511650085,
715
+ 0.8455232977867126,
716
+ 0.9984153509140015,
717
+ 0.44011861085891724,
718
+ 0.5573714971542358,
719
+ 0.45769622921943665,
720
+ 0.5591621994972229,
721
+ 0.42212074995040894,
722
+ 0.5001301765441895,
723
+ 0.4822878837585449,
724
+ 0.7320272326469421,
725
+ 0.6362009048461914,
726
+ 0.429107666015625,
727
+ 0.41713282465934753,
728
+ 0.4034157991409302,
729
+ 0.5436654686927795,
730
+ 0.47678494453430176,
731
+ 0.5470160245895386,
732
+ 0.5428171157836914,
733
+ 0.5227158665657043,
734
+ 0.5470259785652161,
735
+ 0.4163595139980316,
736
+ 0.35028159618377686,
737
+ 0.5244665145874023,
738
+ 0.40996211767196655,
739
+ 0.9096313714981079,
740
+ 0.37183111906051636,
741
+ 0.47767525911331177,
742
+ 0.47421643137931824,
743
+ 0.46094876527786255,
744
+ 0.47423312067985535,
745
+ 0.4818762242794037,
746
+ 0.4068610668182373,
747
+ 0.497257262468338,
748
+ 0.559116542339325,
749
+ 0.5229799747467041,
750
+ 0.725604772567749
751
+ ],
752
+ "eval_code_hard_runtime": 57.8139,
753
+ "eval_code_hard_samples_per_second": 1.73,
754
+ "eval_code_hard_steps_per_second": 0.035,
755
+ "epoch": 0.88,
756
+ "step": 50
757
+ },
758
+ {
759
+ "loss": 0.1258,
760
+ "learning_rate": 0.0002,
761
+ "epoch": 0.89,
762
+ "step": 51
763
+ },
764
+ {
765
+ "loss": 0.0618,
766
+ "learning_rate": 0.0002,
767
+ "epoch": 0.91,
768
+ "step": 52
769
+ },
770
+ {
771
+ "loss": 0.2478,
772
+ "learning_rate": 0.0002,
773
+ "epoch": 0.93,
774
+ "step": 53
775
+ },
776
+ {
777
+ "loss": 0.4399,
778
+ "learning_rate": 0.0002,
779
+ "epoch": 0.95,
780
+ "step": 54
781
+ },
782
+ {
783
+ "loss": 0.1653,
784
+ "learning_rate": 0.0002,
785
+ "epoch": 0.96,
786
+ "step": 55
787
+ },
788
+ {
789
+ "loss": 0.0441,
790
+ "learning_rate": 0.0002,
791
+ "epoch": 0.98,
792
+ "step": 56
793
+ },
794
+ {
795
+ "loss": 0.0223,
796
+ "learning_rate": 0.0002,
797
+ "epoch": 1.0,
798
+ "step": 57
799
+ },
800
+ {
801
+ "loss": 0.0056,
802
+ "learning_rate": 0.0002,
803
+ "epoch": 1.02,
804
+ "step": 58
805
+ },
806
+ {
807
+ "loss": 0.1784,
808
+ "learning_rate": 0.0002,
809
+ "epoch": 1.04,
810
+ "step": 59
811
+ },
812
+ {
813
+ "loss": 0.0642,
814
+ "learning_rate": 0.0002,
815
+ "epoch": 1.05,
816
+ "step": 60
817
+ },
818
+ {
819
+ "loss": 0.1516,
820
+ "learning_rate": 0.0002,
821
+ "epoch": 1.07,
822
+ "step": 61
823
+ },
824
+ {
825
+ "loss": 0.0548,
826
+ "learning_rate": 0.0002,
827
+ "epoch": 1.09,
828
+ "step": 62
829
+ },
830
+ {
831
+ "loss": 0.0663,
832
+ "learning_rate": 0.0002,
833
+ "epoch": 1.11,
834
+ "step": 63
835
+ },
836
+ {
837
+ "loss": 0.1384,
838
+ "learning_rate": 0.0002,
839
+ "epoch": 1.12,
840
+ "step": 64
841
+ },
842
+ {
843
+ "loss": 0.0505,
844
+ "learning_rate": 0.0002,
845
+ "epoch": 1.14,
846
+ "step": 65
847
+ },
848
+ {
849
+ "loss": 0.0631,
850
+ "learning_rate": 0.0002,
851
+ "epoch": 1.16,
852
+ "step": 66
853
+ },
854
+ {
855
+ "loss": 0.0532,
856
+ "learning_rate": 0.0002,
857
+ "epoch": 1.18,
858
+ "step": 67
859
+ },
860
+ {
861
+ "loss": 0.028,
862
+ "learning_rate": 0.0002,
863
+ "epoch": 1.19,
864
+ "step": 68
865
+ },
866
+ {
867
+ "loss": 0.0891,
868
+ "learning_rate": 0.0002,
869
+ "epoch": 1.21,
870
+ "step": 69
871
+ },
872
+ {
873
+ "loss": 0.0735,
874
+ "learning_rate": 0.0002,
875
+ "epoch": 1.23,
876
+ "step": 70
877
+ },
878
+ {
879
+ "loss": 0.0397,
880
+ "learning_rate": 0.0002,
881
+ "epoch": 1.25,
882
+ "step": 71
883
+ },
884
+ {
885
+ "loss": 0.0342,
886
+ "learning_rate": 0.0002,
887
+ "epoch": 1.26,
888
+ "step": 72
889
+ },
890
+ {
891
+ "loss": 0.0234,
892
+ "learning_rate": 0.0002,
893
+ "epoch": 1.28,
894
+ "step": 73
895
+ },
896
+ {
897
+ "loss": 0.0108,
898
+ "learning_rate": 0.0002,
899
+ "epoch": 1.3,
900
+ "step": 74
901
+ },
902
+ {
903
+ "loss": 0.0525,
904
+ "learning_rate": 0.0002,
905
+ "epoch": 1.32,
906
+ "step": 75
907
+ },
908
+ {
909
+ "eval_code_easy_loss": 0.22187194228172302,
910
+ "eval_code_easy_score": -0.044156476855278015,
911
+ "eval_code_easy_brier_score": 0.044156476855278015,
912
+ "eval_code_easy_average_probability": 0.9226246476173401,
913
+ "eval_code_easy_accuracy": 0.95,
914
+ "eval_code_easy_probabilities": [
915
+ 0.9997438788414001,
916
+ 0.9999991655349731,
917
+ 0.9999997615814209,
918
+ 0.9999998807907104,
919
+ 0.9999997615814209,
920
+ 0.9999947547912598,
921
+ 0.8262179493904114,
922
+ 0.9995021820068359,
923
+ 0.998319685459137,
924
+ 0.988502025604248,
925
+ 0.9749259352684021,
926
+ 0.9808527827262878,
927
+ 0.9996113181114197,
928
+ 0.9993595480918884,
929
+ 0.9999059438705444,
930
+ 0.9027615189552307,
931
+ 0.9773653745651245,
932
+ 0.9999016523361206,
933
+ 0.9789873957633972,
934
+ 0.45622462034225464,
935
+ 0.9388629198074341,
936
+ 1.0,
937
+ 1.0,
938
+ 0.9999997615814209,
939
+ 0.9999998807907104,
940
+ 0.9999997615814209,
941
+ 0.9993877410888672,
942
+ 1.0,
943
+ 0.9999998807907104,
944
+ 0.9999998807907104,
945
+ 0.7212077379226685,
946
+ 0.7462506294250488,
947
+ 0.5923525094985962,
948
+ 0.9379615187644958,
949
+ 0.9973413348197937,
950
+ 0.881181001663208,
951
+ 0.9957089424133301,
952
+ 0.9999865293502808,
953
+ 0.9746310114860535,
954
+ 0.9999599456787109,
955
+ 0.9995377063751221,
956
+ 0.9997233748435974,
957
+ 1.0,
958
+ 1.0,
959
+ 1.0,
960
+ 0.7736052870750427,
961
+ 0.15912780165672302,
962
+ 0.6213001608848572,
963
+ 0.9973452687263489,
964
+ 0.9999977350234985,
965
+ 0.9999613761901855,
966
+ 0.5660737752914429,
967
+ 1.0,
968
+ 0.9999997615814209,
969
+ 0.999991774559021,
970
+ 0.9999953508377075,
971
+ 0.999993085861206,
972
+ 0.9999995231628418,
973
+ 0.6532827615737915,
974
+ 0.999998927116394,
975
+ 0.9999794960021973,
976
+ 0.9999988079071045,
977
+ 0.999998927116394,
978
+ 0.9999998807907104,
979
+ 1.0,
980
+ 1.0,
981
+ 0.997825026512146,
982
+ 0.999733030796051,
983
+ 0.9979919195175171,
984
+ 0.7770313024520874,
985
+ 0.011424303986132145,
986
+ 9.0804387582466e-05,
987
+ 1.0,
988
+ 1.0,
989
+ 0.9999986886978149,
990
+ 0.99936443567276,
991
+ 0.9927093982696533,
992
+ 0.9999185800552368,
993
+ 0.9999991655349731,
994
+ 0.9982544779777527,
995
+ 0.9999996423721313,
996
+ 0.9998847246170044,
997
+ 0.9999909400939941,
998
+ 0.9689223766326904,
999
+ 0.9911876916885376,
1000
+ 0.9976316690444946,
1001
+ 0.9958218336105347,
1002
+ 0.993769109249115,
1003
+ 0.40141063928604126,
1004
+ 0.8972048759460449,
1005
+ 0.9999942779541016,
1006
+ 0.6255294680595398,
1007
+ 0.9922812581062317,
1008
+ 1.0,
1009
+ 1.0,
1010
+ 1.0,
1011
+ 1.0,
1012
+ 1.0,
1013
+ 1.0,
1014
+ 0.9875762462615967
1015
+ ],
1016
+ "eval_code_easy_runtime": 17.1185,
1017
+ "eval_code_easy_samples_per_second": 5.842,
1018
+ "eval_code_easy_steps_per_second": 0.117,
1019
+ "epoch": 1.32,
1020
+ "step": 75
1021
+ },
1022
+ {
1023
+ "eval_code_hard_loss": 0.613934338092804,
1024
+ "eval_code_hard_score": -0.20562565326690674,
1025
+ "eval_code_hard_brier_score": 0.20562565326690674,
1026
+ "eval_code_hard_average_probability": 0.6201408505439758,
1027
+ "eval_code_hard_accuracy": 0.68,
1028
+ "eval_code_hard_probabilities": [
1029
+ 0.9997522234916687,
1030
+ 0.6726481318473816,
1031
+ 0.4837990403175354,
1032
+ 0.9999945163726807,
1033
+ 0.9999929666519165,
1034
+ 0.9999992847442627,
1035
+ 0.5199088454246521,
1036
+ 0.5684794783592224,
1037
+ 0.5020670890808105,
1038
+ 0.6793079376220703,
1039
+ 0.3110067546367645,
1040
+ 0.4360009431838989,
1041
+ 0.999996542930603,
1042
+ 0.9999467134475708,
1043
+ 0.9999498128890991,
1044
+ 0.5130871534347534,
1045
+ 0.5327062010765076,
1046
+ 0.5177708864212036,
1047
+ 0.9999998807907104,
1048
+ 0.9999998807907104,
1049
+ 0.9999998807907104,
1050
+ 0.8731643557548523,
1051
+ 0.9360027313232422,
1052
+ 0.9267758727073669,
1053
+ 0.35389643907546997,
1054
+ 0.39809563755989075,
1055
+ 0.20879632234573364,
1056
+ 0.07375767827033997,
1057
+ 0.5250951051712036,
1058
+ 0.5149143934249878,
1059
+ 0.7871077656745911,
1060
+ 0.7780881524085999,
1061
+ 0.6657141447067261,
1062
+ 0.5761507749557495,
1063
+ 0.758147120475769,
1064
+ 0.6872865557670593,
1065
+ 0.29255661368370056,
1066
+ 0.1509549617767334,
1067
+ 0.004953588359057903,
1068
+ 0.3843843936920166,
1069
+ 0.4454701542854309,
1070
+ 0.5638160705566406,
1071
+ 0.6067726612091064,
1072
+ 0.3357914388179779,
1073
+ 0.5700987577438354,
1074
+ 0.8174778819084167,
1075
+ 0.5854097008705139,
1076
+ 0.5972737073898315,
1077
+ 0.6639817953109741,
1078
+ 0.8354654908180237,
1079
+ 0.8449712991714478,
1080
+ 0.35150453448295593,
1081
+ 0.25781506299972534,
1082
+ 0.7074410319328308,
1083
+ 0.9738778471946716,
1084
+ 0.9760010242462158,
1085
+ 0.9789164066314697,
1086
+ 0.5705886483192444,
1087
+ 0.642711341381073,
1088
+ 0.5328339338302612,
1089
+ 0.999995231628418,
1090
+ 0.9999946355819702,
1091
+ 0.9999946355819702,
1092
+ 0.9326982498168945,
1093
+ 0.7097592353820801,
1094
+ 0.9997654557228088,
1095
+ 0.446959525346756,
1096
+ 0.6458287835121155,
1097
+ 0.47074800729751587,
1098
+ 0.6868018507957458,
1099
+ 0.5448780059814453,
1100
+ 0.5729479789733887,
1101
+ 0.4683343768119812,
1102
+ 0.570155918598175,
1103
+ 0.5829461216926575,
1104
+ 0.4596071243286133,
1105
+ 0.407598614692688,
1106
+ 0.38017627596855164,
1107
+ 0.5608930587768555,
1108
+ 0.49683383107185364,
1109
+ 0.600445568561554,
1110
+ 0.6426953077316284,
1111
+ 0.8385844826698303,
1112
+ 0.6047258377075195,
1113
+ 0.2799336314201355,
1114
+ 0.2773086130619049,
1115
+ 0.6116948127746582,
1116
+ 0.43185508251190186,
1117
+ 0.8329448103904724,
1118
+ 0.4886557459831238,
1119
+ 0.4926038086414337,
1120
+ 0.4693112373352051,
1121
+ 0.4659242630004883,
1122
+ 0.321489155292511,
1123
+ 0.3335782587528229,
1124
+ 0.1558743566274643,
1125
+ 0.5026931762695312,
1126
+ 0.7897409200668335,
1127
+ 0.5024547576904297,
1128
+ 0.9441749453544617
1129
+ ],
1130
+ "eval_code_hard_runtime": 57.778,
1131
+ "eval_code_hard_samples_per_second": 1.731,
1132
+ "eval_code_hard_steps_per_second": 0.035,
1133
+ "epoch": 1.32,
1134
+ "step": 75
1135
+ },
1136
+ {
1137
+ "loss": 0.0344,
1138
+ "learning_rate": 0.0002,
1139
+ "epoch": 1.33,
1140
+ "step": 76
1141
+ },
1142
+ {
1143
+ "loss": 0.0514,
1144
+ "learning_rate": 0.0002,
1145
+ "epoch": 1.35,
1146
+ "step": 77
1147
+ },
1148
+ {
1149
+ "loss": 0.0065,
1150
+ "learning_rate": 0.0002,
1151
+ "epoch": 1.37,
1152
+ "step": 78
1153
+ },
1154
+ {
1155
+ "loss": 0.0893,
1156
+ "learning_rate": 0.0002,
1157
+ "epoch": 1.39,
1158
+ "step": 79
1159
+ },
1160
+ {
1161
+ "loss": 0.0431,
1162
+ "learning_rate": 0.0002,
1163
+ "epoch": 1.4,
1164
+ "step": 80
1165
+ },
1166
+ {
1167
+ "loss": 0.0101,
1168
+ "learning_rate": 0.0002,
1169
+ "epoch": 1.42,
1170
+ "step": 81
1171
+ },
1172
+ {
1173
+ "loss": 0.0622,
1174
+ "learning_rate": 0.0002,
1175
+ "epoch": 1.44,
1176
+ "step": 82
1177
+ },
1178
+ {
1179
+ "loss": 0.0007,
1180
+ "learning_rate": 0.0002,
1181
+ "epoch": 1.46,
1182
+ "step": 83
1183
+ },
1184
+ {
1185
+ "loss": 0.0084,
1186
+ "learning_rate": 0.0002,
1187
+ "epoch": 1.47,
1188
+ "step": 84
1189
+ },
1190
+ {
1191
+ "loss": 0.0307,
1192
+ "learning_rate": 0.0002,
1193
+ "epoch": 1.49,
1194
+ "step": 85
1195
+ },
1196
+ {
1197
+ "loss": 0.0005,
1198
+ "learning_rate": 0.0002,
1199
+ "epoch": 1.51,
1200
+ "step": 86
1201
+ },
1202
+ {
1203
+ "loss": 0.001,
1204
+ "learning_rate": 0.0002,
1205
+ "epoch": 1.53,
1206
+ "step": 87
1207
+ },
1208
+ {
1209
+ "loss": 0.0243,
1210
+ "learning_rate": 0.0002,
1211
+ "epoch": 1.54,
1212
+ "step": 88
1213
+ },
1214
+ {
1215
+ "loss": 0.0417,
1216
+ "learning_rate": 0.0002,
1217
+ "epoch": 1.56,
1218
+ "step": 89
1219
+ },
1220
+ {
1221
+ "loss": 0.0418,
1222
+ "learning_rate": 0.0002,
1223
+ "epoch": 1.58,
1224
+ "step": 90
1225
+ },
1226
+ {
1227
+ "loss": 0.0207,
1228
+ "learning_rate": 0.0002,
1229
+ "epoch": 1.6,
1230
+ "step": 91
1231
+ },
1232
+ {
1233
+ "loss": 0.2464,
1234
+ "learning_rate": 0.0002,
1235
+ "epoch": 1.61,
1236
+ "step": 92
1237
+ },
1238
+ {
1239
+ "loss": 0.1456,
1240
+ "learning_rate": 0.0002,
1241
+ "epoch": 1.63,
1242
+ "step": 93
1243
+ },
1244
+ {
1245
+ "loss": 0.0005,
1246
+ "learning_rate": 0.0002,
1247
+ "epoch": 1.65,
1248
+ "step": 94
1249
+ },
1250
+ {
1251
+ "loss": 0.0025,
1252
+ "learning_rate": 0.0002,
1253
+ "epoch": 1.67,
1254
+ "step": 95
1255
+ },
1256
+ {
1257
+ "loss": 0.0456,
1258
+ "learning_rate": 0.0002,
1259
+ "epoch": 1.68,
1260
+ "step": 96
1261
+ },
1262
+ {
1263
+ "loss": 0.0069,
1264
+ "learning_rate": 0.0002,
1265
+ "epoch": 1.7,
1266
+ "step": 97
1267
+ },
1268
+ {
1269
+ "loss": 0.0092,
1270
+ "learning_rate": 0.0002,
1271
+ "epoch": 1.72,
1272
+ "step": 98
1273
+ },
1274
+ {
1275
+ "loss": 0.1516,
1276
+ "learning_rate": 0.0002,
1277
+ "epoch": 1.74,
1278
+ "step": 99
1279
+ },
1280
+ {
1281
+ "loss": 0.0164,
1282
+ "learning_rate": 0.0002,
1283
+ "epoch": 1.75,
1284
+ "step": 100
1285
+ },
1286
+ {
1287
+ "eval_code_easy_loss": 0.37038832902908325,
1288
+ "eval_code_easy_score": -0.03134830296039581,
1289
+ "eval_code_easy_brier_score": 0.03134830296039581,
1290
+ "eval_code_easy_average_probability": 0.9487545490264893,
1291
+ "eval_code_easy_accuracy": 0.97,
1292
+ "eval_code_easy_probabilities": [
1293
+ 1.0,
1294
+ 1.0,
1295
+ 1.0,
1296
+ 1.0,
1297
+ 1.0,
1298
+ 0.9999997615814209,
1299
+ 0.9346917271614075,
1300
+ 1.0,
1301
+ 0.9999998807907104,
1302
+ 0.9997040629386902,
1303
+ 0.9992467164993286,
1304
+ 0.9994949102401733,
1305
+ 0.9994587302207947,
1306
+ 0.9999791383743286,
1307
+ 0.9999996423721313,
1308
+ 0.9992595314979553,
1309
+ 0.9999998807907104,
1310
+ 1.0,
1311
+ 0.9998413324356079,
1312
+ 0.6475287675857544,
1313
+ 0.9992796778678894,
1314
+ 1.0,
1315
+ 1.0,
1316
+ 1.0,
1317
+ 1.0,
1318
+ 1.0,
1319
+ 1.0,
1320
+ 1.0,
1321
+ 1.0,
1322
+ 1.0,
1323
+ 0.8477630615234375,
1324
+ 0.863884687423706,
1325
+ 0.5967232584953308,
1326
+ 0.8738592267036438,
1327
+ 0.9956537485122681,
1328
+ 0.7420057058334351,
1329
+ 0.9940522313117981,
1330
+ 1.0,
1331
+ 0.9952139854431152,
1332
+ 0.9999994039535522,
1333
+ 1.0,
1334
+ 0.9999545812606812,
1335
+ 1.0,
1336
+ 1.0,
1337
+ 1.0,
1338
+ 0.9897578954696655,
1339
+ 0.2987537980079651,
1340
+ 0.9979766011238098,
1341
+ 0.9999840259552002,
1342
+ 1.0,
1343
+ 1.0,
1344
+ 0.9765376448631287,
1345
+ 1.0,
1346
+ 1.0,
1347
+ 1.0,
1348
+ 1.0,
1349
+ 1.0,
1350
+ 1.0,
1351
+ 0.7584323883056641,
1352
+ 1.0,
1353
+ 1.0,
1354
+ 1.0,
1355
+ 0.9999982118606567,
1356
+ 1.0,
1357
+ 1.0,
1358
+ 1.0,
1359
+ 1.0,
1360
+ 1.0,
1361
+ 1.0,
1362
+ 0.9797248840332031,
1363
+ 1.8104368791682646e-05,
1364
+ 2.9319591199339357e-10,
1365
+ 1.0,
1366
+ 1.0,
1367
+ 1.0,
1368
+ 1.0,
1369
+ 0.9998795986175537,
1370
+ 1.0,
1371
+ 1.0,
1372
+ 0.9999961853027344,
1373
+ 1.0,
1374
+ 0.9750629663467407,
1375
+ 0.9934146404266357,
1376
+ 0.9500468969345093,
1377
+ 1.0,
1378
+ 1.0,
1379
+ 1.0,
1380
+ 0.9999947547912598,
1381
+ 0.843275785446167,
1382
+ 0.9999544620513916,
1383
+ 1.0,
1384
+ 0.6250555515289307,
1385
+ 1.0,
1386
+ 1.0,
1387
+ 1.0,
1388
+ 1.0,
1389
+ 1.0,
1390
+ 1.0,
1391
+ 1.0,
1392
+ 0.999996542930603
1393
+ ],
1394
+ "eval_code_easy_runtime": 17.1259,
1395
+ "eval_code_easy_samples_per_second": 5.839,
1396
+ "eval_code_easy_steps_per_second": 0.117,
1397
+ "epoch": 1.75,
1398
+ "step": 100
1399
+ },
1400
+ {
1401
+ "eval_code_hard_loss": 1.3108476400375366,
1402
+ "eval_code_hard_score": -0.238570898771286,
1403
+ "eval_code_hard_brier_score": 0.238570898771286,
1404
+ "eval_code_hard_average_probability": 0.637829601764679,
1405
+ "eval_code_hard_accuracy": 0.69,
1406
+ "eval_code_hard_probabilities": [
1407
+ 1.0,
1408
+ 0.8749839663505554,
1409
+ 0.6682276129722595,
1410
+ 1.0,
1411
+ 1.0,
1412
+ 1.0,
1413
+ 0.4749336838722229,
1414
+ 0.7184466123580933,
1415
+ 0.4824881851673126,
1416
+ 0.7342566251754761,
1417
+ 0.34355923533439636,
1418
+ 0.3212933838367462,
1419
+ 0.999998927116394,
1420
+ 0.11708329617977142,
1421
+ 0.2533932030200958,
1422
+ 0.4083555340766907,
1423
+ 0.49011072516441345,
1424
+ 0.5354903340339661,
1425
+ 0.9999998807907104,
1426
+ 0.9999995231628418,
1427
+ 0.9999921321868896,
1428
+ 0.9668300151824951,
1429
+ 0.9995100498199463,
1430
+ 0.9995787739753723,
1431
+ 0.5743376612663269,
1432
+ 0.6365594267845154,
1433
+ 0.19516201317310333,
1434
+ 5.900204200770531e-07,
1435
+ 0.6282064914703369,
1436
+ 0.6691750884056091,
1437
+ 0.9999256134033203,
1438
+ 0.9999665021896362,
1439
+ 0.9993390440940857,
1440
+ 0.9999998807907104,
1441
+ 0.9648662805557251,
1442
+ 0.7553066611289978,
1443
+ 0.15391865372657776,
1444
+ 0.0059105814434587955,
1445
+ 0.0041419873014092445,
1446
+ 0.6582466959953308,
1447
+ 0.6763702630996704,
1448
+ 0.8389391303062439,
1449
+ 0.8646246194839478,
1450
+ 0.7645048499107361,
1451
+ 0.873203456401825,
1452
+ 0.9956676959991455,
1453
+ 0.6906198263168335,
1454
+ 0.7347480058670044,
1455
+ 0.9999996423721313,
1456
+ 0.9999585151672363,
1457
+ 0.9992007613182068,
1458
+ 0.034449804574251175,
1459
+ 0.01096381526440382,
1460
+ 0.8154931664466858,
1461
+ 3.0823147767478076e-07,
1462
+ 3.748235144485079e-07,
1463
+ 3.7546107023445074e-07,
1464
+ 0.507102370262146,
1465
+ 0.7094148397445679,
1466
+ 0.6129729151725769,
1467
+ 0.8193482756614685,
1468
+ 0.6906271576881409,
1469
+ 0.7478002905845642,
1470
+ 0.9924389123916626,
1471
+ 0.7537370324134827,
1472
+ 1.0,
1473
+ 0.33823978900909424,
1474
+ 0.8392996788024902,
1475
+ 0.33018726110458374,
1476
+ 0.9978412389755249,
1477
+ 0.9927680492401123,
1478
+ 0.9661868214607239,
1479
+ 0.739952027797699,
1480
+ 0.757354199886322,
1481
+ 0.8434256911277771,
1482
+ 0.5655504465103149,
1483
+ 0.09736324101686478,
1484
+ 0.15342624485492706,
1485
+ 0.5846419930458069,
1486
+ 0.5172314047813416,
1487
+ 0.7640779614448547,
1488
+ 0.7927785515785217,
1489
+ 0.9958668947219849,
1490
+ 0.6929405331611633,
1491
+ 0.4140860140323639,
1492
+ 0.4588907063007355,
1493
+ 0.8826824426651001,
1494
+ 0.4302871525287628,
1495
+ 0.8620820045471191,
1496
+ 0.5388514995574951,
1497
+ 0.49043917655944824,
1498
+ 0.4464437961578369,
1499
+ 0.5461203455924988,
1500
+ 0.01837478019297123,
1501
+ 0.01984582096338272,
1502
+ 0.0083817383274436,
1503
+ 0.417593389749527,
1504
+ 0.9964190721511841,
1505
+ 0.5195274949073792,
1506
+ 0.9980224370956421
1507
+ ],
1508
+ "eval_code_hard_runtime": 57.8261,
1509
+ "eval_code_hard_samples_per_second": 1.729,
1510
+ "eval_code_hard_steps_per_second": 0.035,
1511
+ "epoch": 1.75,
1512
+ "step": 100
1513
+ },
1514
+ {
1515
+ "train_runtime": 1718.4456,
1516
+ "train_samples_per_second": 1.862,
1517
+ "train_steps_per_second": 0.058,
1518
+ "total_flos": 0.0,
1519
+ "train_loss": 0.1626638910168549,
1520
+ "epoch": 1.75,
1521
+ "step": 100
1522
+ }
1523
+ ]