amirali1985 commited on
Commit
eec6d9b
·
verified ·
1 Parent(s): d72298f

Upload add_sub_sorl_v1_abs30_100K

Browse files
add_sub_sorl_v1_abs30_100K/metrics.json CHANGED
The diff for this file is too large to render. See raw diff
 
add_sub_sorl_v1_abs30_100K/model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7332e8a91d9f010ea6aaaa8db3a4f5b2d0da3000748bf222741ae40b08e20106
3
  size 650385300
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e953953ca0d08b182e12ee4ae014c135cbe7dbf8676e52d960d3206b93abebed
3
  size 650385300
add_sub_sorl_v1_abs30_100K/train_config.json CHANGED
@@ -17,10 +17,10 @@
17
  "target_vocab_util": 0.8,
18
  "min_abs_ppl": 0.0,
19
  "zipf_alpha": 1.0,
20
- "lr": 4e-05,
21
  "emb_lr_mult": 1.0,
22
  "weight_decay": 0.01,
23
- "warmup_steps": 468,
24
  "cooldown_frac": 0.4,
25
  "max_grad_norm": 1.0,
26
  "vq_abs_pretrain_steps": 0,
@@ -30,13 +30,13 @@
30
  "vq_abs_pretrain_target_vectors": 20000,
31
  "batch_size": 64,
32
  "gradient_accumulation_steps": 1,
33
- "num_epochs": 10,
34
  "emb_warmup_steps": 0,
35
  "log_every": 50,
36
  "eval_every": 1562,
37
  "save_every": 999999,
38
  "eval_samples": 100,
39
- "output_dir": "ckpt/sweep/as_sorl_abs30_K4_100K",
40
  "eval_K": 4,
41
  "alpha_traj": 0.0,
42
  "corrupt_method": "shuffle",
@@ -69,16 +69,16 @@
69
  "no_wandb": false,
70
  "n_params": 162519662,
71
  "run_name": "add_sub_sorl_v1_abs30_100K",
72
- "git_commit": "f447da529caceac8c7d256cbb2cd185cbc50feac",
73
- "timestamp": "2026-04-12T12:09:53.734351+00:00",
74
  "tokenizer": "Qwen/Qwen3-0.6B",
75
  "dataset_repo": "thoughtworks/arithmetic-sorl-data",
76
  "dataset_config": "add_sub_6digit",
77
  "model_repo": "thoughtworks/arithmetic-sorl",
78
  "trainer_version": "v1",
79
- "wandb_run_id": "iffr93ko",
80
- "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/iffr93ko",
81
- "final_accuracy": 0.96125,
82
- "sft_accuracy": 0.515,
83
  "eval_method": "ArithmeticEvaluator"
84
  }
 
17
  "target_vocab_util": 0.8,
18
  "min_abs_ppl": 0.0,
19
  "zipf_alpha": 1.0,
20
+ "lr": 8e-05,
21
  "emb_lr_mult": 1.0,
22
  "weight_decay": 0.01,
23
+ "warmup_steps": 937,
24
  "cooldown_frac": 0.4,
25
  "max_grad_norm": 1.0,
26
  "vq_abs_pretrain_steps": 0,
 
30
  "vq_abs_pretrain_target_vectors": 20000,
31
  "batch_size": 64,
32
  "gradient_accumulation_steps": 1,
33
+ "num_epochs": 20,
34
  "emb_warmup_steps": 0,
35
  "log_every": 50,
36
  "eval_every": 1562,
37
  "save_every": 999999,
38
  "eval_samples": 100,
39
+ "output_dir": "ckpt/sweep/as_sorl_abs30_100K",
40
  "eval_K": 4,
41
  "alpha_traj": 0.0,
42
  "corrupt_method": "shuffle",
 
69
  "no_wandb": false,
70
  "n_params": 162519662,
71
  "run_name": "add_sub_sorl_v1_abs30_100K",
72
+ "git_commit": "57deaa28d9c21e39ddac5ef448d6e1be992fba91",
73
+ "timestamp": "2026-04-13T09:55:13.635116+00:00",
74
  "tokenizer": "Qwen/Qwen3-0.6B",
75
  "dataset_repo": "thoughtworks/arithmetic-sorl-data",
76
  "dataset_config": "add_sub_6digit",
77
  "model_repo": "thoughtworks/arithmetic-sorl",
78
  "trainer_version": "v1",
79
+ "wandb_run_id": "17z5a7rj",
80
+ "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/17z5a7rj",
81
+ "final_accuracy": 1.0,
82
+ "sft_accuracy": 1.0,
83
  "eval_method": "ArithmeticEvaluator"
84
  }