amirali1985 commited on
Commit
726aa8c
·
verified ·
1 Parent(s): 63d70e4

Upload add_sub_baseline_500K_1L2H256d

Browse files
add_sub_baseline_500K_1L2H256d/config.json ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "SorlModelWrapper"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": null,
8
+ "dtype": "float32",
9
+ "eos_token_id": null,
10
+ "head_dim": 128,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 256,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 1024,
15
+ "layer_types": [
16
+ "full_attention"
17
+ ],
18
+ "max_position_embeddings": 128,
19
+ "max_window_layers": 28,
20
+ "model_type": "qwen3",
21
+ "num_attention_heads": 2,
22
+ "num_hidden_layers": 1,
23
+ "num_key_value_heads": 2,
24
+ "pad_token_id": null,
25
+ "rms_norm_eps": 1e-06,
26
+ "rope_parameters": {
27
+ "rope_theta": 10000.0,
28
+ "rope_type": "default"
29
+ },
30
+ "sliding_window": null,
31
+ "tie_word_embeddings": false,
32
+ "transformers_version": "5.5.0",
33
+ "use_cache": true,
34
+ "use_sliding_window": false,
35
+ "vocab_size": 151645
36
+ }
add_sub_baseline_500K_1L2H256d/generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "output_attentions": false,
4
+ "output_hidden_states": false,
5
+ "transformers_version": "5.5.0",
6
+ "use_cache": true
7
+ }
add_sub_baseline_500K_1L2H256d/metrics.json ADDED
The diff for this file is too large to render. See raw diff
 
add_sub_baseline_500K_1L2H256d/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ccee7a335f07c97fa59cdfc681065552f7c63209954b782e2621114d3c8bd391
3
+ size 315072674
add_sub_baseline_500K_1L2H256d/train_config.json ADDED
@@ -0,0 +1,35 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "mode": "baseline",
3
+ "ops": "add_sub",
4
+ "n_digits": 6,
5
+ "n_layer": 1,
6
+ "n_head": 2,
7
+ "n_embd": 256,
8
+ "abs_vocab": 0,
9
+ "K": 4,
10
+ "alpha_info_gain": 10.0,
11
+ "alpha_abs": 0.1,
12
+ "alpha_soft_zipf": 1.0,
13
+ "batch_size": 64,
14
+ "num_epochs": 20,
15
+ "dataset_size": 500000,
16
+ "lr": 8e-05,
17
+ "output_dir": "ckpt/sweep/as_baseline_500K_1L2H256d",
18
+ "device": "cuda",
19
+ "push_to_hub": true,
20
+ "no_wandb": false,
21
+ "n_params": 78691840,
22
+ "run_name": "add_sub_baseline_500K_1L2H256d",
23
+ "git_commit": "800625019270114adcda289bbd550c4f1109a514",
24
+ "timestamp": "2026-04-12T02:50:01.630872+00:00",
25
+ "tokenizer": "Qwen/Qwen3-0.6B",
26
+ "dataset_repo": "thoughtworks/arithmetic-sorl-data",
27
+ "dataset_config": "add_sub_6digit",
28
+ "model_repo": "thoughtworks/arithmetic-sorl",
29
+ "trainer_version": "sft",
30
+ "wandb_run_id": "ra0btz03",
31
+ "wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/ra0btz03",
32
+ "final_accuracy": 0.5914285714285714,
33
+ "sft_accuracy": 0.5914285714285714,
34
+ "eval_method": "ArithmeticEvaluator"
35
+ }