gamemaker1 commited on
Commit
33ed363
·
verified ·
1 Parent(s): 1497f87

Upload folder using huggingface_hub

Browse files
.gitattributes CHANGED
@@ -36,3 +36,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
36
  part1/dense/translations.jsonl filter=lfs diff=lfs merge=lfs -text
37
  part1/moe-shared/routing-heatmap.png filter=lfs diff=lfs merge=lfs -text
38
  part1/moe-shared/translations.jsonl filter=lfs diff=lfs merge=lfs -text
 
 
 
36
  part1/dense/translations.jsonl filter=lfs diff=lfs merge=lfs -text
37
  part1/moe-shared/routing-heatmap.png filter=lfs diff=lfs merge=lfs -text
38
  part1/moe-shared/translations.jsonl filter=lfs diff=lfs merge=lfs -text
39
+ part1/moe-top1/routing-heatmap.png filter=lfs diff=lfs merge=lfs -text
40
+ part1/moe-top1/translations.jsonl filter=lfs diff=lfs merge=lfs -text
part1/moe-top1/config.json ADDED
@@ -0,0 +1,62 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": {
3
+ "d_model": 512,
4
+ "n_layers": 6,
5
+ "n_heads": 8,
6
+ "n_ctx": 256,
7
+ "d_ff": 1536,
8
+ "dropout": 0.0,
9
+ "use_moe": true,
10
+ "n_experts": 4,
11
+ "top_k": 1,
12
+ "n_shared": 0,
13
+ "d_expert": 384
14
+ },
15
+ "seed": 0,
16
+ "data": {
17
+ "name": "belumind/en-vi-ja-curated-500k-triplets",
18
+ "splits": {
19
+ "train": "train",
20
+ "val": "validation",
21
+ "test": "test"
22
+ },
23
+ "columns": {
24
+ "en": "en",
25
+ "vi": "vi",
26
+ "ja": "ja"
27
+ },
28
+ "seed": 0,
29
+ "limit": null
30
+ },
31
+ "tokenizer": {
32
+ "file": "unigram-16k.json",
33
+ "vocab_size": 16000,
34
+ "sample_triplets": 200000
35
+ },
36
+ "train": {
37
+ "max_tokens": 60000000,
38
+ "batch_size": 128,
39
+ "lr": 0.001,
40
+ "betas": [
41
+ 0.9,
42
+ 0.98
43
+ ],
44
+ "weight_decay": 0.1,
45
+ "grad_clip": 1.0,
46
+ "warmup_fraction": 0.05,
47
+ "min_lr_ratio": 0.1,
48
+ "eval_every_fraction": 0.1,
49
+ "log_every": 50
50
+ },
51
+ "evaluate": {
52
+ "batch_size": 128,
53
+ "val_pairs": 4000,
54
+ "max_new_tokens": 128,
55
+ "bleu_limit": null
56
+ },
57
+ "wandb": {
58
+ "enabled": true,
59
+ "project": "anlp-assignment-2"
60
+ },
61
+ "name": "moe-top1"
62
+ }
part1/moe-top1/metrics.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
part1/moe-top1/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d40cea35438a32c32be131adbce51b7a54055ee5f7f552e46153dfb5981480ae
3
+ size 147439711
part1/moe-top1/results.json ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "moe-top1",
3
+ "parameters": {
4
+ "total": 36850176,
5
+ "active": 26233344
6
+ },
7
+ "complete": true,
8
+ "test_loss": 1.437988726516569,
9
+ "test_perplexity": 4.21221537452151,
10
+ "bleu_vi": 40.88040823373441,
11
+ "bleu_ja": 32.495606327381665,
12
+ "bleu": 36.71750455096259,
13
+ "bleu_pairs": 49583
14
+ }
part1/moe-top1/routing-heatmap.pdf ADDED
Binary file (17.3 kB). View file
 
part1/moe-top1/routing-heatmap.png ADDED

Git LFS Details

  • SHA256: 133c12edbe6a535f18282424d91c1e826c342ae9a287de1dafb804c1a6afe857
  • Pointer size: 131 Bytes
  • Size of remote file: 189 kB
part1/moe-top1/translations.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d0eca1219e63963de2843d7fff9b053742d888fa42eee850b5fefa7becc5f49
3
+ size 16613182