pdhinaka commited on
Commit
c96c192
·
verified ·
1 Parent(s): 4e88ab4

upload specviz toy models

Browse files
README.md ADDED
@@ -0,0 +1,94 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ datasets: [roneneldan/TinyStories]
4
+ tags: [speculative-decoding, toy-model, tinystories]
5
+ ---
6
+ # specviz-tinystories
7
+
8
+ Toy model family for visualizing speculative decoding. Shared 4k byte-level BPE tokenizer.
9
+
10
+ | folder | what | notes |
11
+ |---|---|---|
12
+ | `target/` | 8-layer / 512-d GPT (RoPE, RMSNorm, tied emb) | trained with LayerSkip-style early-exit loss, so any prefix of layers is a usable self-drafter |
13
+ | `draft/` | 2-layer / 256-d GPT | distilled from `target` (KL + CE) |
14
+ | `medusa/` | 4 Medusa heads | on frozen `target` |
15
+ | `hydra/` | 4 Hydra (sequentially dependent) heads | on frozen `target` |
16
+ | `eagle/` | 1-layer EAGLE feature drafter | on frozen `target` |
17
+
18
+ Each folder has `model.safetensors`, `config.json`, `metrics.json` (teacher-forced
19
+ acceptance estimates on validation), and `tokenizer.json`.
20
+
21
+ ## Metrics (validation, teacher-forced)
22
+ ```json
23
+ {
24
+ "draft": {
25
+ "val_ce": 1.647251844406128,
26
+ "kl_to_target": 0.4337791875004769,
27
+ "alpha": 0.7311673790216446,
28
+ "greedy": 0.7380981445312498,
29
+ "iter": 12000
30
+ },
31
+ "eagle": {
32
+ "reg": 0.115620232000947,
33
+ "alpha": 0.7858343034982681,
34
+ "greedy": 0.7855078756809234,
35
+ "iter": 10000
36
+ },
37
+ "hydra": {
38
+ "head1_alpha": 0.6419515252113344,
39
+ "head1_greedy": 0.6670437842607498,
40
+ "head2_alpha": 0.5645744204521179,
41
+ "head2_greedy": 0.5964047521352768,
42
+ "head3_alpha": 0.5303537666797639,
43
+ "head3_greedy": 0.5645859122276305,
44
+ "head4_alpha": 0.5123031288385391,
45
+ "head4_greedy": 0.5469441980123521,
46
+ "iter": 6000
47
+ },
48
+ "medusa": {
49
+ "head1_alpha": 0.4397877633571626,
50
+ "head1_greedy": 0.49165851026773455,
51
+ "head2_alpha": 0.2549418851733207,
52
+ "head2_greedy": 0.31615198105573644,
53
+ "head3_alpha": 0.17144695296883586,
54
+ "head3_greedy": 0.22621102407574656,
55
+ "head4_alpha": 0.13068613037467003,
56
+ "head4_greedy": 0.17579355239868166,
57
+ "iter": 6000
58
+ },
59
+ "target": {
60
+ "val_ce": 1.2464708268642428,
61
+ "exit1_alpha": 0.4845910012722016,
62
+ "exit1_greedy": 0.5295845031738281,
63
+ "exit2_alpha": 0.6358921855688096,
64
+ "exit2_greedy": 0.6550628662109373,
65
+ "exit3_alpha": 0.7252050876617432,
66
+ "exit3_greedy": 0.7312286376953127,
67
+ "exit4_alpha": 0.7719815969467161,
68
+ "exit4_greedy": 0.772979736328125,
69
+ "exit5_alpha": 0.8381113588809967,
70
+ "exit5_greedy": 0.8338546752929686,
71
+ "exit6_alpha": 0.8831344306468962,
72
+ "exit6_greedy": 0.8760437011718749,
73
+ "exit7_alpha": 0.9238532721996306,
74
+ "exit7_greedy": 0.9169792175292967,
75
+ "iter": 20000
76
+ },
77
+ "sanity": {
78
+ "draft_model": {
79
+ "greedy_lossless": true,
80
+ "mean_accepted": 2.0,
81
+ "tokens_per_target_call": 3.0
82
+ },
83
+ "layerskip_exit3": {
84
+ "greedy_lossless": true,
85
+ "mean_accepted": 2.4285714285714284,
86
+ "tokens_per_target_call": 3.4285714285714284
87
+ },
88
+ "spec_sample_T1": {
89
+ "mean_accepted": 1.813953488372093,
90
+ "tokens_per_target_call": 2.7906976744186047
91
+ }
92
+ }
93
+ }
94
+ ```
draft/config.json ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "kind": "gpt",
3
+ "role": "draft",
4
+ "model": {
5
+ "vocab_size": 4096,
6
+ "block_size": 512,
7
+ "n_layer": 2,
8
+ "n_head": 4,
9
+ "n_embd": 256,
10
+ "rope_base": 10000.0
11
+ },
12
+ "train": {
13
+ "kd_alpha": 0.9,
14
+ "distilled_from": "target"
15
+ }
16
+ }
draft/metrics.json ADDED
@@ -0,0 +1,95 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "last": {
3
+ "val_ce": 1.647251844406128,
4
+ "kl_to_target": 0.4337791875004769,
5
+ "alpha": 0.7311673790216446,
6
+ "greedy": 0.7380981445312498,
7
+ "iter": 12000
8
+ },
9
+ "history": [
10
+ {
11
+ "val_ce": 2.0099524259567256,
12
+ "kl_to_target": 0.7797438204288483,
13
+ "alpha": 0.622846382856369,
14
+ "greedy": 0.6417129516601563,
15
+ "iter": 1000
16
+ },
17
+ {
18
+ "val_ce": 1.8566814422607423,
19
+ "kl_to_target": 0.6337778121232986,
20
+ "alpha": 0.6686296522617341,
21
+ "greedy": 0.6780601501464842,
22
+ "iter": 2000
23
+ },
24
+ {
25
+ "val_ce": 1.792548328638077,
26
+ "kl_to_target": 0.5724014222621918,
27
+ "alpha": 0.6858189970254898,
28
+ "greedy": 0.6960624694824219,
29
+ "iter": 3000
30
+ },
31
+ {
32
+ "val_ce": 1.7606663465499879,
33
+ "kl_to_target": 0.5405576616525649,
34
+ "alpha": 0.6973596751689911,
35
+ "greedy": 0.7049896240234375,
36
+ "iter": 4000
37
+ },
38
+ {
39
+ "val_ce": 1.746587711572647,
40
+ "kl_to_target": 0.5206510752439498,
41
+ "alpha": 0.7006595581769943,
42
+ "greedy": 0.7090461730957033,
43
+ "iter": 5000
44
+ },
45
+ {
46
+ "val_ce": 1.7139503300189969,
47
+ "kl_to_target": 0.5007144927978516,
48
+ "alpha": 0.7079429775476455,
49
+ "greedy": 0.7160675048828127,
50
+ "iter": 6000
51
+ },
52
+ {
53
+ "val_ce": 1.7016730666160584,
54
+ "kl_to_target": 0.4825879991054535,
55
+ "alpha": 0.7134584307670593,
56
+ "greedy": 0.7211517333984375,
57
+ "iter": 7000
58
+ },
59
+ {
60
+ "val_ce": 1.6721126437187195,
61
+ "kl_to_target": 0.46867714226245877,
62
+ "alpha": 0.7192983746528624,
63
+ "greedy": 0.7264602661132813,
64
+ "iter": 8000
65
+ },
66
+ {
67
+ "val_ce": 1.673528206348419,
68
+ "kl_to_target": 0.4556530773639678,
69
+ "alpha": 0.7239526838064192,
70
+ "greedy": 0.7305328369140626,
71
+ "iter": 9000
72
+ },
73
+ {
74
+ "val_ce": 1.6636894285678863,
75
+ "kl_to_target": 0.4467218339443207,
76
+ "alpha": 0.7272016823291778,
77
+ "greedy": 0.7334213256835938,
78
+ "iter": 10000
79
+ },
80
+ {
81
+ "val_ce": 1.6511635184288023,
82
+ "kl_to_target": 0.43592406362295144,
83
+ "alpha": 0.7305143564939498,
84
+ "greedy": 0.7367881774902344,
85
+ "iter": 11000
86
+ },
87
+ {
88
+ "val_ce": 1.647251844406128,
89
+ "kl_to_target": 0.4337791875004769,
90
+ "alpha": 0.7311673790216446,
91
+ "greedy": 0.7380981445312498,
92
+ "iter": 12000
93
+ }
94
+ ]
95
+ }
draft/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b04a1aaf47be6790513bf2bda575a4130348fe42a82bf26507825ee81e7034b1
3
+ size 10492208
draft/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
eagle/config.json ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "kind": "eagle",
3
+ "model": {
4
+ "vocab_size": 4096,
5
+ "block_size": 512,
6
+ "n_layer": 8,
7
+ "n_head": 8,
8
+ "n_embd": 512,
9
+ "rope_base": 10000.0
10
+ },
11
+ "target": "target",
12
+ "train": {
13
+ "w_cls": 0.5,
14
+ "feat_noise": 0.1
15
+ }
16
+ }
eagle/metrics.json ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "last": {
3
+ "reg": 0.115620232000947,
4
+ "alpha": 0.7858343034982681,
5
+ "greedy": 0.7855078756809234,
6
+ "iter": 10000
7
+ },
8
+ "history": [
9
+ {
10
+ "reg": 0.16248844116926192,
11
+ "alpha": 0.7045249938964844,
12
+ "greedy": 0.7131298869848252,
13
+ "iter": 1000
14
+ },
15
+ {
16
+ "reg": 0.1393620289862156,
17
+ "alpha": 0.7411754161119462,
18
+ "greedy": 0.7447376430034637,
19
+ "iter": 2000
20
+ },
21
+ {
22
+ "reg": 0.13170703276991844,
23
+ "alpha": 0.7544642329216005,
24
+ "greedy": 0.7579302132129669,
25
+ "iter": 3000
26
+ },
27
+ {
28
+ "reg": 0.1273528974503279,
29
+ "alpha": 0.7619078665971756,
30
+ "greedy": 0.7615092933177947,
31
+ "iter": 4000
32
+ },
33
+ {
34
+ "reg": 0.1248361334204674,
35
+ "alpha": 0.7684480398893356,
36
+ "greedy": 0.7694517374038696,
37
+ "iter": 5000
38
+ },
39
+ {
40
+ "reg": 0.12067152783274653,
41
+ "alpha": 0.7745981186628341,
42
+ "greedy": 0.7758852064609528,
43
+ "iter": 6000
44
+ },
45
+ {
46
+ "reg": 0.11911505199968815,
47
+ "alpha": 0.7785812288522721,
48
+ "greedy": 0.7784430116415023,
49
+ "iter": 7000
50
+ },
51
+ {
52
+ "reg": 0.11806977093219759,
53
+ "alpha": 0.7828734844923019,
54
+ "greedy": 0.7825571775436401,
55
+ "iter": 8000
56
+ },
57
+ {
58
+ "reg": 0.11849058754742146,
59
+ "alpha": 0.7845947831869126,
60
+ "greedy": 0.7849345535039902,
61
+ "iter": 9000
62
+ },
63
+ {
64
+ "reg": 0.115620232000947,
65
+ "alpha": 0.7858343034982681,
66
+ "greedy": 0.7855078756809234,
67
+ "iter": 10000
68
+ }
69
+ ]
70
+ }
eagle/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0ba82601b36db042de1bbc469ace334de868e977cb746b29bb44e901669b0176
3
+ size 14686912
eagle/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
hydra/config.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "kind": "hydra",
3
+ "K": 4,
4
+ "target": "target",
5
+ "target_model": {
6
+ "vocab_size": 4096,
7
+ "block_size": 512,
8
+ "n_layer": 8,
9
+ "n_head": 8,
10
+ "n_embd": 512,
11
+ "rope_base": 10000.0
12
+ }
13
+ }
hydra/metrics.json ADDED
@@ -0,0 +1,81 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "last": {
3
+ "head1_alpha": 0.6419515252113344,
4
+ "head1_greedy": 0.6670437842607498,
5
+ "head2_alpha": 0.5645744204521179,
6
+ "head2_greedy": 0.5964047521352768,
7
+ "head3_alpha": 0.5303537666797639,
8
+ "head3_greedy": 0.5645859122276305,
9
+ "head4_alpha": 0.5123031288385391,
10
+ "head4_greedy": 0.5469441980123521,
11
+ "iter": 6000
12
+ },
13
+ "history": [
14
+ {
15
+ "head1_alpha": 0.5846165388822556,
16
+ "head1_greedy": 0.6149431198835373,
17
+ "head2_alpha": 0.48963863402605057,
18
+ "head2_greedy": 0.524047213792801,
19
+ "head3_alpha": 0.4546733692288399,
20
+ "head3_greedy": 0.4870641157031059,
21
+ "head4_alpha": 0.4275311261415482,
22
+ "head4_greedy": 0.4656050026416778,
23
+ "iter": 1000
24
+ },
25
+ {
26
+ "head1_alpha": 0.6139427363872528,
27
+ "head1_greedy": 0.641572892665863,
28
+ "head2_alpha": 0.5227782577276229,
29
+ "head2_greedy": 0.5582904785871506,
30
+ "head3_alpha": 0.48853647857904436,
31
+ "head3_greedy": 0.5266699612140655,
32
+ "head4_alpha": 0.4622075036168098,
33
+ "head4_greedy": 0.501694755256176,
34
+ "iter": 2000
35
+ },
36
+ {
37
+ "head1_alpha": 0.629333007335663,
38
+ "head1_greedy": 0.6546905517578125,
39
+ "head2_alpha": 0.5431721359491348,
40
+ "head2_greedy": 0.5781771183013916,
41
+ "head3_alpha": 0.510604503750801,
42
+ "head3_greedy": 0.5455780595541,
43
+ "head4_alpha": 0.4853551983833313,
44
+ "head4_greedy": 0.5253983169794083,
45
+ "iter": 3000
46
+ },
47
+ {
48
+ "head1_alpha": 0.634756550192833,
49
+ "head1_greedy": 0.6599574863910674,
50
+ "head2_alpha": 0.5565346986055374,
51
+ "head2_greedy": 0.5882981330156325,
52
+ "head3_alpha": 0.5199958384037018,
53
+ "head3_greedy": 0.5554257839918137,
54
+ "head4_alpha": 0.5002639740705489,
55
+ "head4_greedy": 0.5371001482009888,
56
+ "iter": 4000
57
+ },
58
+ {
59
+ "head1_alpha": 0.6393299520015717,
60
+ "head1_greedy": 0.6642306625843047,
61
+ "head2_alpha": 0.5620603740215302,
62
+ "head2_greedy": 0.5925521194934846,
63
+ "head3_alpha": 0.5261775463819504,
64
+ "head3_greedy": 0.5606366783380508,
65
+ "head4_alpha": 0.5070496618747711,
66
+ "head4_greedy": 0.5432009696960449,
67
+ "iter": 5000
68
+ },
69
+ {
70
+ "head1_alpha": 0.6419515252113344,
71
+ "head1_greedy": 0.6670437842607498,
72
+ "head2_alpha": 0.5645744204521179,
73
+ "head2_greedy": 0.5964047521352768,
74
+ "head3_alpha": 0.5303537666797639,
75
+ "head3_greedy": 0.5645859122276305,
76
+ "head4_alpha": 0.5123031288385391,
77
+ "head4_greedy": 0.5469441980123521,
78
+ "iter": 6000
79
+ }
80
+ ]
81
+ }
hydra/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:361c980c34a10c60090bc7d1d9622c2c92e435dfc9ce02515ca0a4218b987cb3
3
+ size 18892080
hydra/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
medusa/config.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "kind": "medusa",
3
+ "K": 4,
4
+ "target": "target",
5
+ "target_model": {
6
+ "vocab_size": 4096,
7
+ "block_size": 512,
8
+ "n_layer": 8,
9
+ "n_head": 8,
10
+ "n_embd": 512,
11
+ "rope_base": 10000.0
12
+ }
13
+ }
medusa/metrics.json ADDED
@@ -0,0 +1,81 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "last": {
3
+ "head1_alpha": 0.4397877633571626,
4
+ "head1_greedy": 0.49165851026773455,
5
+ "head2_alpha": 0.2549418851733207,
6
+ "head2_greedy": 0.31615198105573644,
7
+ "head3_alpha": 0.17144695296883586,
8
+ "head3_greedy": 0.22621102407574656,
9
+ "head4_alpha": 0.13068613037467003,
10
+ "head4_greedy": 0.17579355239868166,
11
+ "iter": 6000
12
+ },
13
+ "history": [
14
+ {
15
+ "head1_alpha": 0.42103663235902783,
16
+ "head1_greedy": 0.471053996682167,
17
+ "head2_alpha": 0.24204960912466045,
18
+ "head2_greedy": 0.3009237289428711,
19
+ "head3_alpha": 0.16207220479846002,
20
+ "head3_greedy": 0.21406250670552254,
21
+ "head4_alpha": 0.1241915762424469,
22
+ "head4_greedy": 0.1651267223060131,
23
+ "iter": 1000
24
+ },
25
+ {
26
+ "head1_alpha": 0.4280305802822113,
27
+ "head1_greedy": 0.4776082381606102,
28
+ "head2_alpha": 0.24610844999551776,
29
+ "head2_greedy": 0.30440258979797363,
30
+ "head3_alpha": 0.16589450091123584,
31
+ "head3_greedy": 0.2168037891387939,
32
+ "head4_alpha": 0.1272031757980585,
33
+ "head4_greedy": 0.16875307410955426,
34
+ "iter": 2000
35
+ },
36
+ {
37
+ "head1_alpha": 0.43427005112171174,
38
+ "head1_greedy": 0.4846318483352661,
39
+ "head2_alpha": 0.25126655772328377,
40
+ "head2_greedy": 0.31229781210422514,
41
+ "head3_alpha": 0.16857100352644916,
42
+ "head3_greedy": 0.2231105804443359,
43
+ "head4_alpha": 0.12878014594316484,
44
+ "head4_greedy": 0.17323603406548505,
45
+ "iter": 3000
46
+ },
47
+ {
48
+ "head1_alpha": 0.4371277198195458,
49
+ "head1_greedy": 0.488535037636757,
50
+ "head2_alpha": 0.2534819707274437,
51
+ "head2_greedy": 0.3144929721951485,
52
+ "head3_alpha": 0.17099412232637404,
53
+ "head3_greedy": 0.22533614486455916,
54
+ "head4_alpha": 0.13038476184010506,
55
+ "head4_greedy": 0.17381120920181278,
56
+ "iter": 4000
57
+ },
58
+ {
59
+ "head1_alpha": 0.43966121226549154,
60
+ "head1_greedy": 0.49001803547143935,
61
+ "head2_alpha": 0.2547260284423828,
62
+ "head2_greedy": 0.31685357391834257,
63
+ "head3_alpha": 0.1713761068880558,
64
+ "head3_greedy": 0.226174183934927,
65
+ "head4_alpha": 0.1306981660425663,
66
+ "head4_greedy": 0.17601808533072472,
67
+ "iter": 5000
68
+ },
69
+ {
70
+ "head1_alpha": 0.4397877633571626,
71
+ "head1_greedy": 0.49165851026773455,
72
+ "head2_alpha": 0.2549418851733207,
73
+ "head2_greedy": 0.31615198105573644,
74
+ "head3_alpha": 0.17144695296883586,
75
+ "head3_greedy": 0.22621102407574656,
76
+ "head4_alpha": 0.13068613037467003,
77
+ "head4_greedy": 0.17579355239868166,
78
+ "iter": 6000
79
+ }
80
+ ]
81
+ }
medusa/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:57670da9538e3bd0638da006ab1eaacd98da80889ba9242afa78deb46e09f6d8
3
+ size 37757960
medusa/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
sanity.json ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "draft_model": {
3
+ "greedy_lossless": true,
4
+ "mean_accepted": 2.0,
5
+ "tokens_per_target_call": 3.0
6
+ },
7
+ "layerskip_exit3": {
8
+ "greedy_lossless": true,
9
+ "mean_accepted": 2.4285714285714284,
10
+ "tokens_per_target_call": 3.4285714285714284
11
+ },
12
+ "spec_sample_T1": {
13
+ "mean_accepted": 1.813953488372093,
14
+ "tokens_per_target_call": 2.7906976744186047
15
+ }
16
+ }
target/config.json ADDED
@@ -0,0 +1,15 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "kind": "gpt",
3
+ "role": "target",
4
+ "model": {
5
+ "vocab_size": 4096,
6
+ "block_size": 512,
7
+ "n_layer": 8,
8
+ "n_head": 8,
9
+ "n_embd": 512,
10
+ "rope_base": 10000.0
11
+ },
12
+ "train": {
13
+ "early_exit": 0.25
14
+ }
15
+ }
target/metrics.json ADDED
@@ -0,0 +1,382 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "last": {
3
+ "val_ce": 1.2464708268642428,
4
+ "exit1_alpha": 0.4845910012722016,
5
+ "exit1_greedy": 0.5295845031738281,
6
+ "exit2_alpha": 0.6358921855688096,
7
+ "exit2_greedy": 0.6550628662109373,
8
+ "exit3_alpha": 0.7252050876617432,
9
+ "exit3_greedy": 0.7312286376953127,
10
+ "exit4_alpha": 0.7719815969467161,
11
+ "exit4_greedy": 0.772979736328125,
12
+ "exit5_alpha": 0.8381113588809967,
13
+ "exit5_greedy": 0.8338546752929686,
14
+ "exit6_alpha": 0.8831344306468962,
15
+ "exit6_greedy": 0.8760437011718749,
16
+ "exit7_alpha": 0.9238532721996306,
17
+ "exit7_greedy": 0.9169792175292967,
18
+ "iter": 20000
19
+ },
20
+ "history": [
21
+ {
22
+ "val_ce": 1.771557992696762,
23
+ "exit1_alpha": 0.40818989574909204,
24
+ "exit1_greedy": 0.4042259216308594,
25
+ "exit2_alpha": 0.5836701780557633,
26
+ "exit2_greedy": 0.5889640808105467,
27
+ "exit3_alpha": 0.7103229671716691,
28
+ "exit3_greedy": 0.7000419616699218,
29
+ "exit4_alpha": 0.770584425330162,
30
+ "exit4_greedy": 0.7579452514648438,
31
+ "exit5_alpha": 0.8510826379060744,
32
+ "exit5_greedy": 0.8368385314941407,
33
+ "exit6_alpha": 0.897156646847725,
34
+ "exit6_greedy": 0.884052276611328,
35
+ "exit7_alpha": 0.9390101432800292,
36
+ "exit7_greedy": 0.9295768737792968,
37
+ "iter": 1000
38
+ },
39
+ {
40
+ "val_ce": 1.578459221124649,
41
+ "exit1_alpha": 0.40710014700889596,
42
+ "exit1_greedy": 0.42163848876953125,
43
+ "exit2_alpha": 0.5925680071115492,
44
+ "exit2_greedy": 0.6041282653808594,
45
+ "exit3_alpha": 0.7103522688150407,
46
+ "exit3_greedy": 0.7058708190917969,
47
+ "exit4_alpha": 0.7660878688097,
48
+ "exit4_greedy": 0.7575103759765625,
49
+ "exit5_alpha": 0.8463849425315857,
50
+ "exit5_greedy": 0.8351921081542968,
51
+ "exit6_alpha": 0.8927793353796003,
52
+ "exit6_greedy": 0.8805702209472657,
53
+ "exit7_alpha": 0.9349214017391205,
54
+ "exit7_greedy": 0.9257751464843752,
55
+ "iter": 2000
56
+ },
57
+ {
58
+ "val_ce": 1.5022810697555542,
59
+ "exit1_alpha": 0.4170039817690849,
60
+ "exit1_greedy": 0.44345245361328134,
61
+ "exit2_alpha": 0.5942453950643539,
62
+ "exit2_greedy": 0.6097404479980467,
63
+ "exit3_alpha": 0.708651602268219,
64
+ "exit3_greedy": 0.7068855285644531,
65
+ "exit4_alpha": 0.7627668380737305,
66
+ "exit4_greedy": 0.7567466735839844,
67
+ "exit5_alpha": 0.8417250126600265,
68
+ "exit5_greedy": 0.8329322814941403,
69
+ "exit6_alpha": 0.8888692647218703,
70
+ "exit6_greedy": 0.8786048889160155,
71
+ "exit7_alpha": 0.9317562252283097,
72
+ "exit7_greedy": 0.9234054565429688,
73
+ "iter": 3000
74
+ },
75
+ {
76
+ "val_ce": 1.4586917817592617,
77
+ "exit1_alpha": 0.433258655667305,
78
+ "exit1_greedy": 0.4643768310546875,
79
+ "exit2_alpha": 0.6047763884067535,
80
+ "exit2_greedy": 0.6220146179199219,
81
+ "exit3_alpha": 0.7129614770412446,
82
+ "exit3_greedy": 0.7139602661132812,
83
+ "exit4_alpha": 0.765247431397438,
84
+ "exit4_greedy": 0.7619926452636717,
85
+ "exit5_alpha": 0.8431548476219176,
86
+ "exit5_greedy": 0.8360275268554689,
87
+ "exit6_alpha": 0.8900604933500288,
88
+ "exit6_greedy": 0.8808296203613284,
89
+ "exit7_alpha": 0.9319884389638898,
90
+ "exit7_greedy": 0.9244018554687499,
91
+ "iter": 4000
92
+ },
93
+ {
94
+ "val_ce": 1.433526623249054,
95
+ "exit1_alpha": 0.44229043573141097,
96
+ "exit1_greedy": 0.4734825134277343,
97
+ "exit2_alpha": 0.6097689032554625,
98
+ "exit2_greedy": 0.6249656677246095,
99
+ "exit3_alpha": 0.7152164906263351,
100
+ "exit3_greedy": 0.7157157897949219,
101
+ "exit4_alpha": 0.7679181784391402,
102
+ "exit4_greedy": 0.7638214111328127,
103
+ "exit5_alpha": 0.842146846652031,
104
+ "exit5_greedy": 0.8342453002929687,
105
+ "exit6_alpha": 0.8888639360666275,
106
+ "exit6_greedy": 0.8797622680664063,
107
+ "exit7_alpha": 0.930886721611023,
108
+ "exit7_greedy": 0.9237228393554686,
109
+ "iter": 5000
110
+ },
111
+ {
112
+ "val_ce": 1.39659908413887,
113
+ "exit1_alpha": 0.4487991124391556,
114
+ "exit1_greedy": 0.4879180908203125,
115
+ "exit2_alpha": 0.6146820276975632,
116
+ "exit2_greedy": 0.6330322265624999,
117
+ "exit3_alpha": 0.717953586578369,
118
+ "exit3_greedy": 0.7202896118164063,
119
+ "exit4_alpha": 0.7694938749074937,
120
+ "exit4_greedy": 0.7674110412597656,
121
+ "exit5_alpha": 0.84256731569767,
122
+ "exit5_greedy": 0.8361511230468749,
123
+ "exit6_alpha": 0.8886137694120406,
124
+ "exit6_greedy": 0.8800559997558592,
125
+ "exit7_alpha": 0.9299631059169766,
126
+ "exit7_greedy": 0.9224426269531251,
127
+ "iter": 6000
128
+ },
129
+ {
130
+ "val_ce": 1.383489817380905,
131
+ "exit1_alpha": 0.45552693754434587,
132
+ "exit1_greedy": 0.4913726806640625,
133
+ "exit2_alpha": 0.6181112408638,
134
+ "exit2_greedy": 0.6341178894042968,
135
+ "exit3_alpha": 0.7176161557435988,
136
+ "exit3_greedy": 0.7191116333007813,
137
+ "exit4_alpha": 0.7689908534288407,
138
+ "exit4_greedy": 0.7660240173339844,
139
+ "exit5_alpha": 0.8409761995077134,
140
+ "exit5_greedy": 0.834588623046875,
141
+ "exit6_alpha": 0.8872842937707902,
142
+ "exit6_greedy": 0.8789833068847654,
143
+ "exit7_alpha": 0.9289374917745589,
144
+ "exit7_greedy": 0.9214012145996094,
145
+ "iter": 7000
146
+ },
147
+ {
148
+ "val_ce": 1.3478731572628022,
149
+ "exit1_alpha": 0.4623634889721871,
150
+ "exit1_greedy": 0.4993957519531251,
151
+ "exit2_alpha": 0.621457040309906,
152
+ "exit2_greedy": 0.6377197265625,
153
+ "exit3_alpha": 0.7188945442438125,
154
+ "exit3_greedy": 0.7204681396484375,
155
+ "exit4_alpha": 0.7694859355688096,
156
+ "exit4_greedy": 0.7671745300292967,
157
+ "exit5_alpha": 0.8412264496088029,
158
+ "exit5_greedy": 0.8346000671386719,
159
+ "exit6_alpha": 0.886994031071663,
160
+ "exit6_greedy": 0.8790596008300781,
161
+ "exit7_alpha": 0.9285726338624954,
162
+ "exit7_greedy": 0.9216156005859374,
163
+ "iter": 8000
164
+ },
165
+ {
166
+ "val_ce": 1.3483326613903046,
167
+ "exit1_alpha": 0.4615658625960351,
168
+ "exit1_greedy": 0.5015533447265625,
169
+ "exit2_alpha": 0.6214722603559494,
170
+ "exit2_greedy": 0.6372215270996094,
171
+ "exit3_alpha": 0.7183588057756425,
172
+ "exit3_greedy": 0.7196212768554687,
173
+ "exit4_alpha": 0.768870809674263,
174
+ "exit4_greedy": 0.7661483764648437,
175
+ "exit5_alpha": 0.839602640271187,
176
+ "exit5_greedy": 0.8330055236816406,
177
+ "exit6_alpha": 0.8852128148078918,
178
+ "exit6_greedy": 0.8765960693359375,
179
+ "exit7_alpha": 0.9268552124500274,
180
+ "exit7_greedy": 0.9186050415039063,
181
+ "iter": 9000
182
+ },
183
+ {
184
+ "val_ce": 1.3316114962100982,
185
+ "exit1_alpha": 0.4700911849737167,
186
+ "exit1_greedy": 0.5082527160644531,
187
+ "exit2_alpha": 0.6264044821262361,
188
+ "exit2_greedy": 0.6428421020507813,
189
+ "exit3_alpha": 0.7207205474376678,
190
+ "exit3_greedy": 0.7237045288085937,
191
+ "exit4_alpha": 0.7705703556537628,
192
+ "exit4_greedy": 0.7688064575195312,
193
+ "exit5_alpha": 0.8408509403467177,
194
+ "exit5_greedy": 0.8349418640136721,
195
+ "exit6_alpha": 0.886443716287613,
196
+ "exit6_greedy": 0.8784645080566407,
197
+ "exit7_alpha": 0.9278432577848436,
198
+ "exit7_greedy": 0.9202407836914063,
199
+ "iter": 10000
200
+ },
201
+ {
202
+ "val_ce": 1.3176176846027374,
203
+ "exit1_alpha": 0.4697788402438164,
204
+ "exit1_greedy": 0.5085655212402345,
205
+ "exit2_alpha": 0.6254494339227676,
206
+ "exit2_greedy": 0.64173583984375,
207
+ "exit3_alpha": 0.7195282131433487,
208
+ "exit3_greedy": 0.7223030090332031,
209
+ "exit4_alpha": 0.768567582964897,
210
+ "exit4_greedy": 0.7674957275390626,
211
+ "exit5_alpha": 0.8386009097099303,
212
+ "exit5_greedy": 0.8323188781738281,
213
+ "exit6_alpha": 0.8845931142568589,
214
+ "exit6_greedy": 0.8763595581054688,
215
+ "exit7_alpha": 0.9260558307170867,
216
+ "exit7_greedy": 0.9184646606445311,
217
+ "iter": 11000
218
+ },
219
+ {
220
+ "val_ce": 1.3059154570102693,
221
+ "exit1_alpha": 0.4728017628192901,
222
+ "exit1_greedy": 0.512060546875,
223
+ "exit2_alpha": 0.6275364607572556,
224
+ "exit2_greedy": 0.6449378967285156,
225
+ "exit3_alpha": 0.7219109445810318,
226
+ "exit3_greedy": 0.7250289916992186,
227
+ "exit4_alpha": 0.771248546242714,
228
+ "exit4_greedy": 0.7697929382324218,
229
+ "exit5_alpha": 0.8400995552539825,
230
+ "exit5_greedy": 0.8334976196289063,
231
+ "exit6_alpha": 0.8853256195783613,
232
+ "exit6_greedy": 0.8770217895507812,
233
+ "exit7_alpha": 0.9262120515108109,
234
+ "exit7_greedy": 0.9189743041992188,
235
+ "iter": 12000
236
+ },
237
+ {
238
+ "val_ce": 1.2948933601379393,
239
+ "exit1_alpha": 0.4731743648648262,
240
+ "exit1_greedy": 0.5167854309082032,
241
+ "exit2_alpha": 0.6282133609056473,
242
+ "exit2_greedy": 0.6462585449218751,
243
+ "exit3_alpha": 0.7214778035879135,
244
+ "exit3_greedy": 0.7259895324707033,
245
+ "exit4_alpha": 0.7699974924325942,
246
+ "exit4_greedy": 0.7699966430664062,
247
+ "exit5_alpha": 0.8384121149778363,
248
+ "exit5_greedy": 0.8336082458496096,
249
+ "exit6_alpha": 0.884169813990593,
250
+ "exit6_greedy": 0.8764663696289063,
251
+ "exit7_alpha": 0.9253626286983491,
252
+ "exit7_greedy": 0.9174041748046873,
253
+ "iter": 13000
254
+ },
255
+ {
256
+ "val_ce": 1.2827632665634154,
257
+ "exit1_alpha": 0.47814678400754934,
258
+ "exit1_greedy": 0.5198051452636718,
259
+ "exit2_alpha": 0.6312671095132828,
260
+ "exit2_greedy": 0.6491500854492187,
261
+ "exit3_alpha": 0.7226436764001848,
262
+ "exit3_greedy": 0.72613525390625,
263
+ "exit4_alpha": 0.7707418739795684,
264
+ "exit4_greedy": 0.7703193664550783,
265
+ "exit5_alpha": 0.8391160368919371,
266
+ "exit5_greedy": 0.8340530395507814,
267
+ "exit6_alpha": 0.8843333840370178,
268
+ "exit6_greedy": 0.8766525268554689,
269
+ "exit7_alpha": 0.9253863930702211,
270
+ "exit7_greedy": 0.9177177429199219,
271
+ "iter": 14000
272
+ },
273
+ {
274
+ "val_ce": 1.2767474710941316,
275
+ "exit1_alpha": 0.4784462839365005,
276
+ "exit1_greedy": 0.5214790344238283,
277
+ "exit2_alpha": 0.6313386857509614,
278
+ "exit2_greedy": 0.6499969482421876,
279
+ "exit3_alpha": 0.7227032780647278,
280
+ "exit3_greedy": 0.7276084899902343,
281
+ "exit4_alpha": 0.7713228404521941,
282
+ "exit4_greedy": 0.7715507507324219,
283
+ "exit5_alpha": 0.8385546714067458,
284
+ "exit5_greedy": 0.8338310241699219,
285
+ "exit6_alpha": 0.8839981228113175,
286
+ "exit6_greedy": 0.8766891479492187,
287
+ "exit7_alpha": 0.9249983906745911,
288
+ "exit7_greedy": 0.9178123474121093,
289
+ "iter": 15000
290
+ },
291
+ {
292
+ "val_ce": 1.2510948657989502,
293
+ "exit1_alpha": 0.4791820466518403,
294
+ "exit1_greedy": 0.522113037109375,
295
+ "exit2_alpha": 0.6319353550672531,
296
+ "exit2_greedy": 0.6504577636718749,
297
+ "exit3_alpha": 0.7221516221761702,
298
+ "exit3_greedy": 0.7268997192382812,
299
+ "exit4_alpha": 0.7699011594057082,
300
+ "exit4_greedy": 0.7703872680664061,
301
+ "exit5_alpha": 0.8373788088560106,
302
+ "exit5_greedy": 0.8321022033691405,
303
+ "exit6_alpha": 0.88275009393692,
304
+ "exit6_greedy": 0.8752052307128906,
305
+ "exit7_alpha": 0.9238883793354034,
306
+ "exit7_greedy": 0.9170501708984374,
307
+ "iter": 16000
308
+ },
309
+ {
310
+ "val_ce": 1.2593918740749364,
311
+ "exit1_alpha": 0.48289879858493817,
312
+ "exit1_greedy": 0.5271568298339844,
313
+ "exit2_alpha": 0.6351416736841201,
314
+ "exit2_greedy": 0.654430389404297,
315
+ "exit3_alpha": 0.7253278106451035,
316
+ "exit3_greedy": 0.7305152893066407,
317
+ "exit4_alpha": 0.7729058593511582,
318
+ "exit4_greedy": 0.7742111206054688,
319
+ "exit5_alpha": 0.8391105562448502,
320
+ "exit5_greedy": 0.8342247009277343,
321
+ "exit6_alpha": 0.8841602981090545,
322
+ "exit6_greedy": 0.8770401000976563,
323
+ "exit7_alpha": 0.9245910406112671,
324
+ "exit7_greedy": 0.9174201965332032,
325
+ "iter": 17000
326
+ },
327
+ {
328
+ "val_ce": 1.2452915370464326,
329
+ "exit1_alpha": 0.4827954232692719,
330
+ "exit1_greedy": 0.5270515441894532,
331
+ "exit2_alpha": 0.6341491907835006,
332
+ "exit2_greedy": 0.6547569274902343,
333
+ "exit3_alpha": 0.72407668530941,
334
+ "exit3_greedy": 0.7304679870605468,
335
+ "exit4_alpha": 0.7716502159833908,
336
+ "exit4_greedy": 0.773052978515625,
337
+ "exit5_alpha": 0.8385924726724626,
338
+ "exit5_greedy": 0.8343406677246094,
339
+ "exit6_alpha": 0.8831680685281753,
340
+ "exit6_greedy": 0.8763328552246094,
341
+ "exit7_alpha": 0.9239696323871613,
342
+ "exit7_greedy": 0.9172851562500002,
343
+ "iter": 18000
344
+ },
345
+ {
346
+ "val_ce": 1.2473996341228482,
347
+ "exit1_alpha": 0.484264649450779,
348
+ "exit1_greedy": 0.5282669067382812,
349
+ "exit2_alpha": 0.6363260746002196,
350
+ "exit2_greedy": 0.6554977416992187,
351
+ "exit3_alpha": 0.7256934702396393,
352
+ "exit3_greedy": 0.7315994262695313,
353
+ "exit4_alpha": 0.7724919855594633,
354
+ "exit4_greedy": 0.7737304687500001,
355
+ "exit5_alpha": 0.8385471880435944,
356
+ "exit5_greedy": 0.834428405761719,
357
+ "exit6_alpha": 0.8831394642591474,
358
+ "exit6_greedy": 0.8758834838867188,
359
+ "exit7_alpha": 0.9239964902400971,
360
+ "exit7_greedy": 0.9169631958007811,
361
+ "iter": 19000
362
+ },
363
+ {
364
+ "val_ce": 1.2464708268642428,
365
+ "exit1_alpha": 0.4845910012722016,
366
+ "exit1_greedy": 0.5295845031738281,
367
+ "exit2_alpha": 0.6358921855688096,
368
+ "exit2_greedy": 0.6550628662109373,
369
+ "exit3_alpha": 0.7252050876617432,
370
+ "exit3_greedy": 0.7312286376953127,
371
+ "exit4_alpha": 0.7719815969467161,
372
+ "exit4_greedy": 0.772979736328125,
373
+ "exit5_alpha": 0.8381113588809967,
374
+ "exit5_greedy": 0.8338546752929686,
375
+ "exit6_alpha": 0.8831344306468962,
376
+ "exit6_greedy": 0.8760437011718749,
377
+ "exit7_alpha": 0.9238532721996306,
378
+ "exit7_greedy": 0.9169792175292967,
379
+ "iter": 20000
380
+ }
381
+ ]
382
+ }
target/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:92219629bfd8d9a35ac6139edd833c6b4400e1970558fc5cf0509bab6928fd0e
3
+ size 109091456
target/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff