Hugging Face's logo

lapp0
/
distily_bench_obj_cross_v2.13_gpt2

TensorBoard
Safetensors
Distily
gpt2
Generated from Trainer
8-bit precision
bitsandbytes
Model card Files Metrics Community
lapp0 commited on
Commit
62c0f77
•
1 Parent(s): 0b6f5f3

End of training

Browse files
Files changed (2) hide show
  1. README.md +20 -20
  2. logs/hs_layer_mapper=uniform_last, hs_loss_fn=kl, hs_weight=1.0/events.out.tfevents.1724102840.f383272e719b +3 -0
README.md CHANGED
@@ -16,14 +16,14 @@ This student model is distilled from the teacher model [gpt2](https://huggingfac
16
  The [Distily](https://github.com/lapp0/distily) library was used for this distillation.
17
 
18
  It achieves the following results on the evaluation set:
19
- - eval_enwikippl: 2144.0
20
- - eval_frwikippl: 8704.0
21
- - eval_zhwikippl: 148480.0
22
- - eval_tinystoriesppl: 1648.0
23
- - eval_loss: 3.2443
24
- - eval_runtime: 12.9651
25
- - eval_samples_per_second: 46.278
26
- - eval_steps_per_second: 11.569
27
 
28
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
29
  should probably proofread and complete it, then remove this comment.
@@ -46,7 +46,7 @@ More information needed
46
  ### Training hyperparameters
47
 
48
  The following hyperparameters were used during training:
49
- - distillation_objective: DistillationObjective(logits_loss_component=LossComponent(label=logits, weight=1, loss_fn=kl, layer_mapper=None, projector=None), hs_loss_component=LossComponent(label=hs, weight=1.0, loss_fn=kl, layer_mapper=uniform_cons, projector=None), attn_loss_component=LossComponent(label=attn, weight=0, loss_fn=None, layer_mapper=None, projector=None))
50
  - train_embeddings: True
51
  - learning_rate: 0.0001
52
  - train_batch_size: 8
@@ -64,17 +64,17 @@ Peak GPU Memory: 8.0905 GB
64
  | step | epoch | enwikippl | frwikippl | loss | runtime | samples_per_second | steps_per_second | tinystoriesppl | zhwikippl |
65
  | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
66
  | **teacher eval** | | 43.75 | 61.75 | | | | | 11.8125 | 19.125 |
67
- | 0 | 0 | 1821066133504.0 | 158329674399744.0 | 25.4650 | 12.9053 | 46.493 | 11.623 | 12079595520.0 | 98956046499840.0 |
68
- | 750 | 0.1010 | 2144.0 | 8704.0 | 3.2443 | 12.9651 | 46.278 | 11.569 | 1648.0 | 148480.0 |
69
- | 1500 | 0.2020 | 776.0 | 4800.0 | 2.2942 | 12.9521 | 46.325 | 11.581 | 568.0 | 4736.0 |
70
- | 2250 | 0.3030 | 448.0 | 2832.0 | 1.9334 | 12.9674 | 46.27 | 11.568 | 358.0 | 592.0 |
71
- | 3000 | 0.4040 | 318.0 | 1424.0 | 1.6643 | 13.0049 | 46.136 | 11.534 | 256.0 | 308.0 |
72
- | 3750 | 0.5051 | 249.0 | 912.0 | 1.4761 | 12.9665 | 46.273 | 11.568 | 198.0 | 474.0 |
73
- | 4500 | 0.6061 | 188.0 | 684.0 | 1.2711 | 12.9804 | 46.224 | 11.556 | 152.0 | 354.0 |
74
- | 5250 | 0.7071 | 147.0 | 560.0 | 1.1017 | 12.9809 | 46.222 | 11.555 | 116.0 | 218.0 |
75
- | 6000 | 0.8081 | 134.0 | 490.0 | 1.0242 | 12.9725 | 46.252 | 11.563 | 105.5 | 186.0 |
76
- | 6750 | 0.9091 | 125.5 | 464.0 | 0.9844 | 12.9741 | 46.246 | 11.561 | 99.0 | 175.0 |
77
- | 7425 | 1.0 | 124.0 | 462.0 | 0.9768 | 12.938 | 46.375 | 11.594 | 97.5 | 165.0 |
78
 
79
  ### Framework versions
80
  - Distily 0.2.0
 
16
  The [Distily](https://github.com/lapp0/distily) library was used for this distillation.
17
 
18
  It achieves the following results on the evaluation set:
19
+ - eval_enwikippl: 2192.0
20
+ - eval_frwikippl: 9344.0
21
+ - eval_zhwikippl: 165888.0
22
+ - eval_tinystoriesppl: 1544.0
23
+ - eval_loss: 3.2467
24
+ - eval_runtime: 12.9561
25
+ - eval_samples_per_second: 46.31
26
+ - eval_steps_per_second: 11.578
27
 
28
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
29
  should probably proofread and complete it, then remove this comment.
 
46
  ### Training hyperparameters
47
 
48
  The following hyperparameters were used during training:
49
+ - distillation_objective: DistillationObjective(logits_loss_component=LossComponent(label=logits, weight=1, loss_fn=kl, layer_mapper=None, projector=None), hs_loss_component=LossComponent(label=hs, weight=1.0, loss_fn=kl, layer_mapper=uniform+last, projector=None), attn_loss_component=LossComponent(label=attn, weight=0, loss_fn=None, layer_mapper=None, projector=None))
50
  - train_embeddings: True
51
  - learning_rate: 0.0001
52
  - train_batch_size: 8
 
64
  | step | epoch | enwikippl | frwikippl | loss | runtime | samples_per_second | steps_per_second | tinystoriesppl | zhwikippl |
65
  | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
66
  | **teacher eval** | | 43.75 | 61.75 | | | | | 11.8125 | 19.125 |
67
+ | 0 | 0 | 1821066133504.0 | 158329674399744.0 | 25.4650 | 12.9114 | 46.47 | 11.618 | 12079595520.0 | 98956046499840.0 |
68
+ | 750 | 0.1010 | 2192.0 | 9344.0 | 3.2467 | 12.9561 | 46.31 | 11.578 | 1544.0 | 165888.0 |
69
+ | 1500 | 0.2020 | 776.0 | 4704.0 | 2.2915 | 12.9788 | 46.229 | 11.557 | 584.0 | 5152.0 |
70
+ | 2250 | 0.3030 | 454.0 | 2800.0 | 1.9392 | 13.006 | 46.133 | 11.533 | 358.0 | 604.0 |
71
+ | 3000 | 0.4040 | 320.0 | 1512.0 | 1.6692 | 12.9768 | 46.236 | 11.559 | 255.0 | 278.0 |
72
+ | 3750 | 0.5051 | 254.0 | 896.0 | 1.4808 | 13.0015 | 46.148 | 11.537 | 199.0 | 364.0 |
73
+ | 4500 | 0.6061 | 185.0 | 684.0 | 1.2715 | 13.0003 | 46.153 | 11.538 | 152.0 | 516.0 |
74
+ | 5250 | 0.7071 | 148.0 | 572.0 | 1.1020 | 12.9814 | 46.22 | 11.555 | 114.5 | 200.0 |
75
+ | 6000 | 0.8081 | 134.0 | 490.0 | 1.0245 | 12.9705 | 46.259 | 11.565 | 104.0 | 178.0 |
76
+ | 6750 | 0.9091 | 126.5 | 464.0 | 0.9858 | 13.0461 | 45.991 | 11.498 | 97.5 | 168.0 |
77
+ | 7425 | 1.0 | 124.5 | 462.0 | 0.9781 | 13.0812 | 45.867 | 11.467 | 96.5 | 163.0 |
78
 
79
  ### Framework versions
80
  - Distily 0.2.0
logs/hs_layer_mapper=uniform_last, hs_loss_fn=kl, hs_weight=1.0/events.out.tfevents.1724102840.f383272e719b ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:951902fee99ee2a40d21dbe83047fe539c68433df55b855eba1025219b6c1aa6
3
+ size 578