End of training

README.md

@@ -1,6 +1,7 @@
 ---
-license: mit
 base_model: gpt2
+library_name: Distily
+license: mit
 tags:
 - generated_from_trainer
 model-index:
@@ -8,14 +9,23 @@ model-index:
   results: []
 ---
 
-<!-- This model card has been generated automatically according to the information the Trainer had access to. You
-should probably proofread and complete it, then remove this comment. -->
-
 # distily_bench_gpt2_attn
 
-This model is a fine-tuned version of [gpt2](https://huggingface.co/gpt2) on an unknown dataset.
+This student model is distilled from the teacher model [gpt2](https://huggingface.co/gpt2) using the dataset (unspecified).
+
+The [Distily](https://github.com/lapp0/distily) library was used for this distillation.
+
 It achieves the following results on the evaluation set:
-- Loss: 1.1698
+- eval_enwikippl: 212.2672
+- eval_frwikippl: 1352.8285
+- eval_zhwikippl: 811.8465
+- eval_loss: 1.2429
+- eval_runtime: 17.2351
+- eval_samples_per_second: 58.021
+- eval_steps_per_second: 7.253
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment.
 
 ## Model description
 
@@ -28,12 +38,15 @@ More information needed
 ## Training and evaluation data
 
 More information needed
+-->
 
 ## Training procedure
 
 ### Training hyperparameters
 
 The following hyperparameters were used during training:
+- 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=0, loss_fn=None, layer_mapper=None, projector=None), attn_loss_component=LossComponent(label=attn, weight=2.0, loss_fn=mse, layer_mapper=None, projector=None))
+- train_embeddings: True
 - learning_rate: 4e-05
 - train_batch_size: 8
 - eval_batch_size: 8
@@ -42,28 +55,30 @@ The following hyperparameters were used during training:
 - lr_scheduler_type: constant
 - num_epochs: 1.0
 
-### Training results
-
-| Training Loss | Epoch  | Step  | Validation Loss |
-|:-------------:|:------:|:-----:|:---------------:|
-| No log        | 0      | 0     | 6.0237          |
-| 2.0831        | 0.0808 | 1000  | 2.0503          |
-| 1.9875        | 0.1616 | 2000  | 1.8577          |
-| 1.7222        | 0.2424 | 3000  | 1.7444          |
-| 1.625         | 0.3232 | 4000  | 1.6547          |
-| 1.5466        | 0.4040 | 5000  | 1.5704          |
-| 1.482         | 0.4848 | 6000  | 1.4916          |
-| 1.3055        | 0.5657 | 7000  | 1.4296          |
-| 1.3487        | 0.6465 | 8000  | 1.3749          |
-| 1.5284        | 0.7273 | 9000  | 1.3261          |
-| 1.3073        | 0.8081 | 10000 | 1.2775          |
-| 1.2886        | 0.8889 | 11000 | 1.2243          |
-| 1.2192        | 0.9697 | 12000 | 1.1816          |
-
+### Resource Usage
+Peak GPU Memory: 8.2202 GB
+
+### Eval-Phase Metrics
+| step | epoch | enwikippl | frwikippl | loss | runtime | samples_per_second | steps_per_second | zhwikippl |
+| --- | --- | --- | --- | --- | --- | --- | --- | --- |
+| **teacher eval** |  | 30.2086 | 57.2728 |  |  |  |  | 18.1784 |
+| 0 | 0 | 58954.875 | 57690.6602 | 5.9433 | 17.2113 | 58.101 | 7.263 | 54707.1133 |
+| 1000 | 0.0808 | 710.2846 | 4396.9824 | 1.9296 | 17.117 | 58.421 | 7.303 | 18078.0879 |
+| 2000 | 0.1616 | 503.9694 | 3086.3142 | 1.7454 | 17.2903 | 57.836 | 7.229 | 2703.2686 |
+| 3000 | 0.2424 | 420.6563 | 2955.9736 | 1.6378 | 17.2224 | 58.064 | 7.258 | 1477.8748 |
+| 4000 | 0.3232 | 367.0064 | 2704.3167 | 1.5544 | 17.2208 | 58.069 | 7.259 | 851.8279 |
+| 5000 | 0.4040 | 317.6482 | 2113.5315 | 1.4722 | 17.1418 | 58.337 | 7.292 | 1214.1425 |
+| 6000 | 0.4848 | 276.7272 | 1629.8280 | 1.3995 | 17.1258 | 58.392 | 7.299 | 813.5826 |
+| 7000 | 0.5657 | 250.8947 | 1553.0933 | 1.3412 | 17.2475 | 57.979 | 7.247 | 773.1216 |
+| 8000 | 0.6465 | 228.6603 | 1347.1174 | 1.2915 | 17.2115 | 58.101 | 7.263 | 716.5538 |
+| 9000 | 0.7273 | 212.2672 | 1352.8285 | 1.2429 | 17.2351 | 58.021 | 7.253 | 811.8465 |
+| 10000 | 0.8081 | 193.2158 | 1189.5732 | 1.1981 | 17.1888 | 58.177 | 7.272 | 670.6308 |
+| 11000 | 0.8889 | 178.6132 | 1058.1842 | 1.1502 | 17.2336 | 58.026 | 7.253 | 653.9169 |
+| 12000 | 0.9697 | 165.5636 | 977.5611 | 1.1143 | 17.2114 | 58.101 | 7.263 | 509.6881 |
+| 12375 | 1.0 | 160.1887 | 948.9035 | 1.0983 | 17.1765 | 58.219 | 7.277 | 518.8907 |
 
 ### Framework versions
-
+- Distily 0.2.0
 - Transformers 4.44.0
 - Pytorch 2.3.0
 - Datasets 2.20.0
-- Tokenizers 0.19.1

logs/attn_loss_fn=mse, attn_weight=2.0/events.out.tfevents.1723660225.93d6cbb3ad53

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:17a33e40555070df61c3730cf97614fbbead92007ee047df21579527a03296a1
+size 249