add: training_parameter.json for reproducability
Browse files- training_parameter.json +34 -0
training_parameter.json
ADDED
|
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"ppo_config": {
|
| 3 |
+
"model_name": "R-best-fine-tuned-bart-base-full-ft-reward_short_sentences_and_words-2023-07-13T06-49-08",
|
| 4 |
+
"steps": 20000,
|
| 5 |
+
"note": "Stopped manually after 65 steps",
|
| 6 |
+
"learning_rate": 1e-06,
|
| 7 |
+
"target": 6,
|
| 8 |
+
"batch_size": 64,
|
| 9 |
+
"mini_batch_size": 1,
|
| 10 |
+
"ppo_epochs": 4,
|
| 11 |
+
"gradient_accumulation_steps": 1,
|
| 12 |
+
"log_with": "tensorboard",
|
| 13 |
+
"accelerator_kwargs": {
|
| 14 |
+
"logging_dir": "/content/results/R-best-fine-tuned-bart-base-full-ft-reward_short_sentences_and_words-2023-07-13T06-49-08/runs"
|
| 15 |
+
},
|
| 16 |
+
"early_stopping": false,
|
| 17 |
+
"push_to_hub_if_best_kwargs": {
|
| 18 |
+
"repo_id": "nlp-lab-2023-seq2seq/R-best-fine-tuned-bart-base-full-ft-reward_short_sentences_and_words-2023-07-13T06-49-08",
|
| 19 |
+
"private": true,
|
| 20 |
+
"commit_message": "add: new best model"
|
| 21 |
+
},
|
| 22 |
+
"compare_steps": 100
|
| 23 |
+
},
|
| 24 |
+
"generation_kwargs": {
|
| 25 |
+
"max_new_tokens": 414,
|
| 26 |
+
"num_beams": 1,
|
| 27 |
+
"early_stopping": true,
|
| 28 |
+
"pad_token_id": 1,
|
| 29 |
+
"eos_token_id": 2
|
| 30 |
+
},
|
| 31 |
+
"translation_model": "nlp-lab-2023-seq2seq/R-facebook-bart-base-full-ft-without-tum-nlp-german-gpt2_easy-prior-pp-bb5c579f",
|
| 32 |
+
"reward_fn": "reward_bleu_short_sentences_and_words",
|
| 33 |
+
"finetuning_strategy": "full"
|
| 34 |
+
}
|