Emma974 commited on
Commit
df22f29
·
verified ·
1 Parent(s): 7257f57

Add GPT-2 Small Mixed FT LOO without ArXiv

Browse files
gpt2-small/Mixed_FT_LOO_without_ArXiv/model/model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e3748079aa759e41a38e7bc9ecc2f0d754b96dbcf82487378163cba91d33a721
3
  size 497774208
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3766aba88da20bf33dd086a4df2794b269f7657c1e443896225dc55ffbcdd5c1
3
  size 497774208
gpt2-small/Mixed_FT_LOO_without_ArXiv/model/source_dataset_stats.csv CHANGED
@@ -1,11 +1,10 @@
1
  num_documents_used,raw_tokens_seen,requested_token_budget,usable_token_budget,num_sequences,context_length,effective_tokens,source,hf_dataset,hf_split,target_total_sequences,target_train_sequences,target_val_sequences,available_unique_sequences,unique_train_sequences,unique_val_sequences,num_train_sequences,num_val_sequences,train_was_upsampled,val_was_upsampled,train_extra_resampled_sequences,val_extra_resampled_sequences,train_upsampling_factor,val_upsampling_factor,train_tokens,val_tokens,final_mixture_tokens
2
- 131,512162,512000,512000,1000,512,512000,DM Mathematics,timaeus/pile-dm_mathematics,train,1000,900,100,1000,900,100,900,100,False,False,0,0,1.0,1.0,460800,51200,512000
3
- 16704,495307,512000,512000,1000,512,512000,EuroParl,Helsinki-NLP/europarl,train,1000,900,100,1000,900,100,900,100,False,False,0,0,1.0,1.0,460800,51200,512000
4
- 118,512941,512000,512000,1000,512,512000,FreeLaw,timaeus/pile-freelaw,train,1000,900,100,1000,900,100,900,100,False,False,0,0,1.0,1.0,460800,51200,512000
5
- 237,535171,512000,512000,1000,512,512000,GitHub,timaeus/pile-github,train,1000,900,100,1000,900,100,900,100,False,False,0,0,1.0,1.0,460800,51200,512000
6
- 1,1200631,512000,512000,1000,512,512000,PG-19,emozilla/pg19,train,1000,900,100,1000,900,100,900,100,False,False,0,0,1.0,1.0,460800,51200,512000
7
- 524,512454,512000,512000,1000,512,512000,OWT2,suolyer/pile_openwebtext2,validation,1000,900,100,1000,900,100,900,100,False,False,0,0,1.0,1.0,460800,51200,512000
8
- 1768,510522,512000,512000,1000,512,512000,PubMed Abstracts,timaeus/pile-pubmed_abstracts,train,1000,900,100,1000,900,100,900,100,False,False,0,0,1.0,1.0,460800,51200,512000
9
- 65,515020,512000,512000,1000,512,512000,PubMed Central,datajuicer/the-pile-pubmed-central-refined-by-data-juicer,train,1000,900,100,1000,900,100,900,100,False,False,0,0,1.0,1.0,460800,51200,512000
10
- 2505,509582,512000,512000,1000,512,512000,StackExchange,flax-sentence-embeddings/stackexchange_title_body_jsonl,train,1000,900,100,1000,900,100,900,100,False,False,0,0,1.0,1.0,460800,51200,512000
11
- 75,518416,512000,512000,1000,512,512000,Wikipedia,wikimedia/wikipedia,train,1000,900,100,1000,900,100,900,100,False,False,0,0,1.0,1.0,460800,51200,512000
 
1
  num_documents_used,raw_tokens_seen,requested_token_budget,usable_token_budget,num_sequences,context_length,effective_tokens,source,hf_dataset,hf_split,target_total_sequences,target_train_sequences,target_val_sequences,available_unique_sequences,unique_train_sequences,unique_val_sequences,num_train_sequences,num_val_sequences,train_was_upsampled,val_was_upsampled,train_extra_resampled_sequences,val_extra_resampled_sequences,train_upsampling_factor,val_upsampling_factor,train_tokens,val_tokens,final_mixture_tokens
2
+ 146,572733,569344,569344,1112,512,569344,DM Mathematics,timaeus/pile-dm_mathematics,train,1112,1000,112,1112,1000,112,1000,112,False,False,0,0,1.0,1.0,512000,57344,569344
3
+ 18574,550276,568832,568832,1111,512,568832,EuroParl,Helsinki-NLP/europarl,train,1111,1000,111,1111,1000,111,1000,111,False,False,0,0,1.0,1.0,512000,56832,568832
4
+ 254,569039,568832,568832,1111,512,568832,GitHub,timaeus/pile-github,train,1111,1000,111,1111,1000,111,1000,111,False,False,0,0,1.0,1.0,512000,56832,568832
5
+ 1,1200631,568832,568832,1111,512,568832,PG-19,emozilla/pg19,train,1111,1000,111,1111,1000,111,1000,111,False,False,0,0,1.0,1.0,512000,56832,568832
6
+ 573,568366,568832,568832,1111,512,568832,OWT2,suolyer/pile_openwebtext2,validation,1111,1000,111,1111,1000,111,1000,111,False,False,0,0,1.0,1.0,512000,56832,568832
7
+ 1965,566870,568832,568832,1111,512,568832,PubMed Abstracts,timaeus/pile-pubmed_abstracts,train,1111,1000,111,1111,1000,111,1000,111,False,False,0,0,1.0,1.0,512000,56832,568832
8
+ 73,578162,568832,568832,1111,512,568832,PubMed Central,datajuicer/the-pile-pubmed-central-refined-by-data-juicer,train,1111,1000,111,1111,1000,111,1000,111,False,False,0,0,1.0,1.0,512000,56832,568832
9
+ 2792,566077,568832,568832,1111,512,568832,StackExchange,flax-sentence-embeddings/stackexchange_title_body_jsonl,train,1111,1000,111,1111,1000,111,1000,111,False,False,0,0,1.0,1.0,512000,56832,568832
10
+ 82,570941,568832,568832,1111,512,568832,Wikipedia,wikimedia/wikipedia,train,1111,1000,111,1111,1000,111,1000,111,False,False,0,0,1.0,1.0,512000,56832,568832
 
gpt2-small/Mixed_FT_LOO_without_ArXiv/model/training_history.csv CHANGED
@@ -1,3 +1,3 @@
1
  epoch,optimizer_step,cumulative_tokens_seen,optimization_loss,train_loss,val_loss,train_ppl,val_ppl,train_acc,val_acc,train_eval_tokens,val_eval_tokens,learning_rate
2
- 1,563,4608000,2.8710845625135635,2.6182988525781377,2.6666656621366434,13.71237695797125,14.391901638045345,0.48472042088151274,0.47924279084760485,4579151,508763,5e-05
3
- 2,1126,9216000,2.711718646155463,2.5176119115834052,2.619658354975389,12.398951487590203,13.73103164499028,0.4976972805657643,0.4855207631058076,4579151,508763,5e-05
 
1
  epoch,optimizer_step,cumulative_tokens_seen,optimization_loss,train_loss,val_loss,train_ppl,val_ppl,train_acc,val_acc,train_eval_tokens,val_eval_tokens,learning_rate
2
+ 1,563,4608000,2.877319888856676,2.6247260226666445,2.6966310096508392,13.800792563706548,14.829686400735582,0.4814001421867838,0.4737239555166616,4577078,508505,5e-05
3
+ 2,1126,9216000,2.715038715786404,2.5237715881367846,2.648399594960042,12.475560720178658,14.13140456726937,0.4938589641688431,0.47967276624615296,4577078,508505,5e-05
gpt2-small/Mixed_FT_LOO_without_ArXiv/model/training_summary.csv CHANGED
@@ -1,2 +1,2 @@
1
  method,model_name,hf_repo_id,hf_subfolder,held_out_target,sources,num_sources,mixture_type,seed,context_length,requested_total_max_tokens,effective_total_source_tokens,batch_size,gradient_accumulation_steps,effective_batch_size,learning_rate,weight_decay,val_split_ratio,max_epochs,final_optimizer_step,final_tokens_seen,final_train_loss,final_val_loss,final_train_ppl,final_val_ppl,final_train_acc,final_val_acc
2
- Mixed FT Leave-One-Out,gpt2,alignment-decision-lab/robustness-model-bank,gpt2-small/Mixed_FT_LOO_without_ArXiv/model,ArXiv,DM Mathematics | EuroParl | FreeLaw | GitHub | PG-19 | OWT2 | PubMed Abstracts | PubMed Central | StackExchange | Wikipedia,10,uniform_source_mixture,42,512,5120000,5120000,8,2,16,5e-05,0.0,0.1,2,1126,9216000,2.5176119115834052,2.619658354975389,12.398951487590203,13.73103164499028,0.4976972805657643,0.4855207631058076
 
1
  method,model_name,hf_repo_id,hf_subfolder,held_out_target,sources,num_sources,mixture_type,seed,context_length,requested_total_max_tokens,effective_total_source_tokens,batch_size,gradient_accumulation_steps,effective_batch_size,learning_rate,weight_decay,val_split_ratio,max_epochs,final_optimizer_step,final_tokens_seen,final_train_loss,final_val_loss,final_train_ppl,final_val_ppl,final_train_acc,final_val_acc
2
+ Mixed FT Leave-One-Out,gpt2,alignment-decision-lab/robustness-model-bank,gpt2-small/Mixed_FT_LOO_without_ArXiv/model,ArXiv,DM Mathematics | EuroParl | GitHub | PG-19 | OWT2 | PubMed Abstracts | PubMed Central | StackExchange | Wikipedia,9,uniform_source_mixture,42,512,5120000,5120000,8,2,16,5e-05,0.0,0.1,2,1126,9216000,2.5237715881367846,2.648399594960042,12.475560720178658,14.13140456726937,0.4938589641688431,0.47967276624615296