Update README.md

README.md

@@ -61,8 +61,8 @@ wandb_watch:
 wandb_run_id:
 wandb_log_model: 
 
-gradient_accumulation_steps: 8
-micro_batch_size: 1
+gradient_accumulation_steps: 2
+micro_batch_size: 4
 num_epochs: 1
 optimizer: paged_adamw_32bit
 lr_scheduler: constant
@@ -72,18 +72,18 @@ learning_rate: 0.00005
 
 train_on_inputs: true
 group_by_length: false
-bf16: false
+bf16: true
 fp16: false
-tf32: true
+tf32: false
 
 gradient_checkpointing: false
 early_stopping_patience:
 resume_from_checkpoint:
-auto_resume_from_checkpoints: false
+auto_resume_from_checkpoints: True
 local_rank:
 logging_steps: 1
 xformers_attention:
-flash_attention: false
+flash_attention: true
 flash_attn_cross_entropy: false
 flash_attn_rms_norm: true
 flash_attn_fuse_qkv: false
@@ -111,7 +111,7 @@ special_tokens:
 
 # TinyMistral-StructureEvaluator
 
-This model was further trained on the epfl-llm/guidelines and JeanKaddour/minipile datasets.
+This model was further trained on the epfl-llm/guidelines and JeanKaddour/minipile datasets for 1 epoch.
 
 ## Model description
 
@@ -131,10 +131,10 @@ More information needed
 
 The following hyperparameters were used during training:
 - learning_rate: 5e-05
-- train_batch_size: 1
-- eval_batch_size: 1
+- train_batch_size: 4
+- eval_batch_size: 4
 - seed: 42
-- gradient_accumulation_steps: 8
+- gradient_accumulation_steps: 2
 - total_train_batch_size: 8
 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
 - lr_scheduler_type: constant