Upload training_config.yml with huggingface_hub

training_config.yml

@@ -5,8 +5,8 @@ model:
   - v_proj
   apply_lora_to_mlp: false
   apply_lora_to_output: false
-  lora_rank: 16
-  lora_alpha: 32
+  lora_rank: 32
+  lora_alpha: 64
   perception_tokens: 2
   use_clip: false
 tokenizer:
@@ -24,9 +24,9 @@ checkpointer:
 resume_from_checkpoint: false
 interim_checkpoint_steps: 5000
 interim_gen_steps: null
-max_new_tokens: 100
+max_new_tokens: 170
 temperature: 0.8
-top_k: 230
+top_k: 200
 dataset:
   _component_: ds.EvenBatcher
   buffer_size: 36
@@ -77,7 +77,7 @@ inference:
     {video}
 
     Caption the previous video.'
-  max_new_tokens: 100
-  temperature: 0.6
-  top_k: 170
+  max_new_tokens: 300
+  temperature: 0.7
+  top_k: 250
   quantizer: null