Update model_config.yaml

model_config.yaml

@@ -80,7 +80,7 @@ reduce_amax: true
 use_emha: false
 optim:
   name: distributed_fused_adam
-  lr: 3.002e-07
+  lr: 3.001e-07
   weight_decay: 0.1
   betas:
   - 0.9
@@ -110,7 +110,7 @@ data:
   num_workers: 2
   dataloader_type: single
   train_ds:
-    file_path: /dataset/daring-anteater_commercial.shuf.removelong.jsonl
+    file_path: /dataset/train.jsonl
     global_batch_size: 128
     micro_batch_size: 1
     shuffle: true
@@ -153,7 +153,7 @@ data:
     hf_dataset: true
     truncation_method: right
   validation_ds:
-    file_path: /dataset/daring-anteater_commercial.shuf.removelong.jsonl
+    file_path: /dataset/val.jsonl
     names: null
     global_batch_size: 128
     micro_batch_size: 1
@@ -238,13 +238,13 @@ data:
   index_mapping_dir: /indexmap_dir
   data_prefix:
     train:
-    - /datasets/v30_benign-walrus_clip153600.jsonl
+    - /datasets/train.jsonl
     validation:
-    - /datasets/v30_benign-walrus_clip153600.jsonl
+    - /datasets/val.jsonl
     test:
-    - /datasets/v30_benign-walrus_clip153600.jsonl
+    - /datasets/val.jsonl
 answer_only_loss: true
-restore_from_path: /models/340B_100p_CT_100B
+restore_from_path: /models/340B_base
 save_nemo_on_validation_end: true
 use_flash_attention: null
 pipeline_model_parallel_split_rank: 0