llama 2

config/llama2_hf.yaml

@@ -0,0 +1,13 @@
+# Model configuration
+model:
+  type: llama
+  seq_len: 2048
+  hidden_dim: 4096
+  intermediate_dim: 11008
+  num_layers: 32
+  num_heads: 32
+  num_kv_heads: 32
+  use_flash_attention: true
+  flash_attention_block_size: 512
+  use_bias: false
+  use_layer_norm_weight: false