diff --git a/llama2_hf_convert.yaml b/llama2_hf_convert.yaml new file mode 100644 index 000000000..4394c327c --- /dev/null +++ b/llama2_hf_convert.yaml @@ -0,0 +1,10 @@ +model: # 7B class 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: 1024 \ No newline at end of file