Add autoregressive config

config.json

@@ -103,7 +103,7 @@
     "length_penalty": 1.0,
     "max_length": 20,
     "min_length": 0,
-    "model_type": "clip_speech_model",
+    "model_type": "clvp_speech_model",
     "no_repeat_ngram_size": 0,
     "num_attention_heads": 12,
     "num_beam_groups": 1,
@@ -138,5 +138,28 @@
     "use_rotary_embedding": true,
     "summary_type": "mean"
   },
-  "speech_config_dict": null
+  "speech_config_dict": null,
+  "autoregressive_config": {
+    "activation_function": "gelu_new",
+    "attn_pdrop": 0.1,
+    "bos_token_id": 8192,
+    "embd_pdrop": 0.1,
+    "eos_token_id": 8193,
+    "initializer_range": 0.02,
+    "layer_norm_epsilon": 1e-05,
+    "model_type": "clvp_autoregressive_model",
+    "n_ctx": 1012,
+    "n_embd": 1024,
+    "n_head": 16,
+    "n_layer": 30,
+    "n_positions": 1012,
+    "resid_pdrop": 0.1,
+    "summary_activation": null,
+    "summary_first_dropout": 0.1,
+    "summary_proj_to_labels": true,
+    "summary_type": "cls_index",
+    "summary_use_proj": true,
+    "vocab_size": 8194
+    },
+  "autoregressive_config_dict": null
 }