Upload MERaLiON3ForConditionalGeneration

config.json

@@ -1,10 +1,14 @@
 {
+  "architectures": [
+    "MERaLiON3ForConditionalGeneration"
+  ],
   "auto_map": {
     "AutoConfig": "configuration_meralion3.MERaLiON3Config",
     "AutoModelForSpeechSeq2Seq": "modeling_meralion3.MERaLiON3ForConditionalGeneration",
     "AutoProcessor": "processing_meralion3.MERaLiON3Processor"
   },
   "cache_implementation": "hybrid",
+  "dtype": "bfloat16",
   "fixed_speech_embeds_length": 300,
   "head_dim": 256,
   "hidden_size": 3584,

generation_config.json

@@ -0,0 +1,8 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 2,
+  "cache_implementation": "hybrid",
+  "eos_token_id": 107,
+  "pad_token_id": 0,
+  "transformers_version": "4.56.2"
+}

model-00001-of-00005.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a7555b487dd50ca205ff51db1e6c2460989313aa55128bd98142018f29d663d
+size 4910473064

model-00002-of-00005.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9429286e3ffbbaf4d743119dd1989a921d9fcb3d5c648ba2c098a0a8cfefbfbc
+size 4962223136

model-00003-of-00005.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:03aa21069bcc425e1ee670244fc2174180857bb413ccfda74c7c8e6d309a7d2f
+size 4947572720

model-00004-of-00005.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bfeff15dab020e5749de4cc258ae6f00a7401cc2c129527ae02448735cbcbb2e
+size 4962223200

model-00005-of-00005.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b3427bf09571f4b197c553765131fd0f21776525fffdfdbdb20da71de4766de2
+size 102797064

model.safetensors.index.json

@@ -0,0 +1,966 @@
+{
+  "metadata": {
+    "total_parameters": 9942585120,
+    "total_size": 19885170240
+  },
+  "weight_map": {
+    "ln_speech.bias": "model-00001-of-00005.safetensors",
+    "ln_speech.weight": "model-00001-of-00005.safetensors",
+    "speech_audio_adapter.mlp_adapter.0.bias": "model-00001-of-00005.safetensors",
+    "speech_audio_adapter.mlp_adapter.0.weight": "model-00001-of-00005.safetensors",
+    "speech_audio_adapter.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_audio_adapter.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.conv1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.conv1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.conv2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.conv2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.embed_positions.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.0.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.0.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.0.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.0.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.0.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.0.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.0.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.0.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.0.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.0.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.0.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.0.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.0.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.0.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.0.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.1.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.1.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.1.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.1.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.1.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.1.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.1.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.1.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.1.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.1.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.1.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.1.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.1.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.1.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.1.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.10.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.10.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.10.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.10.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.10.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.10.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.10.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.10.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.10.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.10.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.10.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.10.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.10.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.10.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.10.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.11.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.11.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.11.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.11.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.11.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.11.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.11.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.11.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.11.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.11.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.11.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.11.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.11.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.11.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.11.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.12.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.12.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.12.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.12.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.12.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.12.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.12.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.12.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.12.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.12.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.12.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.12.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.12.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.12.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.12.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.13.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.13.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.13.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.13.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.13.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.13.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.13.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.13.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.13.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.13.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.13.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.13.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.13.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.13.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.13.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.14.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.14.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.14.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.14.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.14.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.14.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.14.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.14.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.14.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.14.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.14.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.14.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.14.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.14.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.14.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.15.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.15.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.15.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.15.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.15.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.15.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.15.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.15.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.15.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.15.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.15.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.15.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.15.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.15.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.15.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.16.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.16.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.16.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.16.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.16.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.16.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.16.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.16.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.16.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.16.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.16.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.16.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.16.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.16.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.16.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.17.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.17.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.17.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.17.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.17.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.17.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.17.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.17.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.17.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.17.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.17.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.17.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.17.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.17.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.17.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.18.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.18.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.18.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.18.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.18.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.18.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.18.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.18.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.18.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.18.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.18.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.18.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.18.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.18.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.18.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.19.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.19.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.19.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.19.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.19.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.19.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.19.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.19.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.19.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.19.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.19.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.19.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.19.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.19.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.19.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.2.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.2.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.2.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.2.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.2.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.2.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.2.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.2.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.2.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.2.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.2.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.2.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.2.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.2.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.2.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.20.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.20.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.20.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.20.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.20.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.20.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.20.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.20.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.20.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.20.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.20.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.20.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.20.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.20.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.20.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.21.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.21.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.21.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.21.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.21.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.21.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.21.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.21.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.21.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.21.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.21.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.21.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.21.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.21.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.21.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.22.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.22.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.22.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.22.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.22.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.22.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.22.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.22.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.22.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.22.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.22.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.22.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.22.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.22.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.22.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.23.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.23.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.23.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.23.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.23.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.23.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.23.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.23.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.23.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.23.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.23.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.23.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.23.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.23.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.23.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.24.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.24.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.24.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.24.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.24.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.24.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.24.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.24.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.24.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.24.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.24.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.24.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.24.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.24.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.24.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.25.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.25.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.25.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.25.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.25.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.25.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.25.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.25.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.25.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.25.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.25.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.25.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.25.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.25.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.25.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.26.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.26.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.26.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.26.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.26.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.26.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.26.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.26.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.26.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.26.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.26.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.26.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.26.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.26.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.26.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.27.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.27.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.27.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.27.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.27.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.27.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.27.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.27.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.27.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.27.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.27.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.27.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.27.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.27.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.27.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.28.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.28.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.28.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.28.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.28.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.28.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.28.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.28.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.28.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.28.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.28.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.28.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.28.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.28.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.28.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.29.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.29.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.29.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.29.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.29.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.29.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.29.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.29.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.29.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.29.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.29.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.29.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.29.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.29.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.29.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.3.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.3.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.3.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.3.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.3.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.3.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.3.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.3.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.3.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.3.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.3.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.3.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.3.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.3.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.3.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.30.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.30.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.30.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.30.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.30.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.30.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.30.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.30.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.30.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.30.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.30.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.30.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.30.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.30.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.30.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.31.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.31.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.31.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.31.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.31.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.31.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.31.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.31.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.31.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.31.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.31.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.31.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.31.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.31.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.31.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.4.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.4.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.4.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.4.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.4.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.4.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.4.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.4.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.4.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.4.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.4.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.4.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.4.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.4.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.4.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.5.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.5.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.5.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.5.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.5.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.5.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.5.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.5.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.5.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.5.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.5.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.5.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.5.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.5.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.5.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.6.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.6.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.6.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.6.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.6.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.6.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.6.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.6.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.6.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.6.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.6.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.6.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.6.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.6.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.6.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.7.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.7.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.7.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.7.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.7.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.7.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.7.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.7.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.7.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.7.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.7.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.7.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.7.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.7.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.7.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.8.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.8.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.8.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.8.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.8.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.8.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.8.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.8.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.8.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.8.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.8.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.8.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.8.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.8.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.8.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.9.fc1.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.9.fc1.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.9.fc2.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.9.fc2.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.9.final_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.9.final_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.9.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.9.self_attn.out_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.9.self_attn.out_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.9.self_attn.q_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.9.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.9.self_attn.v_proj.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.9.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.9.self_attn_layer_norm.bias": "model-00001-of-00005.safetensors",
+    "speech_encoder.layers.9.self_attn_layer_norm.weight": "model-00001-of-00005.safetensors",
+    "speech_encoder_layer_weighted_sum.weights": "model-00001-of-00005.safetensors",
+    "text_decoder.model.embed_tokens.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.0.input_layernorm.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.0.mlp.down_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.0.mlp.gate_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.0.mlp.up_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.0.post_attention_layernorm.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.0.post_feedforward_layernorm.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.0.pre_feedforward_layernorm.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.0.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.0.self_attn.o_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.0.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.0.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.1.input_layernorm.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.1.mlp.down_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.1.mlp.gate_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.1.mlp.up_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.1.post_attention_layernorm.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.1.post_feedforward_layernorm.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.1.pre_feedforward_layernorm.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.1.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.1.self_attn.o_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.1.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.1.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.10.input_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.10.mlp.down_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.10.mlp.gate_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.10.mlp.up_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.10.post_attention_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.10.post_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.10.pre_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.10.self_attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.10.self_attn.o_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.10.self_attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.10.self_attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.11.input_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.11.mlp.down_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.11.mlp.gate_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.11.mlp.up_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.11.post_attention_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.11.post_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.11.pre_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.11.self_attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.11.self_attn.o_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.11.self_attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.11.self_attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.12.input_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.12.mlp.down_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.12.mlp.gate_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.12.mlp.up_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.12.post_attention_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.12.post_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.12.pre_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.12.self_attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.12.self_attn.o_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.12.self_attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.12.self_attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.13.input_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.13.mlp.down_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.13.mlp.gate_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.13.mlp.up_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.13.post_attention_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.13.post_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.13.pre_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.13.self_attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.13.self_attn.o_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.13.self_attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.13.self_attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.14.input_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.14.mlp.down_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.14.mlp.gate_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.14.mlp.up_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.14.post_attention_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.14.post_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.14.pre_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.14.self_attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.14.self_attn.o_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.14.self_attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.14.self_attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.15.input_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.15.mlp.down_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.15.mlp.gate_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.15.mlp.up_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.15.post_attention_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.15.post_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.15.pre_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.15.self_attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.15.self_attn.o_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.15.self_attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.15.self_attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.16.input_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.16.mlp.down_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.16.mlp.gate_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.16.mlp.up_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.16.post_attention_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.16.post_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.16.pre_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.16.self_attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.16.self_attn.o_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.16.self_attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.16.self_attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.17.input_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.17.mlp.down_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.17.mlp.gate_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.17.mlp.up_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.17.post_attention_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.17.post_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.17.pre_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.17.self_attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.17.self_attn.o_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.17.self_attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.17.self_attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.18.input_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.18.mlp.down_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.18.mlp.gate_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.18.mlp.up_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.18.post_attention_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.18.post_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.18.pre_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.18.self_attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.18.self_attn.o_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.18.self_attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.18.self_attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.19.input_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.19.mlp.down_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.19.mlp.gate_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.19.mlp.up_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.19.post_attention_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.19.post_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.19.pre_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.19.self_attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.19.self_attn.o_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.19.self_attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.19.self_attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.2.input_layernorm.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.2.mlp.down_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.2.mlp.gate_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.2.mlp.up_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.2.post_attention_layernorm.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.2.post_feedforward_layernorm.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.2.pre_feedforward_layernorm.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.2.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.2.self_attn.o_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.2.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.2.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.20.input_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.20.mlp.down_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.20.mlp.gate_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.20.mlp.up_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.20.post_attention_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.20.post_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.20.pre_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.20.self_attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.20.self_attn.o_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.20.self_attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.20.self_attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.21.input_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.21.mlp.down_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.21.mlp.gate_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.21.mlp.up_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.21.post_attention_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.21.post_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.21.pre_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.21.self_attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.21.self_attn.o_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.21.self_attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.21.self_attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.22.input_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.22.mlp.down_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.22.mlp.gate_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.22.mlp.up_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.22.post_attention_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.22.post_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.22.pre_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.22.self_attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.22.self_attn.o_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.22.self_attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.22.self_attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.23.input_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.23.mlp.down_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.23.mlp.gate_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.23.mlp.up_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.23.post_attention_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.23.post_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.23.pre_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.23.self_attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.23.self_attn.o_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.23.self_attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.23.self_attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.24.input_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.24.mlp.down_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.24.mlp.gate_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.24.mlp.up_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.24.post_attention_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.24.post_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.24.pre_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.24.self_attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.24.self_attn.o_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.24.self_attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.24.self_attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.25.input_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.25.mlp.down_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.25.mlp.gate_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.25.mlp.up_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.25.post_attention_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.25.post_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.25.pre_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.25.self_attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.25.self_attn.o_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.25.self_attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.25.self_attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.26.input_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.26.mlp.down_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.26.mlp.gate_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.26.mlp.up_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.26.post_attention_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.26.post_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.26.pre_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.26.self_attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.26.self_attn.o_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.26.self_attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.26.self_attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.27.input_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.27.mlp.down_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.27.mlp.gate_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.27.mlp.up_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.27.post_attention_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.27.post_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.27.pre_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.27.self_attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.27.self_attn.o_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.27.self_attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.27.self_attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.28.input_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.28.mlp.down_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.28.mlp.gate_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.28.mlp.up_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.28.post_attention_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.28.post_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.28.pre_feedforward_layernorm.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.28.self_attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.28.self_attn.o_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.28.self_attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.28.self_attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.29.input_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.29.mlp.down_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.29.mlp.gate_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.29.mlp.up_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.29.post_attention_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.29.post_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.29.pre_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.29.self_attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.29.self_attn.o_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.29.self_attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.29.self_attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "text_decoder.model.layers.3.input_layernorm.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.3.mlp.down_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.3.mlp.gate_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.3.mlp.up_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.3.post_attention_layernorm.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.3.post_feedforward_layernorm.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.3.pre_feedforward_layernorm.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.3.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.3.self_attn.o_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.3.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.3.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.30.input_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.30.mlp.down_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.30.mlp.gate_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.30.mlp.up_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.30.post_attention_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.30.post_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.30.pre_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.30.self_attn.k_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.30.self_attn.o_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.30.self_attn.q_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.30.self_attn.v_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.31.input_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.31.mlp.down_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.31.mlp.gate_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.31.mlp.up_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.31.post_attention_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.31.post_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.31.pre_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.31.self_attn.k_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.31.self_attn.o_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.31.self_attn.q_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.31.self_attn.v_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.32.input_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.32.mlp.down_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.32.mlp.gate_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.32.mlp.up_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.32.post_attention_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.32.post_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.32.pre_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.32.self_attn.k_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.32.self_attn.o_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.32.self_attn.q_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.32.self_attn.v_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.33.input_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.33.mlp.down_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.33.mlp.gate_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.33.mlp.up_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.33.post_attention_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.33.post_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.33.pre_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.33.self_attn.k_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.33.self_attn.o_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.33.self_attn.q_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.33.self_attn.v_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.34.input_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.34.mlp.down_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.34.mlp.gate_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.34.mlp.up_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.34.post_attention_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.34.post_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.34.pre_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.34.self_attn.k_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.34.self_attn.o_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.34.self_attn.q_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.34.self_attn.v_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.35.input_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.35.mlp.down_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.35.mlp.gate_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.35.mlp.up_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.35.post_attention_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.35.post_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.35.pre_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.35.self_attn.k_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.35.self_attn.o_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.35.self_attn.q_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.35.self_attn.v_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.36.input_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.36.mlp.down_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.36.mlp.gate_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.36.mlp.up_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.36.post_attention_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.36.post_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.36.pre_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.36.self_attn.k_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.36.self_attn.o_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.36.self_attn.q_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.36.self_attn.v_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.37.input_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.37.mlp.down_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.37.mlp.gate_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.37.mlp.up_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.37.post_attention_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.37.post_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.37.pre_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.37.self_attn.k_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.37.self_attn.o_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.37.self_attn.q_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.37.self_attn.v_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.38.input_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.38.mlp.down_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.38.mlp.gate_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.38.mlp.up_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.38.post_attention_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.38.post_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.38.pre_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.38.self_attn.k_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.38.self_attn.o_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.38.self_attn.q_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.38.self_attn.v_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.39.input_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.39.mlp.down_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.39.mlp.gate_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.39.mlp.up_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.39.post_attention_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.39.post_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.39.pre_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.39.self_attn.k_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.39.self_attn.o_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.39.self_attn.q_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.39.self_attn.v_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.4.input_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.4.mlp.down_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.4.mlp.gate_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.4.mlp.up_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.4.post_attention_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.4.post_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.4.pre_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.4.self_attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.4.self_attn.o_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.4.self_attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.4.self_attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "text_decoder.model.layers.40.input_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.40.mlp.down_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.40.mlp.gate_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.40.mlp.up_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.40.post_attention_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.40.post_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.40.pre_feedforward_layernorm.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.40.self_attn.k_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.40.self_attn.o_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.40.self_attn.q_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.40.self_attn.v_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.41.input_layernorm.weight": "model-00005-of-00005.safetensors",
+    "text_decoder.model.layers.41.mlp.down_proj.weight": "model-00005-of-00005.safetensors",
+    "text_decoder.model.layers.41.mlp.gate_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.41.mlp.up_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.41.post_attention_layernorm.weight": "model-00005-of-00005.safetensors",
+    "text_decoder.model.layers.41.post_feedforward_layernorm.weight": "model-00005-of-00005.safetensors",
+    "text_decoder.model.layers.41.pre_feedforward_layernorm.weight": "model-00005-of-00005.safetensors",
+    "text_decoder.model.layers.41.self_attn.k_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.41.self_attn.o_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.41.self_attn.q_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.41.self_attn.v_proj.weight": "model-00004-of-00005.safetensors",
+    "text_decoder.model.layers.5.input_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.5.mlp.down_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.5.mlp.gate_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.5.mlp.up_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.5.post_attention_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.5.post_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.5.pre_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.5.self_attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.5.self_attn.o_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.5.self_attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.5.self_attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.6.input_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.6.mlp.down_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.6.mlp.gate_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.6.mlp.up_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.6.post_attention_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.6.post_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.6.pre_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.6.self_attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.6.self_attn.o_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.6.self_attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.6.self_attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.7.input_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.7.mlp.down_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.7.mlp.gate_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.7.mlp.up_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.7.post_attention_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.7.post_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.7.pre_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.7.self_attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.7.self_attn.o_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.7.self_attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.7.self_attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.8.input_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.8.mlp.down_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.8.mlp.gate_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.8.mlp.up_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.8.post_attention_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.8.post_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.8.pre_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.8.self_attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.8.self_attn.o_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.8.self_attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.8.self_attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.9.input_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.9.mlp.down_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.9.mlp.gate_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.9.mlp.up_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.9.post_attention_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.9.post_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.9.pre_feedforward_layernorm.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.9.self_attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.9.self_attn.o_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.9.self_attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.layers.9.self_attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "text_decoder.model.norm.weight": "model-00005-of-00005.safetensors"
+  }
+}

modeling_meralion3.py

@@ -0,0 +1,613 @@
+"""PyTorch MERaLiON3 model."""
+
+from dataclasses import dataclass
+from typing import List, Optional, Tuple, Union
+
+import torch
+import torch.utils.checkpoint
+from torch import nn
+
+from transformers import Gemma2ForCausalLM
+from transformers.models.whisper.modeling_whisper import WhisperEncoder
+from transformers.cache_utils import HybridCache
+from transformers.generation import GenerationMixin
+from transformers.modeling_outputs import ModelOutput
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import (
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+    logging,
+    replace_return_docstrings,
+)
+
+from .configuration_meralion3 import MERaLiON3Config
+
+
+logger = logging.get_logger(__name__)
+
+_CONFIG_FOR_DOC = "MERaLiON3Config"
+
+
+# Copied from transformers.models.llama.modeling_llama._prepare_4d_causal_attention_mask_with_cache_position
+def _prepare_4d_causal_attention_mask_with_cache_position(
+    attention_mask: torch.Tensor,
+    sequence_length: int,
+    target_length: int,
+    dtype: torch.dtype,
+    device: torch.device,
+    min_dtype: float,
+    cache_position: torch.Tensor,
+    batch_size: int,
+):
+    """
+    Creates a causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` from a 2D mask of shape
+    `(batch_size, key_value_length)`, or if the input `attention_mask` is already 4D, do nothing.
+
+    Args:
+        attention_mask (`torch.Tensor`):
+            A 2D attention mask of shape `(batch_size, key_value_length)` or a 4D attention mask of shape `(batch_size, 1, query_length, key_value_length)`.
+        sequence_length (`int`):
+            The sequence length being processed.
+        target_length (`int`):
+            The target length: when generating with static cache, the mask should be as long as the static cache, to account for the 0 padding, the part of the cache that is not filled yet.
+        dtype (`torch.dtype`):
+            The dtype to use for the 4D attention mask.
+        device (`torch.device`):
+            The device to plcae the 4D attention mask on.
+        min_dtype (`float`):
+            The minimum value representable with the dtype `dtype`.
+        cache_position (`torch.Tensor`):
+            Indices depicting the position of the input sequence tokens in the sequence.
+        batch_size (`torch.Tensor`):
+            Batch size.
+    """
+    if attention_mask is not None and attention_mask.dim() == 4:
+        # In this case we assume that the mask comes already in inverted form and requires no inversion or slicing.
+        causal_mask = attention_mask
+    else:
+        causal_mask = torch.full((sequence_length, target_length), fill_value=min_dtype, dtype=dtype, device=device)
+        if sequence_length != 1:
+            causal_mask = torch.triu(causal_mask, diagonal=1)
+        causal_mask *= torch.arange(target_length, device=device) > cache_position.reshape(-1, 1)
+        causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
+        if attention_mask is not None:
+            causal_mask = causal_mask.clone()  # copy to contiguous memory for in-place edit
+            mask_length = attention_mask.shape[-1]
+            padding_mask = causal_mask[:, :, :, :mask_length] + attention_mask[:, None, None, :]
+            padding_mask = padding_mask == 0
+            causal_mask[:, :, :, :mask_length] = causal_mask[:, :, :, :mask_length].masked_fill(
+                padding_mask, min_dtype
+            )
+    return causal_mask
+
+
+# copied from Qwen3AudioCausalLMOutputWithPast
+@dataclass
+class MERaLiON3OutputWithPast(ModelOutput):
+    """
+    Base class for MERaLiON3 causal language model (or autoregressive) outputs.
+
+    Args:
+        loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
+            Language modeling loss (for next-token prediction).
+        logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
+            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
+        past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
+            `(batch_size, num_heads, sequence_length, embed_size_per_head)`)
+
+            Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see
+            `past_key_values` input) to speed up sequential decoding.
+        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, +
+            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
+        attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+        attention_mask (`torch.FloatTensor`, *optional*):
+            Attentions mask, used to update attention mask and position_ids.
+    """
+
+    loss: Optional[torch.FloatTensor] = None
+    logits: torch.FloatTensor = None
+    past_key_values: Optional[List[torch.FloatTensor]] = None
+    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
+    attentions: Optional[Tuple[torch.FloatTensor]] = None
+    attention_mask: Optional[torch.FloatTensor] = None
+
+
+MERALION_START_DOCSTRING = r"""
+    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
+    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
+    etc.)
+
+    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
+    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
+    and behavior.
+
+    Parameters:
+        config ([`MERaLiON3Config`]):
+            Model configuration class with all the parameters of the model. Initializing with a config file does not
+            load the weights associated with the model, only the configuration. Check out the
+            [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+
+@add_start_docstrings(
+    "The bare MERaLiON3 Model outputting raw hidden-states without any specific head on top.",
+    MERALION_START_DOCSTRING,
+)
+class MERaLiON3PreTrainedModel(PreTrainedModel):
+    config_class = MERaLiON3Config
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["WhisperEncoderLayer", "WhisperDecoderLayer", "Gemma2DecoderLayer"]
+    _supports_flash_attn_2 = True
+    _supports_sdpa = True
+    _supports_cache_class = True
+    _supports_static_cache = True
+
+    def _init_weights(self, module):
+        # important: this ported version of Qwen2Audio isn't meant for training from scratch - only
+        # inference and fine-tuning - so the proper init weights code has been removed
+        std = self.config.init_std if hasattr(self.config, "init_std") else self.config.speech_config.init_std
+
+        if isinstance(module, (nn.Linear, nn.Conv1d)):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+
+    @property
+    def _supports_sdpa(self):
+        """
+        Retrieve language_model's attribute to check whether the model supports
+        SDPA or not.
+        """
+        # During initialization, text_decoder may not exist yet, so fall back to True
+        if hasattr(self, 'text_decoder'):
+            return self.text_decoder._supports_sdpa
+        return True
+
+class MERaLiON3SpeechAudioAdaper(nn.Module):
+    def __init__(
+        self,
+        config,
+        **kwargs
+    ):
+        super(MERaLiON3SpeechAudioAdaper, self).__init__()
+        speech_audio_encoder_output_dim = config.speech_config.d_model
+        llm_input_hidden_size = config.text_config.hidden_size
+        speech_mlp_scale_factor = config.speech_mlp_scale_factor
+
+        self.speech_mlp_scale_factor = speech_mlp_scale_factor
+        self.mlp_adapter = nn.Sequential(
+            nn.Linear(
+                in_features=speech_audio_encoder_output_dim * speech_mlp_scale_factor,
+                out_features=speech_audio_encoder_output_dim
+            ),
+            nn.SiLU(),
+            nn.Dropout(0.1),
+        )
+
+        self.speech_llm_proj = nn.Sequential(
+                nn.Linear(
+                    speech_audio_encoder_output_dim,
+                    speech_audio_encoder_output_dim * 4
+                ),
+                nn.SiLU(),
+                nn.Dropout(0.1),
+
+                nn.Linear(
+                    speech_audio_encoder_output_dim * 4,
+                    llm_input_hidden_size
+                ),
+            )
+
+    def forward(self, speech_embeds, **kwargs):
+        B, T, C = speech_embeds.shape
+        speech_embeds = self.mlp_adapter(
+            speech_embeds.reshape(
+                B,
+                T // self.speech_mlp_scale_factor,
+                C * self.speech_mlp_scale_factor,
+            )
+        )
+        return self.speech_llm_proj(speech_embeds)
+    
+
+class MERaLiON3SpeechAudioAdaperLarge(nn.Module):
+    def __init__(
+        self,
+        config,
+        **kwargs
+    ):
+        super(MERaLiON3SpeechAudioAdaperLarge, self).__init__()
+        speech_audio_encoder_output_dim = config.speech_config.d_model
+        llm_input_hidden_size = config.text_config.hidden_size
+        
+        self.speech_mlp_use_projection = config.speech_mlp_use_projection
+        self.speech_mlp_scale_factor = config.speech_mlp_scale_factor
+
+        self.mlp_adapter = nn.Sequential(
+            nn.Linear(
+                in_features=speech_audio_encoder_output_dim * self.speech_mlp_scale_factor,
+                out_features=speech_audio_encoder_output_dim * 5,
+            ),
+            nn.SiLU(),
+            nn.Dropout(0.01),
+        )
+
+        if self.speech_mlp_use_projection:
+            self.gate_proj = nn.Linear(
+                    in_features=speech_audio_encoder_output_dim * 5,
+                    out_features=speech_audio_encoder_output_dim * 5,
+                )
+            
+            self.pool_proj = nn.Linear(
+                    in_features=speech_audio_encoder_output_dim * 5,
+                    out_features=speech_audio_encoder_output_dim * 5,
+            )
+            self.act_fn = nn.SiLU()
+
+        self.out_proj = nn.Linear(
+                speech_audio_encoder_output_dim * 5,
+                llm_input_hidden_size,
+        )
+
+
+    def forward(self, speech_embeds, **kwargs):
+        B, T, C = speech_embeds.shape
+        speech_embeds = self.mlp_adapter(
+            speech_embeds.reshape(
+                B,
+                T // self.speech_mlp_scale_factor,
+                C * self.speech_mlp_scale_factor,
+            )
+        )
+        if self.speech_mlp_use_projection:
+            speech_embeds = self.act_fn(self.gate_proj(speech_embeds)) * self.pool_proj(speech_embeds)
+        speech_embeds = self.out_proj(speech_embeds)
+        return speech_embeds
+
+
+MERALION_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+            Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
+            it.
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        input_features (`torch.FloatTensor` of shape `(batch_size, feature_size, feature_sequence_length)`, *optional*):
+            Float values mel features extracted from the raw speech waveform. Raw speech waveform can be obtained by
+            loading a `.flac` or `.wav` audio file into an array of type `List[float]` or a `numpy.ndarray`, *e.g.* via
+            the soundfile library (`pip install soundfile`). To prepare the array into `input_features`, the
+            [`AutoFeatureExtractor`] should be used for extracting the mel features, padding and conversion into a
+            tensor of type `torch.FloatTensor`. See [`~WhisperFeatureExtractor.__call__`]
+        attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            [What are attention masks?](../glossary#attention-mask)
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            If `past_key_values` is used, optionally only the last `decoder_input_ids` have to be input (see
+            `past_key_values`).
+
+            If you want to change padding behavior, you should read [`modeling_opt._prepare_decoder_attention_mask`]
+            and modify to your needs. See diagram 1 in [the paper](https://arxiv.org/abs/1910.13461) for more
+            information on the default strategy.
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+        feature_attention_mask (`torch.Tensor` of shape `(batch_size, feature_sequence_length)`, *optional*):
+            Mask to avoid performing attention on padding feature indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+        position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
+            config.n_positions - 1]`. [What are position IDs?](../glossary#position-ids)
+        past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
+            `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and 2 additional tensors of shape
+            `(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)`.
+
+            Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
+            blocks) that can be used (see `past_key_values` input) to speed up sequential decoding.
+
+            If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those that
+            don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of all
+            `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+        inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
+            is useful if you want more control over how to convert `input_ids` indices into associated vectors than the
+            model's internal embedding lookup matrix.
+        use_cache (`bool`, *optional*):
+            If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see
+            `past_key_values`).
+        output_attentions (`bool`, *optional*):
+            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
+            tensors for more detail.
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+"""
+
+class MERaLiON3SpeechLayerWeightedSum(nn.Module):
+    """Weighted sum over speech encoder hidden states across all layers."""
+
+    def __init__(self, num_layers: int):
+        super().__init__()
+        self.weights = nn.Parameter(torch.zeros(num_layers))
+
+    def forward(self, hidden_states: tuple) -> torch.Tensor:
+        norm_weights = torch.softmax(self.weights, dim=-1)
+        stacked = torch.stack(list(hidden_states), dim=0)  # [num_layers, batch, seq, hidden]
+        origin_shape = stacked.shape[1:]
+        num_layers = stacked.shape[0]
+        stacked = stacked.view(num_layers, -1)
+        result = (norm_weights.unsqueeze(-1) * stacked).sum(dim=0)
+        return result.view(origin_shape)
+
+
+@add_start_docstrings(
+    """The MERALION model which consists of a audio backbone and a language model.""",
+    MERALION_START_DOCSTRING,
+)
+class MERaLiON3ForConditionalGeneration(MERaLiON3PreTrainedModel, GenerationMixin):
+    def __init__(self, config: MERaLiON3Config):
+        config.text_config._attn_implementation = config._attn_implementation
+        config.speech_config._attn_implementation = config._attn_implementation
+
+        super().__init__(config)
+
+        self.speech_encoder = WhisperEncoder(config.speech_config)
+        # self.speech_encoder = AutoModel.from_config(config.audio_config, attn_implementation=config._attn_implementation)
+
+        if config.speech_config.use_weighted_layer_sum:
+            self.speech_encoder_layer_weighted_sum = MERaLiON3SpeechLayerWeightedSum(
+                num_layers=config.speech_config.encoder_layers
+            )
+
+        self.ln_speech = nn.LayerNorm(config.speech_config.d_model)
+        self.speech_audio_adapter = MERaLiON3SpeechAudioAdaperLarge(config)
+        self.vocab_size = config.text_config.vocab_size
+        self.text_decoder = Gemma2ForCausalLM(config.text_config)
+        self.pad_token_id = self.config.pad_token_id if self.config.pad_token_id is not None else -1
+        self._padding_side = "left"  # set it to left by default, user can use setter to change padding_sides
+        self.post_init()
+
+    @property
+    def padding_side(self):
+        return self._padding_side
+
+    @padding_side.setter
+    def padding_side(self, padding_side: str):
+        if padding_side not in ["left", "right"]:
+            raise ValueError(f"{padding_side} is not `left` or `right`.")
+        self._padding_side = padding_side
+
+    # Copied from transformers.models.llava.modeling_llava.LlavaForConditionalGeneration.get_input_embeddings
+    def get_input_embeddings(self):
+        return self.text_decoder.get_input_embeddings()
+
+    # Copied from transformers.models.llava.modeling_llava.LlavaForConditionalGeneration.set_input_embeddings
+    def set_input_embeddings(self, value):
+        self.text_decoder.set_input_embeddings(value)
+
+    # Copied from transformers.models.llava.modeling_llava.LlavaForConditionalGeneration.get_output_embeddings
+    def get_output_embeddings(self):
+        return self.text_decoder.get_output_embeddings()
+
+    # Copied from transformers.models.llava.modeling_llava.LlavaForConditionalGeneration.set_output_embeddings
+    def set_output_embeddings(self, new_embeddings):
+        self.text_decoder.set_output_embeddings(new_embeddings)
+
+    # Copied from transformers.models.llava.modeling_llava.LlavaForConditionalGeneration.set_decoder
+    def set_decoder(self, decoder):
+        self.text_decoder.set_decoder(decoder)
+
+    # Copied from transformers.models.llava.modeling_llava.LlavaForConditionalGeneration.get_decoder
+    def get_decoder(self):
+        return self.text_decoder.get_decoder()
+
+    # Copied from transformers.models.llava.modeling_llava.LlavaForConditionalGeneration.tie_weights
+    def tie_weights(self):
+        return self.text_decoder.tie_weights()
+
+    # Copied from transformers.models.llava.modeling_llava.LlavaForConditionalGeneration.resize_token_embeddings
+    def resize_token_embeddings(self, new_num_tokens: Optional[int] = None, pad_to_multiple_of=None) -> nn.Embedding:
+        model_embeds = self.text_decoder.resize_token_embeddings(new_num_tokens, pad_to_multiple_of)
+        # update vocab size
+        self.config.text_config.vocab_size = model_embeds.num_embeddings
+        self.vocab_size = model_embeds.num_embeddings
+        return model_embeds
+
+    @add_start_docstrings_to_model_forward(MERALION_INPUTS_DOCSTRING)
+    @replace_return_docstrings(output_type=MERaLiON3OutputWithPast, config_class=_CONFIG_FOR_DOC)
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        input_features: torch.FloatTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        feature_attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, MERaLiON3OutputWithPast]:
+        r"""
+        Args:
+            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+                config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+                (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        Returns:
+        """
+
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        speech_encoder_device = self.speech_encoder.device
+
+        if input_features is not None:
+            input_features = input_features.to(speech_encoder_device)
+            feature_attention_mask = feature_attention_mask.to(speech_encoder_device)
+
+            if inputs_embeds is None:
+                if self.config.speech_config.use_weighted_layer_sum:
+                    speech_encoder_output = self.speech_encoder(
+                        input_features,
+                        attention_mask=feature_attention_mask,
+                        output_hidden_states=True,
+                    )
+                    # hidden_states[0] is the input embedding; [1:] are the per-layer outputs
+                    speech_contexts_embeds = self.speech_encoder_layer_weighted_sum(
+                        speech_encoder_output.hidden_states[1:]
+                    )
+                else:
+                    speech_contexts_embeds = self.speech_encoder(
+                        input_features, attention_mask=feature_attention_mask
+                    ).last_hidden_state
+                speech_contexts_embeds = self.ln_speech(speech_contexts_embeds)
+                speech_audio_contexts_embeds = self.speech_audio_adapter(speech_contexts_embeds)
+
+                inputs_embeds = self.text_decoder.base_model.embed_tokens(input_ids)
+
+                speech_mask = (input_ids == self.config.speech_token_index).unsqueeze(-1)
+                speech_mask = speech_mask.expand_as(inputs_embeds).to(inputs_embeds.device)
+
+                inputs_embeds = inputs_embeds.masked_scatter(speech_mask, speech_audio_contexts_embeds)
+
+                input_ids = None
+
+        outputs = self.text_decoder(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            labels=labels
+        )
+
+        return outputs
+
+    # from transformers.models.gemma2.modeling_gemma2.Gemma2ForCausalLM.prepare_inputs_for_generation
+    def prepare_inputs_for_generation(
+        self,
+        input_ids,
+        attention_mask=None,
+        input_features=None,
+        feature_attention_mask=None,
+        past_key_values=None,
+        inputs_embeds=None,
+        cache_position=None,
+        position_ids=None,
+        use_cache=None,
+        **kwargs,
+    ):
+        # If we have cache: let's slice `input_ids` through `cache_position`, to keep only the unprocessed tokens
+        # Exception 1: when passing input_embeds, input_ids may be missing entries
+        # Exception 2: some generation methods do special slicing of input_ids, so we don't need to do it here
+        is_first_step = cache_position[0].item() == 0
+        if past_key_values is not None:
+            if inputs_embeds is not None:  # Exception 1
+                input_ids = input_ids[:, -cache_position.shape[0] :]
+            elif input_ids.shape[1] != cache_position.shape[0]:  # Default case (the "else", a no op, is Exception 2)
+                input_ids = input_ids[:, cache_position]
+
+        if attention_mask is not None and position_ids is None:
+            # create position_ids on the fly for batch generation
+            position_ids = attention_mask.long().cumsum(-1) - 1
+            position_ids.masked_fill_(attention_mask == 0, 1)
+            if past_key_values:
+                position_ids = position_ids[:, -input_ids.shape[1] :]
+                # This `clone` call is needed to avoid recapturing cuda graphs with `torch.compile`'s
+                # `mode="reduce-overhead`, as otherwise the input `position_ids` would have various stride
+                # during the decoding. Here, simply using `.contiguous()` is not sufficient as in the
+                # batch size = 1 case, `position_ids` is already contiguous but with varying stride
+                # which retriggers a capture.
+                position_ids = position_ids.clone(memory_format=torch.contiguous_format)
+
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and is_first_step:
+            model_inputs = {"inputs_embeds": inputs_embeds, "input_ids": None}
+        else:
+            # The clone here is for the same reason as for `position_ids`.
+            model_inputs = {"input_ids": input_ids.clone(memory_format=torch.contiguous_format), "inputs_embeds": None}
+
+        if (
+            isinstance(past_key_values, HybridCache)
+            and attention_mask.ndim == 2
+            and not self.config._attn_implementation == "flash_attention_2"
+        ):
+            if model_inputs["inputs_embeds"] is not None:
+                batch_size, sequence_length, _ = model_inputs["inputs_embeds"].shape
+                device = model_inputs["inputs_embeds"].device
+            else:
+                batch_size, sequence_length = model_inputs["input_ids"].shape
+                device = model_inputs["input_ids"].device
+            dtype = self.text_decoder.lm_head.weight.dtype
+            min_dtype = torch.finfo(dtype).min
+            attention_mask = _prepare_4d_causal_attention_mask_with_cache_position(
+                attention_mask,
+                sequence_length=sequence_length,
+                target_length=past_key_values.get_max_cache_shape(),
+                dtype=dtype,
+                device=device,
+                min_dtype=min_dtype,
+                cache_position=cache_position,
+                batch_size=batch_size,
+            )
+
+        model_inputs.update(
+            {
+                "attention_mask": attention_mask,
+                "position_ids": position_ids,
+                "cache_position": cache_position,
+                "past_key_values": past_key_values,
+                "use_cache": use_cache
+            }
+        )
+
+        # Input ids will only be used from the second step. 
+        if is_first_step:
+            model_inputs["input_features"] = input_features
+            model_inputs["feature_attention_mask"] = feature_attention_mask
+
+        return model_inputs
+
+    def _reorder_cache(self, *args, **kwargs):
+        return self.text_decoder._reorder_cache(*args, **kwargs)