initial

config.json

@@ -0,0 +1,22 @@
+{
+  "architectures": [
+    "StepAudioForCausalLM"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_stepaudio.StepAudioConfig",
+    "AutoModelForCausalLM": "modeling_stepaudio.StepAudioForCausalLM"
+  },
+  "model_type": "step_audio",
+  "bos_token_id": 1,
+  "pad_token_id": 0,
+  "eos_token_id": 3,
+  "hidden_size": 12288,
+  "intermediate_size": 31232,
+  "num_attention_heads": 96,
+  "num_attention_groups": 8,
+  "num_hidden_layers": 88,
+  "max_seq_len": 32768,
+  "vocab_size": 74752,
+  "rms_norm_eps": 1e-05,
+  "torch_dtype": "bfloat16"
+}

configuration_stepaudio.py

@@ -0,0 +1,41 @@
+from typing import Optional, List, Any, Dict
+from transformers.configuration_utils import PretrainedConfig
+
+
+
+class StepAudioConfig(PretrainedConfig):
+    model_type = "step_audio"
+    keys_to_ignore_at_inference = ["past_key_values"]
+    
+    def __init__(
+        self,
+        hidden_size: int = 5120,
+        intermediate_size: int = 13312,
+        num_attention_heads: int = 40,
+        num_attention_groups: int = 8,
+        num_hidden_layers: int = 48,
+        max_seq_len: int = 4096,
+        vocab_size: int = 65536,
+        rms_norm_eps: float = 1e-5,
+        bos_token_id: int = 1,
+        eos_token_id: int = 3,
+        pad_token_id: int = 0,
+        **kwargs,
+    ) -> None:
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_attention_heads = num_attention_heads
+        self.num_attention_groups = num_attention_groups
+        self.num_hidden_layers = num_hidden_layers
+        self.max_seq_len = max_seq_len
+        self.vocab_size = vocab_size
+        self.rms_norm_eps = rms_norm_eps
+        super().__init__(
+            bos_token_id=bos_token_id,
+            pad_token_id=pad_token_id,
+            eos_token_id=eos_token_id,
+            **kwargs
+        )
+
+
+__all__ = ["StepAudioConfig"]

model-00001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc7a8cae6582530335c162c6f05663990f6ba89ae99fa244e8110f7c875e300a
+size 9940651088

model-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a438b4d547cabe3a791f06444e940553c26d302c2b8e993912d02dfa5d910e81
+size 9940676016

model-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e315707423917f8c8de0a71e0e46862a1a74727b54add03e13b0ff172a890e70
+size 9991007912

model-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa89528e3a3c27e58c17d1070f807f2185644613dc3de7423df30d9622e78c2e
+size 9638661344

model-00005.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c43050406b9d11419500ff5deee701c5bcdfc94a92153c1d1974d2f42cda8ea
+size 9940676048

model-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48bcab471d862820551f889bd8ded1c356bd5cf163fd68401a5f7d5a43213193
+size 9991007944

model-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:496d9af4bb1100f2364b011b65a5005b7a248cf8a5225f8a8659e6de1b244a82
+size 9638661352

model-00008.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:96eaadd8b8648c77ff7b479295b31c0b3332be231ae4e7b16351c475e69827ac
+size 9940676048

model-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d97a574557ada4cffb12e8937c9c8171fb4ec4bdc2b88fc32f072beb11ee657e
+size 9991007944

model-00010.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8db74c008df7e75834d18d73b3a34da484e23bbd764efdda7fed7a25aa570571
+size 9638661352

model-00011.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e3aae77fc4ab6e52c7211738dab91101498fe08f95ce065528b687d82166ee6
+size 9940676048

model-00012.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a7b3630d2b589ede5c1ed345fe4f139e247664479b51ef71ff2f31b844cf5a80
+size 9991007944

model-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:944e2d23c8af7b4112acb14329d65101c36d9352e06651d316d7c22e31c489e8
+size 9638661352

model-00014.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f75fff618fe8abf9093a4297fd7852d5d8963ed117bc7f51e642c9e57d426bca
+size 9940676048

model-00015.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb517627eebd7459e0af1dcba04fca5787025b7fdaf0a99c957313d027c9e313
+size 9991007944

model-00016.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f2e45598fb54be7cd246d64f140dafce1324afcb61f36c3cc03a7adbe8fa43ef
+size 9638661352

model-00017.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f407bbc161785a2053bc18761dd59bbc50ad34fc3d9b5c44e6f3f6bc09e91ae5
+size 9940676048

model-00018.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94db9bcc23d21faf96ebab8f9a981729d9552e07b1b8972b3bbe7fa35d64fc1c
+size 9991007944

model-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51d741d7e68dc3238591450da68b23f62c4742d9364ac34accf12f3e5e1ef528
+size 9638661352

model-00020.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c1e6e3573fe54198aa9850d2360446000ee69b02bfd22c50493dd575edbc4c3f
+size 9940676048

model-00021.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3c7853f2073cf5bf320dce950c82d7dfb58aef6c94c092384ae4833376801b03
+size 9991007944

model-00022.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:58cabbac9b7a6a26bbc532d77bfe2c61257abec36faebcfb5041d4c8ba4357e6
+size 9638661352

model-00023.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:63cb7e5a5894217e3c527988c02087084243402a78244f900c21c69735d07f51
+size 9940676048

model-00024.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cd32354ae817369e845a3ea1f49a7c06ace20a0715db9429ae08df698e79ea19
+size 9991007944

model-00025.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b716da4c011e37e0278285b441ca6731de6305f60028c1d937520d44e7ef263b
+size 9638661352

model-00026.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3204d741da5528c7ca68fba7bd72e089fbcf7f30d06b003da6e6fa9385198058
+size 9940676048

model-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0febcf541aee59e0f6a8d2af926a0e96d191ef158887d30d967e31442be6c133
+size 7449184288

model.safetensors.index.json

@@ -0,0 +1 @@
+{"metadata": {"total_size": 263893180416}, "weight_map": {"model.embed_tokens.weight": "model-00001.safetensors", "model.layers.0.self_attn.o_proj.weight": "model-00001.safetensors", "model.layers.0.input_layernorm.weight": "model-00001.safetensors", "model.layers.0.mlp.down_proj.weight": "model-00001.safetensors", "model.layers.0.post_attention_layernorm.weight": "model-00001.safetensors", "model.layers.0.self_attn.q_proj.weight": "model-00001.safetensors", "model.layers.0.self_attn.k_proj.weight": "model-00001.safetensors", "model.layers.0.self_attn.v_proj.weight": "model-00001.safetensors", "model.layers.0.mlp.gate_proj.weight": "model-00001.safetensors", "model.layers.0.mlp.up_proj.weight": "model-00001.safetensors", "model.layers.1.self_attn.o_proj.weight": "model-00001.safetensors", "model.layers.1.input_layernorm.weight": "model-00001.safetensors", "model.layers.1.mlp.down_proj.weight": "model-00001.safetensors", "model.layers.1.post_attention_layernorm.weight": "model-00001.safetensors", "model.layers.1.self_attn.q_proj.weight": "model-00001.safetensors", "model.layers.1.self_attn.k_proj.weight": "model-00001.safetensors", "model.layers.1.self_attn.v_proj.weight": "model-00001.safetensors", "model.layers.1.mlp.gate_proj.weight": "model-00001.safetensors", "model.layers.1.mlp.up_proj.weight": "model-00001.safetensors", "model.layers.2.self_attn.o_proj.weight": "model-00001.safetensors", "model.layers.2.input_layernorm.weight": "model-00001.safetensors", "model.layers.2.mlp.down_proj.weight": "model-00001.safetensors", "model.layers.2.post_attention_layernorm.weight": "model-00001.safetensors", "model.layers.2.self_attn.q_proj.weight": "model-00001.safetensors", "model.layers.2.self_attn.k_proj.weight": "model-00001.safetensors", "model.layers.2.self_attn.v_proj.weight": "model-00001.safetensors", "model.layers.2.mlp.gate_proj.weight": "model-00001.safetensors", "model.layers.2.mlp.up_proj.weight": "model-00002.safetensors", "model.layers.3.self_attn.o_proj.weight": "model-00002.safetensors", "model.layers.3.input_layernorm.weight": "model-00002.safetensors", "model.layers.3.mlp.down_proj.weight": "model-00002.safetensors", "model.layers.3.post_attention_layernorm.weight": "model-00002.safetensors", "model.layers.3.self_attn.q_proj.weight": "model-00002.safetensors", "model.layers.3.self_attn.k_proj.weight": "model-00002.safetensors", "model.layers.3.self_attn.v_proj.weight": "model-00002.safetensors", "model.layers.3.mlp.gate_proj.weight": "model-00002.safetensors", "model.layers.3.mlp.up_proj.weight": "model-00002.safetensors", "model.layers.4.self_attn.o_proj.weight": "model-00002.safetensors", "model.layers.4.input_layernorm.weight": "model-00002.safetensors", "model.layers.4.mlp.down_proj.weight": "model-00002.safetensors", "model.layers.4.post_attention_layernorm.weight": "model-00002.safetensors", "model.layers.4.self_attn.q_proj.weight": "model-00002.safetensors", "model.layers.4.self_attn.k_proj.weight": "model-00002.safetensors", "model.layers.4.self_attn.v_proj.weight": "model-00002.safetensors", "model.layers.4.mlp.gate_proj.weight": "model-00002.safetensors", "model.layers.4.mlp.up_proj.weight": "model-00002.safetensors", "model.layers.5.self_attn.o_proj.weight": "model-00002.safetensors", "model.layers.5.input_layernorm.weight": "model-00002.safetensors", "model.layers.5.mlp.down_proj.weight": "model-00002.safetensors", "model.layers.5.post_attention_layernorm.weight": "model-00002.safetensors", "model.layers.5.self_attn.q_proj.weight": "model-00002.safetensors", "model.layers.5.self_attn.k_proj.weight": "model-00002.safetensors", "model.layers.5.self_attn.v_proj.weight": "model-00002.safetensors", "model.layers.5.mlp.gate_proj.weight": "model-00002.safetensors", "model.layers.5.mlp.up_proj.weight": "model-00002.safetensors", "model.layers.6.self_attn.o_proj.weight": "model-00002.safetensors", "model.layers.6.input_layernorm.weight": "model-00002.safetensors", "model.layers.6.mlp.down_proj.weight": "model-00003.safetensors", "model.layers.6.post_attention_layernorm.weight": "model-00003.safetensors", "model.layers.6.self_attn.q_proj.weight": "model-00003.safetensors", "model.layers.6.self_attn.k_proj.weight": "model-00003.safetensors", "model.layers.6.self_attn.v_proj.weight": "model-00003.safetensors", "model.layers.6.mlp.gate_proj.weight": "model-00003.safetensors", "model.layers.6.mlp.up_proj.weight": "model-00003.safetensors", "model.layers.7.self_attn.o_proj.weight": "model-00003.safetensors", "model.layers.7.input_layernorm.weight": "model-00003.safetensors", "model.layers.7.mlp.down_proj.weight": "model-00003.safetensors", "model.layers.7.post_attention_layernorm.weight": "model-00003.safetensors", "model.layers.7.self_attn.q_proj.weight": "model-00003.safetensors", "model.layers.7.self_attn.k_proj.weight": "model-00003.safetensors", "model.layers.7.self_attn.v_proj.weight": "model-00003.safetensors", "model.layers.7.mlp.gate_proj.weight": "model-00003.safetensors", "model.layers.7.mlp.up_proj.weight": "model-00003.safetensors", "model.layers.8.self_attn.o_proj.weight": "model-00003.safetensors", "model.layers.8.input_layernorm.weight": "model-00003.safetensors", "model.layers.8.mlp.down_proj.weight": "model-00003.safetensors", "model.layers.8.post_attention_layernorm.weight": "model-00003.safetensors", "model.layers.8.self_attn.q_proj.weight": "model-00003.safetensors", "model.layers.8.self_attn.k_proj.weight": "model-00003.safetensors", "model.layers.8.self_attn.v_proj.weight": "model-00003.safetensors", "model.layers.8.mlp.gate_proj.weight": "model-00003.safetensors", "model.layers.8.mlp.up_proj.weight": "model-00003.safetensors", "model.layers.9.self_attn.o_proj.weight": "model-00003.safetensors", "model.layers.9.input_layernorm.weight": "model-00003.safetensors", "model.layers.9.mlp.down_proj.weight": "model-00003.safetensors", "model.layers.9.post_attention_layernorm.weight": "model-00003.safetensors", "model.layers.9.self_attn.q_proj.weight": "model-00003.safetensors", "model.layers.9.self_attn.k_proj.weight": "model-00003.safetensors", "model.layers.9.self_attn.v_proj.weight": "model-00003.safetensors", "model.layers.9.mlp.gate_proj.weight": "model-00004.safetensors", "model.layers.9.mlp.up_proj.weight": "model-00004.safetensors", "model.layers.10.self_attn.o_proj.weight": "model-00004.safetensors", "model.layers.10.input_layernorm.weight": "model-00004.safetensors", "model.layers.10.mlp.down_proj.weight": "model-00004.safetensors", "model.layers.10.post_attention_layernorm.weight": "model-00004.safetensors", "model.layers.10.self_attn.q_proj.weight": "model-00004.safetensors", "model.layers.10.self_attn.k_proj.weight": "model-00004.safetensors", "model.layers.10.self_attn.v_proj.weight": "model-00004.safetensors", "model.layers.10.mlp.gate_proj.weight": "model-00004.safetensors", "model.layers.10.mlp.up_proj.weight": "model-00004.safetensors", "model.layers.11.self_attn.o_proj.weight": "model-00004.safetensors", "model.layers.11.input_layernorm.weight": "model-00004.safetensors", "model.layers.11.mlp.down_proj.weight": "model-00004.safetensors", "model.layers.11.post_attention_layernorm.weight": "model-00004.safetensors", "model.layers.11.self_attn.q_proj.weight": "model-00004.safetensors", "model.layers.11.self_attn.k_proj.weight": "model-00004.safetensors", "model.layers.11.self_attn.v_proj.weight": "model-00004.safetensors", "model.layers.11.mlp.gate_proj.weight": "model-00004.safetensors", "model.layers.11.mlp.up_proj.weight": "model-00004.safetensors", "model.layers.12.self_attn.o_proj.weight": "model-00004.safetensors", "model.layers.12.input_layernorm.weight": "model-00004.safetensors", "model.layers.12.mlp.down_proj.weight": "model-00004.safetensors", "model.layers.12.post_attention_layernorm.weight": "model-00004.safetensors", "model.layers.12.self_attn.q_proj.weight": "model-00004.safetensors", "model.layers.12.self_attn.k_proj.weight": "model-00004.safetensors", "model.layers.12.self_attn.v_proj.weight": "model-00004.safetensors", "model.layers.12.mlp.gate_proj.weight": "model-00004.safetensors", "model.layers.12.mlp.up_proj.weight": "model-00005.safetensors", "model.layers.13.self_attn.o_proj.weight": "model-00005.safetensors", "model.layers.13.input_layernorm.weight": "model-00005.safetensors", "model.layers.13.mlp.down_proj.weight": "model-00005.safetensors", "model.layers.13.post_attention_layernorm.weight": "model-00005.safetensors", "model.layers.13.self_attn.q_proj.weight": "model-00005.safetensors", "model.layers.13.self_attn.k_proj.weight": "model-00005.safetensors", "model.layers.13.self_attn.v_proj.weight": "model-00005.safetensors", "model.layers.13.mlp.gate_proj.weight": "model-00005.safetensors", "model.layers.13.mlp.up_proj.weight": "model-00005.safetensors", "model.layers.14.self_attn.o_proj.weight": "model-00005.safetensors", "model.layers.14.input_layernorm.weight": "model-00005.safetensors", "model.layers.14.mlp.down_proj.weight": "model-00005.safetensors", "model.layers.14.post_attention_layernorm.weight": "model-00005.safetensors", "model.layers.14.self_attn.q_proj.weight": "model-00005.safetensors", "model.layers.14.self_attn.k_proj.weight": "model-00005.safetensors", "model.layers.14.self_attn.v_proj.weight": "model-00005.safetensors", "model.layers.14.mlp.gate_proj.weight": "model-00005.safetensors", "model.layers.14.mlp.up_proj.weight": "model-00005.safetensors", "model.layers.15.self_attn.o_proj.weight": "model-00005.safetensors", "model.layers.15.input_layernorm.weight": "model-00005.safetensors", "model.layers.15.mlp.down_proj.weight": "model-00005.safetensors", "model.layers.15.post_attention_layernorm.weight": "model-00005.safetensors", "model.layers.15.self_attn.q_proj.weight": "model-00005.safetensors", "model.layers.15.self_attn.k_proj.weight": "model-00005.safetensors", "model.layers.15.self_attn.v_proj.weight": "model-00005.safetensors", "model.layers.15.mlp.gate_proj.weight": "model-00005.safetensors", "model.layers.15.mlp.up_proj.weight": "model-00005.safetensors", "model.layers.16.self_attn.o_proj.weight": "model-00005.safetensors", "model.layers.16.input_layernorm.weight": "model-00005.safetensors", "model.layers.16.mlp.down_proj.weight": "model-00006.safetensors", "model.layers.16.post_attention_layernorm.weight": "model-00006.safetensors", "model.layers.16.self_attn.q_proj.weight": "model-00006.safetensors", "model.layers.16.self_attn.k_proj.weight": "model-00006.safetensors", "model.layers.16.self_attn.v_proj.weight": "model-00006.safetensors", "model.layers.16.mlp.gate_proj.weight": "model-00006.safetensors", "model.layers.16.mlp.up_proj.weight": "model-00006.safetensors", "model.layers.17.self_attn.o_proj.weight": "model-00006.safetensors", "model.layers.17.input_layernorm.weight": "model-00006.safetensors", "model.layers.17.mlp.down_proj.weight": "model-00006.safetensors", "model.layers.17.post_attention_layernorm.weight": "model-00006.safetensors", "model.layers.17.self_attn.q_proj.weight": "model-00006.safetensors", "model.layers.17.self_attn.k_proj.weight": "model-00006.safetensors", "model.layers.17.self_attn.v_proj.weight": "model-00006.safetensors", "model.layers.17.mlp.gate_proj.weight": "model-00006.safetensors", "model.layers.17.mlp.up_proj.weight": "model-00006.safetensors", "model.layers.18.self_attn.o_proj.weight": "model-00006.safetensors", "model.layers.18.input_layernorm.weight": "model-00006.safetensors", "model.layers.18.mlp.down_proj.weight": "model-00006.safetensors", "model.layers.18.post_attention_layernorm.weight": "model-00006.safetensors", "model.layers.18.self_attn.q_proj.weight": "model-00006.safetensors", "model.layers.18.self_attn.k_proj.weight": "model-00006.safetensors", "model.layers.18.self_attn.v_proj.weight": "model-00006.safetensors", "model.layers.18.mlp.gate_proj.weight": "model-00006.safetensors", "model.layers.18.mlp.up_proj.weight": "model-00006.safetensors", "model.layers.19.self_attn.o_proj.weight": "model-00006.safetensors", "model.layers.19.input_layernorm.weight": "model-00006.safetensors", "model.layers.19.mlp.down_proj.weight": "model-00006.safetensors", "model.layers.19.post_attention_layernorm.weight": "model-00006.safetensors", "model.layers.19.self_attn.q_proj.weight": "model-00006.safetensors", "model.layers.19.self_attn.k_proj.weight": "model-00006.safetensors", "model.layers.19.self_attn.v_proj.weight": "model-00006.safetensors", "model.layers.19.mlp.gate_proj.weight": "model-00007.safetensors", "model.layers.19.mlp.up_proj.weight": "model-00007.safetensors", "model.layers.20.self_attn.o_proj.weight": "model-00007.safetensors", "model.layers.20.input_layernorm.weight": "model-00007.safetensors", "model.layers.20.mlp.down_proj.weight": "model-00007.safetensors", "model.layers.20.post_attention_layernorm.weight": "model-00007.safetensors", "model.layers.20.self_attn.q_proj.weight": "model-00007.safetensors", "model.layers.20.self_attn.k_proj.weight": "model-00007.safetensors", "model.layers.20.self_attn.v_proj.weight": "model-00007.safetensors", "model.layers.20.mlp.gate_proj.weight": "model-00007.safetensors", "model.layers.20.mlp.up_proj.weight": "model-00007.safetensors", "model.layers.21.self_attn.o_proj.weight": "model-00007.safetensors", "model.layers.21.input_layernorm.weight": "model-00007.safetensors", "model.layers.21.mlp.down_proj.weight": "model-00007.safetensors", "model.layers.21.post_attention_layernorm.weight": "model-00007.safetensors", "model.layers.21.self_attn.q_proj.weight": "model-00007.safetensors", "model.layers.21.self_attn.k_proj.weight": "model-00007.safetensors", "model.layers.21.self_attn.v_proj.weight": "model-00007.safetensors", "model.layers.21.mlp.gate_proj.weight": "model-00007.safetensors", "model.layers.21.mlp.up_proj.weight": "model-00007.safetensors", "model.layers.22.self_attn.o_proj.weight": "model-00007.safetensors", "model.layers.22.input_layernorm.weight": "model-00007.safetensors", "model.layers.22.mlp.down_proj.weight": "model-00007.safetensors", "model.layers.22.post_attention_layernorm.weight": "model-00007.safetensors", "model.layers.22.self_attn.q_proj.weight": "model-00007.safetensors", "model.layers.22.self_attn.k_proj.weight": "model-00007.safetensors", "model.layers.22.self_attn.v_proj.weight": "model-00007.safetensors", "model.layers.22.mlp.gate_proj.weight": "model-00007.safetensors", "model.layers.22.mlp.up_proj.weight": "model-00008.safetensors", "model.layers.23.self_attn.o_proj.weight": "model-00008.safetensors", "model.layers.23.input_layernorm.weight": "model-00008.safetensors", "model.layers.23.mlp.down_proj.weight": "model-00008.safetensors", "model.layers.23.post_attention_layernorm.weight": "model-00008.safetensors", "model.layers.23.self_attn.q_proj.weight": "model-00008.safetensors", "model.layers.23.self_attn.k_proj.weight": "model-00008.safetensors", "model.layers.23.self_attn.v_proj.weight": "model-00008.safetensors", "model.layers.23.mlp.gate_proj.weight": "model-00008.safetensors", "model.layers.23.mlp.up_proj.weight": "model-00008.safetensors", "model.layers.24.self_attn.o_proj.weight": "model-00008.safetensors", "model.layers.24.input_layernorm.weight": "model-00008.safetensors", "model.layers.24.mlp.down_proj.weight": "model-00008.safetensors", "model.layers.24.post_attention_layernorm.weight": "model-00008.safetensors", "model.layers.24.self_attn.q_proj.weight": "model-00008.safetensors", "model.layers.24.self_attn.k_proj.weight": "model-00008.safetensors", "model.layers.24.self_attn.v_proj.weight": "model-00008.safetensors", "model.layers.24.mlp.gate_proj.weight": "model-00008.safetensors", "model.layers.24.mlp.up_proj.weight": "model-00008.safetensors", "model.layers.25.self_attn.o_proj.weight": "model-00008.safetensors", "model.layers.25.input_layernorm.weight": "model-00008.safetensors", "model.layers.25.mlp.down_proj.weight": "model-00008.safetensors", "model.layers.25.post_attention_layernorm.weight": "model-00008.safetensors", "model.layers.25.self_attn.q_proj.weight": "model-00008.safetensors", "model.layers.25.self_attn.k_proj.weight": "model-00008.safetensors", "model.layers.25.self_attn.v_proj.weight": "model-00008.safetensors", "model.layers.25.mlp.gate_proj.weight": "model-00008.safetensors", "model.layers.25.mlp.up_proj.weight": "model-00008.safetensors", "model.layers.26.self_attn.o_proj.weight": "model-00008.safetensors", "model.layers.26.input_layernorm.weight": "model-00008.safetensors", "model.layers.26.mlp.down_proj.weight": "model-00009.safetensors", "model.layers.26.post_attention_layernorm.weight": "model-00009.safetensors", "model.layers.26.self_attn.q_proj.weight": "model-00009.safetensors", "model.layers.26.self_attn.k_proj.weight": "model-00009.safetensors", "model.layers.26.self_attn.v_proj.weight": "model-00009.safetensors", "model.layers.26.mlp.gate_proj.weight": "model-00009.safetensors", "model.layers.26.mlp.up_proj.weight": "model-00009.safetensors", "model.layers.27.self_attn.o_proj.weight": "model-00009.safetensors", "model.layers.27.input_layernorm.weight": "model-00009.safetensors", "model.layers.27.mlp.down_proj.weight": "model-00009.safetensors", "model.layers.27.post_attention_layernorm.weight": "model-00009.safetensors", "model.layers.27.self_attn.q_proj.weight": "model-00009.safetensors", "model.layers.27.self_attn.k_proj.weight": "model-00009.safetensors", "model.layers.27.self_attn.v_proj.weight": "model-00009.safetensors", "model.layers.27.mlp.gate_proj.weight": "model-00009.safetensors", "model.layers.27.mlp.up_proj.weight": "model-00009.safetensors", "model.layers.28.self_attn.o_proj.weight": "model-00009.safetensors", "model.layers.28.input_layernorm.weight": "model-00009.safetensors", "model.layers.28.mlp.down_proj.weight": "model-00009.safetensors", "model.layers.28.post_attention_layernorm.weight": "model-00009.safetensors", "model.layers.28.self_attn.q_proj.weight": "model-00009.safetensors", "model.layers.28.self_attn.k_proj.weight": "model-00009.safetensors", "model.layers.28.self_attn.v_proj.weight": "model-00009.safetensors", "model.layers.28.mlp.gate_proj.weight": "model-00009.safetensors", "model.layers.28.mlp.up_proj.weight": "model-00009.safetensors", "model.layers.29.self_attn.o_proj.weight": "model-00009.safetensors", "model.layers.29.input_layernorm.weight": "model-00009.safetensors", "model.layers.29.mlp.down_proj.weight": "model-00009.safetensors", "model.layers.29.post_attention_layernorm.weight": "model-00009.safetensors", "model.layers.29.self_attn.q_proj.weight": "model-00009.safetensors", "model.layers.29.self_attn.k_proj.weight": "model-00009.safetensors", "model.layers.29.self_attn.v_proj.weight": "model-00009.safetensors", "model.layers.29.mlp.gate_proj.weight": "model-00010.safetensors", "model.layers.29.mlp.up_proj.weight": "model-00010.safetensors", "model.layers.30.self_attn.o_proj.weight": "model-00010.safetensors", "model.layers.30.input_layernorm.weight": "model-00010.safetensors", "model.layers.30.mlp.down_proj.weight": "model-00010.safetensors", "model.layers.30.post_attention_layernorm.weight": "model-00010.safetensors", "model.layers.30.self_attn.q_proj.weight": "model-00010.safetensors", "model.layers.30.self_attn.k_proj.weight": "model-00010.safetensors", "model.layers.30.self_attn.v_proj.weight": "model-00010.safetensors", "model.layers.30.mlp.gate_proj.weight": "model-00010.safetensors", "model.layers.30.mlp.up_proj.weight": "model-00010.safetensors", "model.layers.31.self_attn.o_proj.weight": "model-00010.safetensors", "model.layers.31.input_layernorm.weight": "model-00010.safetensors", "model.layers.31.mlp.down_proj.weight": "model-00010.safetensors", "model.layers.31.post_attention_layernorm.weight": "model-00010.safetensors", "model.layers.31.self_attn.q_proj.weight": "model-00010.safetensors", "model.layers.31.self_attn.k_proj.weight": "model-00010.safetensors", "model.layers.31.self_attn.v_proj.weight": "model-00010.safetensors", "model.layers.31.mlp.gate_proj.weight": "model-00010.safetensors", "model.layers.31.mlp.up_proj.weight": "model-00010.safetensors", "model.layers.32.self_attn.o_proj.weight": "model-00010.safetensors", "model.layers.32.input_layernorm.weight": "model-00010.safetensors", "model.layers.32.mlp.down_proj.weight": "model-00010.safetensors", "model.layers.32.post_attention_layernorm.weight": "model-00010.safetensors", "model.layers.32.self_attn.q_proj.weight": "model-00010.safetensors", "model.layers.32.self_attn.k_proj.weight": "model-00010.safetensors", "model.layers.32.self_attn.v_proj.weight": "model-00010.safetensors", "model.layers.32.mlp.gate_proj.weight": "model-00010.safetensors", "model.layers.32.mlp.up_proj.weight": "model-00011.safetensors", "model.layers.33.self_attn.o_proj.weight": "model-00011.safetensors", "model.layers.33.input_layernorm.weight": "model-00011.safetensors", "model.layers.33.mlp.down_proj.weight": "model-00011.safetensors", "model.layers.33.post_attention_layernorm.weight": "model-00011.safetensors", "model.layers.33.self_attn.q_proj.weight": "model-00011.safetensors", "model.layers.33.self_attn.k_proj.weight": "model-00011.safetensors", "model.layers.33.self_attn.v_proj.weight": "model-00011.safetensors", "model.layers.33.mlp.gate_proj.weight": "model-00011.safetensors", "model.layers.33.mlp.up_proj.weight": "model-00011.safetensors", "model.layers.34.self_attn.o_proj.weight": "model-00011.safetensors", "model.layers.34.input_layernorm.weight": "model-00011.safetensors", "model.layers.34.mlp.down_proj.weight": "model-00011.safetensors", "model.layers.34.post_attention_layernorm.weight": "model-00011.safetensors", "model.layers.34.self_attn.q_proj.weight": "model-00011.safetensors", "model.layers.34.self_attn.k_proj.weight": "model-00011.safetensors", "model.layers.34.self_attn.v_proj.weight": "model-00011.safetensors", "model.layers.34.mlp.gate_proj.weight": "model-00011.safetensors", "model.layers.34.mlp.up_proj.weight": "model-00011.safetensors", "model.layers.35.self_attn.o_proj.weight": "model-00011.safetensors", "model.layers.35.input_layernorm.weight": "model-00011.safetensors", "model.layers.35.mlp.down_proj.weight": "model-00011.safetensors", "model.layers.35.post_attention_layernorm.weight": "model-00011.safetensors", "model.layers.35.self_attn.q_proj.weight": "model-00011.safetensors", "model.layers.35.self_attn.k_proj.weight": "model-00011.safetensors", "model.layers.35.self_attn.v_proj.weight": "model-00011.safetensors", "model.layers.35.mlp.gate_proj.weight": "model-00011.safetensors", "model.layers.35.mlp.up_proj.weight": "model-00011.safetensors", "model.layers.36.self_attn.o_proj.weight": "model-00011.safetensors", "model.layers.36.input_layernorm.weight": "model-00011.safetensors", "model.layers.36.mlp.down_proj.weight": "model-00012.safetensors", "model.layers.36.post_attention_layernorm.weight": "model-00012.safetensors", "model.layers.36.self_attn.q_proj.weight": "model-00012.safetensors", "model.layers.36.self_attn.k_proj.weight": "model-00012.safetensors", "model.layers.36.self_attn.v_proj.weight": "model-00012.safetensors", "model.layers.36.mlp.gate_proj.weight": "model-00012.safetensors", "model.layers.36.mlp.up_proj.weight": "model-00012.safetensors", "model.layers.37.self_attn.o_proj.weight": "model-00012.safetensors", "model.layers.37.input_layernorm.weight": "model-00012.safetensors", "model.layers.37.mlp.down_proj.weight": "model-00012.safetensors", "model.layers.37.post_attention_layernorm.weight": "model-00012.safetensors", "model.layers.37.self_attn.q_proj.weight": "model-00012.safetensors", "model.layers.37.self_attn.k_proj.weight": "model-00012.safetensors", "model.layers.37.self_attn.v_proj.weight": "model-00012.safetensors", "model.layers.37.mlp.gate_proj.weight": "model-00012.safetensors", "model.layers.37.mlp.up_proj.weight": "model-00012.safetensors", "model.layers.38.self_attn.o_proj.weight": "model-00012.safetensors", "model.layers.38.input_layernorm.weight": "model-00012.safetensors", "model.layers.38.mlp.down_proj.weight": "model-00012.safetensors", "model.layers.38.post_attention_layernorm.weight": "model-00012.safetensors", "model.layers.38.self_attn.q_proj.weight": "model-00012.safetensors", "model.layers.38.self_attn.k_proj.weight": "model-00012.safetensors", "model.layers.38.self_attn.v_proj.weight": "model-00012.safetensors", "model.layers.38.mlp.gate_proj.weight": "model-00012.safetensors", "model.layers.38.mlp.up_proj.weight": "model-00012.safetensors", "model.layers.39.self_attn.o_proj.weight": "model-00012.safetensors", "model.layers.39.input_layernorm.weight": "model-00012.safetensors", "model.layers.39.mlp.down_proj.weight": "model-00012.safetensors", "model.layers.39.post_attention_layernorm.weight": "model-00012.safetensors", "model.layers.39.self_attn.q_proj.weight": "model-00012.safetensors", "model.layers.39.self_attn.k_proj.weight": "model-00012.safetensors", "model.layers.39.self_attn.v_proj.weight": "model-00012.safetensors", "model.layers.39.mlp.gate_proj.weight": "model-00013.safetensors", "model.layers.39.mlp.up_proj.weight": "model-00013.safetensors", "model.layers.40.self_attn.o_proj.weight": "model-00013.safetensors", "model.layers.40.input_layernorm.weight": "model-00013.safetensors", "model.layers.40.mlp.down_proj.weight": "model-00013.safetensors", "model.layers.40.post_attention_layernorm.weight": "model-00013.safetensors", "model.layers.40.self_attn.q_proj.weight": "model-00013.safetensors", "model.layers.40.self_attn.k_proj.weight": "model-00013.safetensors", "model.layers.40.self_attn.v_proj.weight": "model-00013.safetensors", "model.layers.40.mlp.gate_proj.weight": "model-00013.safetensors", "model.layers.40.mlp.up_proj.weight": "model-00013.safetensors", "model.layers.41.self_attn.o_proj.weight": "model-00013.safetensors", "model.layers.41.input_layernorm.weight": "model-00013.safetensors", "model.layers.41.mlp.down_proj.weight": "model-00013.safetensors", "model.layers.41.post_attention_layernorm.weight": "model-00013.safetensors", "model.layers.41.self_attn.q_proj.weight": "model-00013.safetensors", "model.layers.41.self_attn.k_proj.weight": "model-00013.safetensors", "model.layers.41.self_attn.v_proj.weight": "model-00013.safetensors", "model.layers.41.mlp.gate_proj.weight": "model-00013.safetensors", "model.layers.41.mlp.up_proj.weight": "model-00013.safetensors", "model.layers.42.self_attn.o_proj.weight": "model-00013.safetensors", "model.layers.42.input_layernorm.weight": "model-00013.safetensors", "model.layers.42.mlp.down_proj.weight": "model-00013.safetensors", "model.layers.42.post_attention_layernorm.weight": "model-00013.safetensors", "model.layers.42.self_attn.q_proj.weight": "model-00013.safetensors", "model.layers.42.self_attn.k_proj.weight": "model-00013.safetensors", "model.layers.42.self_attn.v_proj.weight": "model-00013.safetensors", "model.layers.42.mlp.gate_proj.weight": "model-00013.safetensors", "model.layers.42.mlp.up_proj.weight": "model-00014.safetensors", "model.layers.43.self_attn.o_proj.weight": "model-00014.safetensors", "model.layers.43.input_layernorm.weight": "model-00014.safetensors", "model.layers.43.mlp.down_proj.weight": "model-00014.safetensors", "model.layers.43.post_attention_layernorm.weight": "model-00014.safetensors", "model.layers.43.self_attn.q_proj.weight": "model-00014.safetensors", "model.layers.43.self_attn.k_proj.weight": "model-00014.safetensors", "model.layers.43.self_attn.v_proj.weight": "model-00014.safetensors", "model.layers.43.mlp.gate_proj.weight": "model-00014.safetensors", "model.layers.43.mlp.up_proj.weight": "model-00014.safetensors", "model.layers.44.self_attn.o_proj.weight": "model-00014.safetensors", "model.layers.44.input_layernorm.weight": "model-00014.safetensors", "model.layers.44.mlp.down_proj.weight": "model-00014.safetensors", "model.layers.44.post_attention_layernorm.weight": "model-00014.safetensors", "model.layers.44.self_attn.q_proj.weight": "model-00014.safetensors", "model.layers.44.self_attn.k_proj.weight": "model-00014.safetensors", "model.layers.44.self_attn.v_proj.weight": "model-00014.safetensors", "model.layers.44.mlp.gate_proj.weight": "model-00014.safetensors", "model.layers.44.mlp.up_proj.weight": "model-00014.safetensors", "model.layers.45.self_attn.o_proj.weight": "model-00014.safetensors", "model.layers.45.input_layernorm.weight": "model-00014.safetensors", "model.layers.45.mlp.down_proj.weight": "model-00014.safetensors", "model.layers.45.post_attention_layernorm.weight": "model-00014.safetensors", "model.layers.45.self_attn.q_proj.weight": "model-00014.safetensors", "model.layers.45.self_attn.k_proj.weight": "model-00014.safetensors", "model.layers.45.self_attn.v_proj.weight": "model-00014.safetensors", "model.layers.45.mlp.gate_proj.weight": "model-00014.safetensors", "model.layers.45.mlp.up_proj.weight": "model-00014.safetensors", "model.layers.46.self_attn.o_proj.weight": "model-00014.safetensors", "model.layers.46.input_layernorm.weight": "model-00014.safetensors", "model.layers.46.mlp.down_proj.weight": "model-00015.safetensors", "model.layers.46.post_attention_layernorm.weight": "model-00015.safetensors", "model.layers.46.self_attn.q_proj.weight": "model-00015.safetensors", "model.layers.46.self_attn.k_proj.weight": "model-00015.safetensors", "model.layers.46.self_attn.v_proj.weight": "model-00015.safetensors", "model.layers.46.mlp.gate_proj.weight": "model-00015.safetensors", "model.layers.46.mlp.up_proj.weight": "model-00015.safetensors", "model.layers.47.self_attn.o_proj.weight": "model-00015.safetensors", "model.layers.47.input_layernorm.weight": "model-00015.safetensors", "model.layers.47.mlp.down_proj.weight": "model-00015.safetensors", "model.layers.47.post_attention_layernorm.weight": "model-00015.safetensors", "model.layers.47.self_attn.q_proj.weight": "model-00015.safetensors", "model.layers.47.self_attn.k_proj.weight": "model-00015.safetensors", "model.layers.47.self_attn.v_proj.weight": "model-00015.safetensors", "model.layers.47.mlp.gate_proj.weight": "model-00015.safetensors", "model.layers.47.mlp.up_proj.weight": "model-00015.safetensors", "model.layers.48.self_attn.o_proj.weight": "model-00015.safetensors", "model.layers.48.input_layernorm.weight": "model-00015.safetensors", "model.layers.48.mlp.down_proj.weight": "model-00015.safetensors", "model.layers.48.post_attention_layernorm.weight": "model-00015.safetensors", "model.layers.48.self_attn.q_proj.weight": "model-00015.safetensors", "model.layers.48.self_attn.k_proj.weight": "model-00015.safetensors", "model.layers.48.self_attn.v_proj.weight": "model-00015.safetensors", "model.layers.48.mlp.gate_proj.weight": "model-00015.safetensors", "model.layers.48.mlp.up_proj.weight": "model-00015.safetensors", "model.layers.49.self_attn.o_proj.weight": "model-00015.safetensors", "model.layers.49.input_layernorm.weight": "model-00015.safetensors", "model.layers.49.mlp.down_proj.weight": "model-00015.safetensors", "model.layers.49.post_attention_layernorm.weight": "model-00015.safetensors", "model.layers.49.self_attn.q_proj.weight": "model-00015.safetensors", "model.layers.49.self_attn.k_proj.weight": "model-00015.safetensors", "model.layers.49.self_attn.v_proj.weight": "model-00015.safetensors", "model.layers.49.mlp.gate_proj.weight": "model-00016.safetensors", "model.layers.49.mlp.up_proj.weight": "model-00016.safetensors", "model.layers.50.self_attn.o_proj.weight": "model-00016.safetensors", "model.layers.50.input_layernorm.weight": "model-00016.safetensors", "model.layers.50.mlp.down_proj.weight": "model-00016.safetensors", "model.layers.50.post_attention_layernorm.weight": "model-00016.safetensors", "model.layers.50.self_attn.q_proj.weight": "model-00016.safetensors", "model.layers.50.self_attn.k_proj.weight": "model-00016.safetensors", "model.layers.50.self_attn.v_proj.weight": "model-00016.safetensors", "model.layers.50.mlp.gate_proj.weight": "model-00016.safetensors", "model.layers.50.mlp.up_proj.weight": "model-00016.safetensors", "model.layers.51.self_attn.o_proj.weight": "model-00016.safetensors", "model.layers.51.input_layernorm.weight": "model-00016.safetensors", "model.layers.51.mlp.down_proj.weight": "model-00016.safetensors", "model.layers.51.post_attention_layernorm.weight": "model-00016.safetensors", "model.layers.51.self_attn.q_proj.weight": "model-00016.safetensors", "model.layers.51.self_attn.k_proj.weight": "model-00016.safetensors", "model.layers.51.self_attn.v_proj.weight": "model-00016.safetensors", "model.layers.51.mlp.gate_proj.weight": "model-00016.safetensors", "model.layers.51.mlp.up_proj.weight": "model-00016.safetensors", "model.layers.52.self_attn.o_proj.weight": "model-00016.safetensors", "model.layers.52.input_layernorm.weight": "model-00016.safetensors", "model.layers.52.mlp.down_proj.weight": "model-00016.safetensors", "model.layers.52.post_attention_layernorm.weight": "model-00016.safetensors", "model.layers.52.self_attn.q_proj.weight": "model-00016.safetensors", "model.layers.52.self_attn.k_proj.weight": "model-00016.safetensors", "model.layers.52.self_attn.v_proj.weight": "model-00016.safetensors", "model.layers.52.mlp.gate_proj.weight": "model-00016.safetensors", "model.layers.52.mlp.up_proj.weight": "model-00017.safetensors", "model.layers.53.self_attn.o_proj.weight": "model-00017.safetensors", "model.layers.53.input_layernorm.weight": "model-00017.safetensors", "model.layers.53.mlp.down_proj.weight": "model-00017.safetensors", "model.layers.53.post_attention_layernorm.weight": "model-00017.safetensors", "model.layers.53.self_attn.q_proj.weight": "model-00017.safetensors", "model.layers.53.self_attn.k_proj.weight": "model-00017.safetensors", "model.layers.53.self_attn.v_proj.weight": "model-00017.safetensors", "model.layers.53.mlp.gate_proj.weight": "model-00017.safetensors", "model.layers.53.mlp.up_proj.weight": "model-00017.safetensors", "model.layers.54.self_attn.o_proj.weight": "model-00017.safetensors", "model.layers.54.input_layernorm.weight": "model-00017.safetensors", "model.layers.54.mlp.down_proj.weight": "model-00017.safetensors", "model.layers.54.post_attention_layernorm.weight": "model-00017.safetensors", "model.layers.54.self_attn.q_proj.weight": "model-00017.safetensors", "model.layers.54.self_attn.k_proj.weight": "model-00017.safetensors", "model.layers.54.self_attn.v_proj.weight": "model-00017.safetensors", "model.layers.54.mlp.gate_proj.weight": "model-00017.safetensors", "model.layers.54.mlp.up_proj.weight": "model-00017.safetensors", "model.layers.55.self_attn.o_proj.weight": "model-00017.safetensors", "model.layers.55.input_layernorm.weight": "model-00017.safetensors", "model.layers.55.mlp.down_proj.weight": "model-00017.safetensors", "model.layers.55.post_attention_layernorm.weight": "model-00017.safetensors", "model.layers.55.self_attn.q_proj.weight": "model-00017.safetensors", "model.layers.55.self_attn.k_proj.weight": "model-00017.safetensors", "model.layers.55.self_attn.v_proj.weight": "model-00017.safetensors", "model.layers.55.mlp.gate_proj.weight": "model-00017.safetensors", "model.layers.55.mlp.up_proj.weight": "model-00017.safetensors", "model.layers.56.self_attn.o_proj.weight": "model-00017.safetensors", "model.layers.56.input_layernorm.weight": "model-00017.safetensors", "model.layers.56.mlp.down_proj.weight": "model-00018.safetensors", "model.layers.56.post_attention_layernorm.weight": "model-00018.safetensors", "model.layers.56.self_attn.q_proj.weight": "model-00018.safetensors", "model.layers.56.self_attn.k_proj.weight": "model-00018.safetensors", "model.layers.56.self_attn.v_proj.weight": "model-00018.safetensors", "model.layers.56.mlp.gate_proj.weight": "model-00018.safetensors", "model.layers.56.mlp.up_proj.weight": "model-00018.safetensors", "model.layers.57.self_attn.o_proj.weight": "model-00018.safetensors", "model.layers.57.input_layernorm.weight": "model-00018.safetensors", "model.layers.57.mlp.down_proj.weight": "model-00018.safetensors", "model.layers.57.post_attention_layernorm.weight": "model-00018.safetensors", "model.layers.57.self_attn.q_proj.weight": "model-00018.safetensors", "model.layers.57.self_attn.k_proj.weight": "model-00018.safetensors", "model.layers.57.self_attn.v_proj.weight": "model-00018.safetensors", "model.layers.57.mlp.gate_proj.weight": "model-00018.safetensors", "model.layers.57.mlp.up_proj.weight": "model-00018.safetensors", "model.layers.58.self_attn.o_proj.weight": "model-00018.safetensors", "model.layers.58.input_layernorm.weight": "model-00018.safetensors", "model.layers.58.mlp.down_proj.weight": "model-00018.safetensors", "model.layers.58.post_attention_layernorm.weight": "model-00018.safetensors", "model.layers.58.self_attn.q_proj.weight": "model-00018.safetensors", "model.layers.58.self_attn.k_proj.weight": "model-00018.safetensors", "model.layers.58.self_attn.v_proj.weight": "model-00018.safetensors", "model.layers.58.mlp.gate_proj.weight": "model-00018.safetensors", "model.layers.58.mlp.up_proj.weight": "model-00018.safetensors", "model.layers.59.self_attn.o_proj.weight": "model-00018.safetensors", "model.layers.59.input_layernorm.weight": "model-00018.safetensors", "model.layers.59.mlp.down_proj.weight": "model-00018.safetensors", "model.layers.59.post_attention_layernorm.weight": "model-00018.safetensors", "model.layers.59.self_attn.q_proj.weight": "model-00018.safetensors", "model.layers.59.self_attn.k_proj.weight": "model-00018.safetensors", "model.layers.59.self_attn.v_proj.weight": "model-00018.safetensors", "model.layers.59.mlp.gate_proj.weight": "model-00019.safetensors", "model.layers.59.mlp.up_proj.weight": "model-00019.safetensors", "model.layers.60.self_attn.o_proj.weight": "model-00019.safetensors", "model.layers.60.input_layernorm.weight": "model-00019.safetensors", "model.layers.60.mlp.down_proj.weight": "model-00019.safetensors", "model.layers.60.post_attention_layernorm.weight": "model-00019.safetensors", "model.layers.60.self_attn.q_proj.weight": "model-00019.safetensors", "model.layers.60.self_attn.k_proj.weight": "model-00019.safetensors", "model.layers.60.self_attn.v_proj.weight": "model-00019.safetensors", "model.layers.60.mlp.gate_proj.weight": "model-00019.safetensors", "model.layers.60.mlp.up_proj.weight": "model-00019.safetensors", "model.layers.61.self_attn.o_proj.weight": "model-00019.safetensors", "model.layers.61.input_layernorm.weight": "model-00019.safetensors", "model.layers.61.mlp.down_proj.weight": "model-00019.safetensors", "model.layers.61.post_attention_layernorm.weight": "model-00019.safetensors", "model.layers.61.self_attn.q_proj.weight": "model-00019.safetensors", "model.layers.61.self_attn.k_proj.weight": "model-00019.safetensors", "model.layers.61.self_attn.v_proj.weight": "model-00019.safetensors", "model.layers.61.mlp.gate_proj.weight": "model-00019.safetensors", "model.layers.61.mlp.up_proj.weight": "model-00019.safetensors", "model.layers.62.self_attn.o_proj.weight": "model-00019.safetensors", "model.layers.62.input_layernorm.weight": "model-00019.safetensors", "model.layers.62.mlp.down_proj.weight": "model-00019.safetensors", "model.layers.62.post_attention_layernorm.weight": "model-00019.safetensors", "model.layers.62.self_attn.q_proj.weight": "model-00019.safetensors", "model.layers.62.self_attn.k_proj.weight": "model-00019.safetensors", "model.layers.62.self_attn.v_proj.weight": "model-00019.safetensors", "model.layers.62.mlp.gate_proj.weight": "model-00019.safetensors", "model.layers.62.mlp.up_proj.weight": "model-00020.safetensors", "model.layers.63.self_attn.o_proj.weight": "model-00020.safetensors", "model.layers.63.input_layernorm.weight": "model-00020.safetensors", "model.layers.63.mlp.down_proj.weight": "model-00020.safetensors", "model.layers.63.post_attention_layernorm.weight": "model-00020.safetensors", "model.layers.63.self_attn.q_proj.weight": "model-00020.safetensors", "model.layers.63.self_attn.k_proj.weight": "model-00020.safetensors", "model.layers.63.self_attn.v_proj.weight": "model-00020.safetensors", "model.layers.63.mlp.gate_proj.weight": "model-00020.safetensors", "model.layers.63.mlp.up_proj.weight": "model-00020.safetensors", "model.layers.64.self_attn.o_proj.weight": "model-00020.safetensors", "model.layers.64.input_layernorm.weight": "model-00020.safetensors", "model.layers.64.mlp.down_proj.weight": "model-00020.safetensors", "model.layers.64.post_attention_layernorm.weight": "model-00020.safetensors", "model.layers.64.self_attn.q_proj.weight": "model-00020.safetensors", "model.layers.64.self_attn.k_proj.weight": "model-00020.safetensors", "model.layers.64.self_attn.v_proj.weight": "model-00020.safetensors", "model.layers.64.mlp.gate_proj.weight": "model-00020.safetensors", "model.layers.64.mlp.up_proj.weight": "model-00020.safetensors", "model.layers.65.self_attn.o_proj.weight": "model-00020.safetensors", "model.layers.65.input_layernorm.weight": "model-00020.safetensors", "model.layers.65.mlp.down_proj.weight": "model-00020.safetensors", "model.layers.65.post_attention_layernorm.weight": "model-00020.safetensors", "model.layers.65.self_attn.q_proj.weight": "model-00020.safetensors", "model.layers.65.self_attn.k_proj.weight": "model-00020.safetensors", "model.layers.65.self_attn.v_proj.weight": "model-00020.safetensors", "model.layers.65.mlp.gate_proj.weight": "model-00020.safetensors", "model.layers.65.mlp.up_proj.weight": "model-00020.safetensors", "model.layers.66.self_attn.o_proj.weight": "model-00020.safetensors", "model.layers.66.input_layernorm.weight": "model-00020.safetensors", "model.layers.66.mlp.down_proj.weight": "model-00021.safetensors", "model.layers.66.post_attention_layernorm.weight": "model-00021.safetensors", "model.layers.66.self_attn.q_proj.weight": "model-00021.safetensors", "model.layers.66.self_attn.k_proj.weight": "model-00021.safetensors", "model.layers.66.self_attn.v_proj.weight": "model-00021.safetensors", "model.layers.66.mlp.gate_proj.weight": "model-00021.safetensors", "model.layers.66.mlp.up_proj.weight": "model-00021.safetensors", "model.layers.67.self_attn.o_proj.weight": "model-00021.safetensors", "model.layers.67.input_layernorm.weight": "model-00021.safetensors", "model.layers.67.mlp.down_proj.weight": "model-00021.safetensors", "model.layers.67.post_attention_layernorm.weight": "model-00021.safetensors", "model.layers.67.self_attn.q_proj.weight": "model-00021.safetensors", "model.layers.67.self_attn.k_proj.weight": "model-00021.safetensors", "model.layers.67.self_attn.v_proj.weight": "model-00021.safetensors", "model.layers.67.mlp.gate_proj.weight": "model-00021.safetensors", "model.layers.67.mlp.up_proj.weight": "model-00021.safetensors", "model.layers.68.self_attn.o_proj.weight": "model-00021.safetensors", "model.layers.68.input_layernorm.weight": "model-00021.safetensors", "model.layers.68.mlp.down_proj.weight": "model-00021.safetensors", "model.layers.68.post_attention_layernorm.weight": "model-00021.safetensors", "model.layers.68.self_attn.q_proj.weight": "model-00021.safetensors", "model.layers.68.self_attn.k_proj.weight": "model-00021.safetensors", "model.layers.68.self_attn.v_proj.weight": "model-00021.safetensors", "model.layers.68.mlp.gate_proj.weight": "model-00021.safetensors", "model.layers.68.mlp.up_proj.weight": "model-00021.safetensors", "model.layers.69.self_attn.o_proj.weight": "model-00021.safetensors", "model.layers.69.input_layernorm.weight": "model-00021.safetensors", "model.layers.69.mlp.down_proj.weight": "model-00021.safetensors", "model.layers.69.post_attention_layernorm.weight": "model-00021.safetensors", "model.layers.69.self_attn.q_proj.weight": "model-00021.safetensors", "model.layers.69.self_attn.k_proj.weight": "model-00021.safetensors", "model.layers.69.self_attn.v_proj.weight": "model-00021.safetensors", "model.layers.69.mlp.gate_proj.weight": "model-00022.safetensors", "model.layers.69.mlp.up_proj.weight": "model-00022.safetensors", "model.layers.70.self_attn.o_proj.weight": "model-00022.safetensors", "model.layers.70.input_layernorm.weight": "model-00022.safetensors", "model.layers.70.mlp.down_proj.weight": "model-00022.safetensors", "model.layers.70.post_attention_layernorm.weight": "model-00022.safetensors", "model.layers.70.self_attn.q_proj.weight": "model-00022.safetensors", "model.layers.70.self_attn.k_proj.weight": "model-00022.safetensors", "model.layers.70.self_attn.v_proj.weight": "model-00022.safetensors", "model.layers.70.mlp.gate_proj.weight": "model-00022.safetensors", "model.layers.70.mlp.up_proj.weight": "model-00022.safetensors", "model.layers.71.self_attn.o_proj.weight": "model-00022.safetensors", "model.layers.71.input_layernorm.weight": "model-00022.safetensors", "model.layers.71.mlp.down_proj.weight": "model-00022.safetensors", "model.layers.71.post_attention_layernorm.weight": "model-00022.safetensors", "model.layers.71.self_attn.q_proj.weight": "model-00022.safetensors", "model.layers.71.self_attn.k_proj.weight": "model-00022.safetensors", "model.layers.71.self_attn.v_proj.weight": "model-00022.safetensors", "model.layers.71.mlp.gate_proj.weight": "model-00022.safetensors", "model.layers.71.mlp.up_proj.weight": "model-00022.safetensors", "model.layers.72.self_attn.o_proj.weight": "model-00022.safetensors", "model.layers.72.input_layernorm.weight": "model-00022.safetensors", "model.layers.72.mlp.down_proj.weight": "model-00022.safetensors", "model.layers.72.post_attention_layernorm.weight": "model-00022.safetensors", "model.layers.72.self_attn.q_proj.weight": "model-00022.safetensors", "model.layers.72.self_attn.k_proj.weight": "model-00022.safetensors", "model.layers.72.self_attn.v_proj.weight": "model-00022.safetensors", "model.layers.72.mlp.gate_proj.weight": "model-00022.safetensors", "model.layers.72.mlp.up_proj.weight": "model-00023.safetensors", "model.layers.73.self_attn.o_proj.weight": "model-00023.safetensors", "model.layers.73.input_layernorm.weight": "model-00023.safetensors", "model.layers.73.mlp.down_proj.weight": "model-00023.safetensors", "model.layers.73.post_attention_layernorm.weight": "model-00023.safetensors", "model.layers.73.self_attn.q_proj.weight": "model-00023.safetensors", "model.layers.73.self_attn.k_proj.weight": "model-00023.safetensors", "model.layers.73.self_attn.v_proj.weight": "model-00023.safetensors", "model.layers.73.mlp.gate_proj.weight": "model-00023.safetensors", "model.layers.73.mlp.up_proj.weight": "model-00023.safetensors", "model.layers.74.self_attn.o_proj.weight": "model-00023.safetensors", "model.layers.74.input_layernorm.weight": "model-00023.safetensors", "model.layers.74.mlp.down_proj.weight": "model-00023.safetensors", "model.layers.74.post_attention_layernorm.weight": "model-00023.safetensors", "model.layers.74.self_attn.q_proj.weight": "model-00023.safetensors", "model.layers.74.self_attn.k_proj.weight": "model-00023.safetensors", "model.layers.74.self_attn.v_proj.weight": "model-00023.safetensors", "model.layers.74.mlp.gate_proj.weight": "model-00023.safetensors", "model.layers.74.mlp.up_proj.weight": "model-00023.safetensors", "model.layers.75.self_attn.o_proj.weight": "model-00023.safetensors", "model.layers.75.input_layernorm.weight": "model-00023.safetensors", "model.layers.75.mlp.down_proj.weight": "model-00023.safetensors", "model.layers.75.post_attention_layernorm.weight": "model-00023.safetensors", "model.layers.75.self_attn.q_proj.weight": "model-00023.safetensors", "model.layers.75.self_attn.k_proj.weight": "model-00023.safetensors", "model.layers.75.self_attn.v_proj.weight": "model-00023.safetensors", "model.layers.75.mlp.gate_proj.weight": "model-00023.safetensors", "model.layers.75.mlp.up_proj.weight": "model-00023.safetensors", "model.layers.76.self_attn.o_proj.weight": "model-00023.safetensors", "model.layers.76.input_layernorm.weight": "model-00023.safetensors", "model.layers.76.mlp.down_proj.weight": "model-00024.safetensors", "model.layers.76.post_attention_layernorm.weight": "model-00024.safetensors", "model.layers.76.self_attn.q_proj.weight": "model-00024.safetensors", "model.layers.76.self_attn.k_proj.weight": "model-00024.safetensors", "model.layers.76.self_attn.v_proj.weight": "model-00024.safetensors", "model.layers.76.mlp.gate_proj.weight": "model-00024.safetensors", "model.layers.76.mlp.up_proj.weight": "model-00024.safetensors", "model.layers.77.self_attn.o_proj.weight": "model-00024.safetensors", "model.layers.77.input_layernorm.weight": "model-00024.safetensors", "model.layers.77.mlp.down_proj.weight": "model-00024.safetensors", "model.layers.77.post_attention_layernorm.weight": "model-00024.safetensors", "model.layers.77.self_attn.q_proj.weight": "model-00024.safetensors", "model.layers.77.self_attn.k_proj.weight": "model-00024.safetensors", "model.layers.77.self_attn.v_proj.weight": "model-00024.safetensors", "model.layers.77.mlp.gate_proj.weight": "model-00024.safetensors", "model.layers.77.mlp.up_proj.weight": "model-00024.safetensors", "model.layers.78.self_attn.o_proj.weight": "model-00024.safetensors", "model.layers.78.input_layernorm.weight": "model-00024.safetensors", "model.layers.78.mlp.down_proj.weight": "model-00024.safetensors", "model.layers.78.post_attention_layernorm.weight": "model-00024.safetensors", "model.layers.78.self_attn.q_proj.weight": "model-00024.safetensors", "model.layers.78.self_attn.k_proj.weight": "model-00024.safetensors", "model.layers.78.self_attn.v_proj.weight": "model-00024.safetensors", "model.layers.78.mlp.gate_proj.weight": "model-00024.safetensors", "model.layers.78.mlp.up_proj.weight": "model-00024.safetensors", "model.layers.79.self_attn.o_proj.weight": "model-00024.safetensors", "model.layers.79.input_layernorm.weight": "model-00024.safetensors", "model.layers.79.mlp.down_proj.weight": "model-00024.safetensors", "model.layers.79.post_attention_layernorm.weight": "model-00024.safetensors", "model.layers.79.self_attn.q_proj.weight": "model-00024.safetensors", "model.layers.79.self_attn.k_proj.weight": "model-00024.safetensors", "model.layers.79.self_attn.v_proj.weight": "model-00024.safetensors", "model.layers.79.mlp.gate_proj.weight": "model-00025.safetensors", "model.layers.79.mlp.up_proj.weight": "model-00025.safetensors", "model.layers.80.self_attn.o_proj.weight": "model-00025.safetensors", "model.layers.80.input_layernorm.weight": "model-00025.safetensors", "model.layers.80.mlp.down_proj.weight": "model-00025.safetensors", "model.layers.80.post_attention_layernorm.weight": "model-00025.safetensors", "model.layers.80.self_attn.q_proj.weight": "model-00025.safetensors", "model.layers.80.self_attn.k_proj.weight": "model-00025.safetensors", "model.layers.80.self_attn.v_proj.weight": "model-00025.safetensors", "model.layers.80.mlp.gate_proj.weight": "model-00025.safetensors", "model.layers.80.mlp.up_proj.weight": "model-00025.safetensors", "model.layers.81.self_attn.o_proj.weight": "model-00025.safetensors", "model.layers.81.input_layernorm.weight": "model-00025.safetensors", "model.layers.81.mlp.down_proj.weight": "model-00025.safetensors", "model.layers.81.post_attention_layernorm.weight": "model-00025.safetensors", "model.layers.81.self_attn.q_proj.weight": "model-00025.safetensors", "model.layers.81.self_attn.k_proj.weight": "model-00025.safetensors", "model.layers.81.self_attn.v_proj.weight": "model-00025.safetensors", "model.layers.81.mlp.gate_proj.weight": "model-00025.safetensors", "model.layers.81.mlp.up_proj.weight": "model-00025.safetensors", "model.layers.82.self_attn.o_proj.weight": "model-00025.safetensors", "model.layers.82.input_layernorm.weight": "model-00025.safetensors", "model.layers.82.mlp.down_proj.weight": "model-00025.safetensors", "model.layers.82.post_attention_layernorm.weight": "model-00025.safetensors", "model.layers.82.self_attn.q_proj.weight": "model-00025.safetensors", "model.layers.82.self_attn.k_proj.weight": "model-00025.safetensors", "model.layers.82.self_attn.v_proj.weight": "model-00025.safetensors", "model.layers.82.mlp.gate_proj.weight": "model-00025.safetensors", "model.layers.82.mlp.up_proj.weight": "model-00026.safetensors", "model.layers.83.self_attn.o_proj.weight": "model-00026.safetensors", "model.layers.83.input_layernorm.weight": "model-00026.safetensors", "model.layers.83.mlp.down_proj.weight": "model-00026.safetensors", "model.layers.83.post_attention_layernorm.weight": "model-00026.safetensors", "model.layers.83.self_attn.q_proj.weight": "model-00026.safetensors", "model.layers.83.self_attn.k_proj.weight": "model-00026.safetensors", "model.layers.83.self_attn.v_proj.weight": "model-00026.safetensors", "model.layers.83.mlp.gate_proj.weight": "model-00026.safetensors", "model.layers.83.mlp.up_proj.weight": "model-00026.safetensors", "model.layers.84.self_attn.o_proj.weight": "model-00026.safetensors", "model.layers.84.input_layernorm.weight": "model-00026.safetensors", "model.layers.84.mlp.down_proj.weight": "model-00026.safetensors", "model.layers.84.post_attention_layernorm.weight": "model-00026.safetensors", "model.layers.84.self_attn.q_proj.weight": "model-00026.safetensors", "model.layers.84.self_attn.k_proj.weight": "model-00026.safetensors", "model.layers.84.self_attn.v_proj.weight": "model-00026.safetensors", "model.layers.84.mlp.gate_proj.weight": "model-00026.safetensors", "model.layers.84.mlp.up_proj.weight": "model-00026.safetensors", "model.layers.85.self_attn.o_proj.weight": "model-00026.safetensors", "model.layers.85.input_layernorm.weight": "model-00026.safetensors", "model.layers.85.mlp.down_proj.weight": "model-00026.safetensors", "model.layers.85.post_attention_layernorm.weight": "model-00026.safetensors", "model.layers.85.self_attn.q_proj.weight": "model-00026.safetensors", "model.layers.85.self_attn.k_proj.weight": "model-00026.safetensors", "model.layers.85.self_attn.v_proj.weight": "model-00026.safetensors", "model.layers.85.mlp.gate_proj.weight": "model-00026.safetensors", "model.layers.85.mlp.up_proj.weight": "model-00026.safetensors", "model.layers.86.self_attn.o_proj.weight": "model-00026.safetensors", "model.layers.86.input_layernorm.weight": "model-00026.safetensors", "model.layers.86.mlp.down_proj.weight": "model-00027.safetensors", "model.layers.86.post_attention_layernorm.weight": "model-00027.safetensors", "model.layers.86.self_attn.q_proj.weight": "model-00027.safetensors", "model.layers.86.self_attn.k_proj.weight": "model-00027.safetensors", "model.layers.86.self_attn.v_proj.weight": "model-00027.safetensors", "model.layers.86.mlp.gate_proj.weight": "model-00027.safetensors", "model.layers.86.mlp.up_proj.weight": "model-00027.safetensors", "model.layers.87.self_attn.o_proj.weight": "model-00027.safetensors", "model.layers.87.input_layernorm.weight": "model-00027.safetensors", "model.layers.87.mlp.down_proj.weight": "model-00027.safetensors", "model.layers.87.post_attention_layernorm.weight": "model-00027.safetensors", "model.layers.87.self_attn.q_proj.weight": "model-00027.safetensors", "model.layers.87.self_attn.k_proj.weight": "model-00027.safetensors", "model.layers.87.self_attn.v_proj.weight": "model-00027.safetensors", "model.layers.87.mlp.gate_proj.weight": "model-00027.safetensors", "model.layers.87.mlp.up_proj.weight": "model-00027.safetensors", "model.norm.weight": "model-00027.safetensors", "lm_head.weight": "model-00027.safetensors"}}

modeling_stepaudio.py

@@ -0,0 +1,392 @@
+import math
+from typing import Optional, Tuple, Union, List
+
+import torch
+import torch.utils.checkpoint
+from torch import nn
+from transformers.generation import GenerationMixin
+
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import logging
+from .configuration_stepaudio import StepAudioConfig
+from transformers.cache_utils import Cache, DynamicCache
+from einops import rearrange
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPast,
+    CausalLMOutputWithPast,
+)
+
+logger = logging.get_logger(__name__)
+
+
+def build_alibi_cache(block_size, n_heads, dtype, device):
+    # get slopes
+    n = 2 ** math.floor(math.log2(n_heads))  # nearest 2**n to n_heads
+    m0 = 2.0 ** (-8.0 / n)
+    # 2^(-8/n), 2^(-8*2/n), 2^(-8*3/n), ...
+    slopes = torch.pow(m0, torch.arange(1, n + 1))
+    if n < n_heads:
+        m1 = 2.0 ** (-4.0 / n)
+        # 2^(-8/(2n)), 2^(-8*3/(2n)), 2^(-8*5/(2n)), ...
+        mm = torch.pow(m1, torch.arange(1, 1 + 2 * (n_heads - n), 2))
+        slopes = torch.cat([slopes, mm])
+    slopes = slopes.to(device)
+
+    tril = torch.tril(torch.ones(1, 1, block_size, block_size, device=device))
+
+    bias_rows = torch.arange(block_size, device=device).view(1, -1)
+    bias_cols = torch.arange(block_size, device=device).view(-1, 1)
+    bias = -torch.sqrt(bias_cols - bias_rows)
+    bias = bias.view(1, block_size, block_size) * slopes.view(-1, 1, 1)
+    bias = bias.masked_fill(tril == 0, float("-inf"))
+
+    return bias.type(dtype)
+
+
+class StepAudioRMSNorm(torch.nn.Module):
+    def __init__(self, hidden_size, eps=1e-5):
+        super().__init__()
+        self.weight = torch.nn.Parameter(torch.ones(hidden_size))
+        self.eps = eps
+
+    def forward(self, x: torch.Tensor):
+        var = x.float().pow(2).mean(-1, keepdim=True)
+        x = x * torch.rsqrt(var + self.eps).to(x.dtype)
+        x = x * self.weight
+        return x
+
+
+class StepAudioAttention(torch.nn.Module):
+    def __init__(self, hidden_size, num_heads, num_groups, layer_idx: int):
+        super().__init__()
+
+        self.num_heads = num_heads
+        self.num_groups = num_groups
+        self.hidden_size = hidden_size
+        self.head_dim = hidden_size // num_heads
+
+        self.q_proj = torch.nn.Linear(hidden_size, hidden_size, bias=False)
+        self.k_proj = torch.nn.Linear(
+            hidden_size, num_groups * self.head_dim, bias=False
+        )
+        self.v_proj = torch.nn.Linear(
+            hidden_size, num_groups * self.head_dim, bias=False
+        )
+        self.o_proj = torch.nn.Linear(hidden_size, hidden_size, bias=False)
+
+        self.layer_idx = layer_idx
+
+    def forward(
+        self,
+        x: torch.Tensor,
+        past_key_value: Optional[Cache] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+    ):
+
+        q: torch.Tensor = self.q_proj(x)
+        k: torch.Tensor = self.k_proj(x)
+        v: torch.Tensor = self.v_proj(x)
+        if past_key_value is not None:
+            cache_kwargs = {"cache_position": cache_position}
+            k, v = past_key_value.update(k, v, self.layer_idx, cache_kwargs)
+
+        q = rearrange(q, "b s (h d) -> b s h d", h=self.num_heads)
+        k = rearrange(k, "b s (g d) -> b s g d", g=self.num_groups)
+        v = rearrange(v, "b s (g d) -> b s g d", g=self.num_groups)
+
+        k = k.repeat_interleave(self.num_heads // self.num_groups, dim=-2)
+        v = v.repeat_interleave(self.num_heads // self.num_groups, dim=-2)
+
+        attention_mask = build_alibi_cache(
+            k.size(1), self.num_heads, dtype=q.dtype, device=q.device
+        )[:, :, -q.size(1) :, :].contiguous()
+
+        q = q.transpose(1, 2)
+        k = k.transpose(1, 2)
+        v = v.transpose(1, 2)
+
+        o: torch.Tensor = torch.nn.functional.scaled_dot_product_attention(
+            q, k, v, attn_mask=attention_mask
+        )
+        o = o.transpose(1, 2).flatten(-2, -1)
+
+        o = self.o_proj(o)
+        return o
+
+
+class StepAudioMLP(torch.nn.Module):
+    def __init__(self, hidden_size, intermediate_size):
+        super().__init__()
+        self.gate_proj = torch.nn.Linear(hidden_size, intermediate_size, bias=False)
+        self.up_proj = torch.nn.Linear(hidden_size, intermediate_size, bias=False)
+        self.down_proj = torch.nn.Linear(intermediate_size, hidden_size, bias=False)
+
+    def forward(self, x):
+        gate = self.gate_proj(x)
+        up = self.up_proj(x)
+        x = torch.nn.functional.silu(gate) * up
+        x = self.down_proj(x)
+        return x
+
+
+class StepAudioLayer(torch.nn.Module):
+    def __init__(self, config: StepAudioConfig, layer_idx: int):
+        super().__init__()
+        self.layer_idx = layer_idx
+        self.self_attn = StepAudioAttention(
+            hidden_size=config.hidden_size,
+            num_heads=config.num_attention_heads,
+            num_groups=config.num_attention_groups,
+            layer_idx=layer_idx,
+        )
+        self.mlp = StepAudioMLP(
+            hidden_size=config.hidden_size,
+            intermediate_size=config.intermediate_size,
+        )
+        self.input_layernorm = StepAudioRMSNorm(
+            hidden_size=config.hidden_size, eps=config.rms_norm_eps
+        )
+        self.post_attention_layernorm = StepAudioRMSNorm(
+            hidden_size=config.hidden_size, eps=config.rms_norm_eps
+        )
+
+    def forward(
+        self,
+        x,
+        attention_mask: Optional[torch.Tensor] = None,
+        past_key_value: Optional[Cache] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+    ):
+        def f(x):
+            x = self.input_layernorm(x)
+            x = self.self_attn(x, past_key_value, attention_mask, cache_position)
+            return x
+
+        x = x + f(x)
+
+        def f(x):
+            x = self.post_attention_layernorm(x)
+            x = self.mlp(x)
+            return x
+
+        x = x + f(x)
+
+        return x
+
+
+class StepAudioPreTrainedModel(PreTrainedModel):
+    config_class = StepAudioConfig
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["StepAudioLayer"]
+    _skip_keys_device_placement = ["past_key_values"]
+    _supports_cache_class = True
+    _supports_static_cache = True
+
+    def _init_weights(self, module):
+        std = self.config.initializer_range
+        if isinstance(module, nn.Linear):
+            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_()
+
+
+class StepAudioModel(StepAudioPreTrainedModel):
+    """
+    Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`LlamaDecoderLayer`]
+
+    Args:
+        config: StepAudioConfig
+    """
+
+    def __init__(self, config: StepAudioConfig):
+        super().__init__(config)
+        self.config = config
+        self.embed_tokens = torch.nn.Embedding(config.vocab_size, config.hidden_size)
+
+        self.layers = torch.nn.Sequential(
+            *[
+                StepAudioLayer(config, layer_idx)
+                for layer_idx in range(config.num_hidden_layers)
+            ]
+        )
+
+        self.norm = StepAudioRMSNorm(
+            hidden_size=config.hidden_size, eps=config.rms_norm_eps
+        )
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.embed_tokens = value
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        past_key_values: Optional[Cache] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        output_attentions = False
+        output_hidden_states = False
+
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = (
+            return_dict if return_dict is not None else self.config.use_return_dict
+        )
+
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError(
+                "You must specify exactly one of input_ids or inputs_embeds"
+            )
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+
+        if use_cache and past_key_values is None:
+            past_key_values = DynamicCache()
+
+        if cache_position is None:
+            past_seen_tokens = (
+                past_key_values.get_seq_length() if past_key_values is not None else 0
+            )
+            cache_position = torch.arange(
+                past_seen_tokens,
+                past_seen_tokens + inputs_embeds.shape[1],
+                device=inputs_embeds.device,
+            )
+
+        causal_mask = attention_mask
+
+        hidden_states = inputs_embeds
+
+        for decoder_layer in self.layers[: self.config.num_hidden_layers]:
+            layer_outputs = decoder_layer(
+                hidden_states,
+                attention_mask=causal_mask,
+                past_key_value=past_key_values,
+                cache_position=cache_position,
+            )
+
+            hidden_states = layer_outputs
+
+        hidden_states = self.norm(hidden_states)
+
+        output = BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values if use_cache else None,
+            hidden_states=hidden_states,
+            attentions=None,
+        )
+        return output if return_dict else output.to_tuple()
+
+
+class StepAudioForCausalLM(StepAudioPreTrainedModel, GenerationMixin):
+    _tied_weights_keys = ["lm_head.weight"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = StepAudioModel(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.embed_tokens = value
+
+    # def get_output_embeddings(self):
+    #     return self.lm_head
+
+    # def set_output_embeddings(self, new_embeddings):
+    #     self.lm_head = new_embeddings
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def get_decoder(self):
+        return self.model
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        num_logits_to_keep: int = 0,
+    ) -> Union[Tuple, CausalLMOutputWithPast]:
+        # output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_attentions = False
+        output_hidden_states = False
+        # 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
+        )
+
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            cache_position=cache_position,
+        )
+
+        hidden_states = outputs[0]
+        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
+
+        logits = self.lm_head(hidden_states)
+
+        # logits = torch.matmul(hidden_states, lm_stat)
+
+        loss = None
+        if labels is not None:
+            loss = self.loss_function(
+                logits=logits,
+                labels=labels,
+                vocab_size=self.config.vocab_size,
+                **kwargs
+            )
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25e122d9205d035033a9994c4d46a6a1b467a938654e4178fc0e5f4f5d610674
+size 1264044

tokenizer_config.json

@@ -0,0 +1,14 @@
+{
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "legacy": false,
+  "model_max_length": 65536,
+  "pad_token": "<unk>",
+  "padding_side": "left",
+  "sp_model_kwargs": {},
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}
+