Upload folder using huggingface_hub

config.json

@@ -0,0 +1,37 @@
+{
+  "architectures": ["MonetForCausalLM"],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "modeling_monet.MonetConfig",
+    "AutoModelForCausalLM": "modeling_monet.MonetForCausalLM"
+  },
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "hidden_act": "relu2",
+  "hidden_size": 3072,
+  "initializer_range": 0.02,
+  "intermediate_size": null,
+  "max_position_embeddings": 2048,
+  "mlp_bias": null,
+  "model_type": "monet",
+  "moe_decompose": "vertical",
+  "moe_dim": 24,
+  "moe_experts": 512,
+  "moe_groups": 4,
+  "moe_heads": 12,
+  "moe_topk": 8,
+  "num_attention_heads": 24,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 24,
+  "output_router_probs": false,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-6,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.44.2",
+  "use_cache": true,
+  "vocab_size": 32000
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "transformers_version": "4.44.2"
+}

interpretation/examples-0.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a6fa25d480ce0f8757def13084b7d7a17abd7029da0781686fed1586ee97ca1
+size 131466479

interpretation/examples-1.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b533695375c5d07b9b7befcd06ffc70b66f4fd037143e726ab7d40aafee8164
+size 370184015

interpretation/examples-2.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3dcbfd290e26c738a148c46b665ea889f0079e140edccc27880e13f6cc519bbc
+size 330305905

interpretation/examples-3.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:257f2e669c7a74c4ac97785a52ab399f7bf8dfd32309549aea2c057b8826f37d
+size 326833157

interpretation/examples-4.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6b95b9dc2bd45055307d090a25b1987df5b940ca8937b66d9155058eac627d6
+size 367214039

interpretation/examples-5.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a7c02f4247c84f286e5ae7bcae85c24b51ad90ed41aacc83b77656a4fd2c6691
+size 396063525

interpretation/examples-6.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:45a6bd63004ce2695e9986c00ab0264df56b3c6438be705fabeb6cf174fc427e
+size 400323805

interpretation/examples-7.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2af2e0def4a0c5c274fdbf2e91e3eca80941fa178562e7b08d19509a8b1d0120
+size 387813131

interpretation/inputs.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d8dfa78c58772da42d1be1995a039672f4d4ed95e319b1fe94fe8a4d0980eb7
+size 393938007

interpretation/routings-0.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f698af3294aea8159251c61a13c7bf279abad7b229b1d7e08f1f0c9f8e71210
+size 9014683873

interpretation/routings-1.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ead12ee49310d45e03c82c49feced14990633df24f9c71b0543f79e7915a9f2
+size 11678536677

interpretation/routings-2.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b98a6a5308987b69e4cf6669b482ec8fe88055b47fc03ba361831ce7b6eb6079
+size 9398043059

interpretation/routings-3.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51984878889eb6eb5b9d8cde97f10c1f37de9bbc5cddd9bbe91267b44245e45d
+size 6722608619

interpretation/routings-4.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a7e5009a156b9c347178e4f4576ec0acf3b242e222d9585806e7a24012509c5f
+size 7572154455

interpretation/routings-5.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:258da30fbdf3df6c128d95289856f21d90d9bc6d04aeb451e6ab7a7a4575122c
+size 8736531053

interpretation/routings-6.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e708210d7d8c5cf4a5f9ff822330159a1fb4d4f98435c40e50fba7ecbef76207
+size 10045081221

interpretation/routings-7.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dfba557f17b0bdbfcf586766db2a9168b0fd1aa902a530cb23c3108d41d189c3
+size 10420552384

model-00001-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cadca3b054a15a013415292c07d45dbd47583ef1f0bdf694c58739f286112d2e
+size 4994582160

model-00002-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:54f1e2964bf923a976612e5ef7defcb18f0b4aff42ea45501baee5a93ac451f0
+size 3352284128

model.safetensors.index.json

@@ -0,0 +1,586 @@
+{
+  "metadata": {
+    "total_size": 8346803072
+  },
+  "weight_map": {
+    "lm_head.weight": "model-00002-of-00002.safetensors",
+    "model.embed_tokens.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.0.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.0.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.0.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.0.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.router.norm1.num_batches_tracked": "model-00001-of-00002.safetensors",
+    "model.layers.0.router.norm1.running_mean": "model-00001-of-00002.safetensors",
+    "model.layers.0.router.norm1.running_var": "model-00001-of-00002.safetensors",
+    "model.layers.0.router.norm2.num_batches_tracked": "model-00001-of-00002.safetensors",
+    "model.layers.0.router.norm2.running_mean": "model-00001-of-00002.safetensors",
+    "model.layers.0.router.norm2.running_var": "model-00001-of-00002.safetensors",
+    "model.layers.0.router.w1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.router.w2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.1.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.1.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.1.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.1.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.10.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.10.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.10.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.10.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.11.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.11.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.11.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.11.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.12.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.12.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.12.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.12.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.router.norm1.num_batches_tracked": "model-00001-of-00002.safetensors",
+    "model.layers.12.router.norm1.running_mean": "model-00001-of-00002.safetensors",
+    "model.layers.12.router.norm1.running_var": "model-00001-of-00002.safetensors",
+    "model.layers.12.router.norm2.num_batches_tracked": "model-00001-of-00002.safetensors",
+    "model.layers.12.router.norm2.running_mean": "model-00001-of-00002.safetensors",
+    "model.layers.12.router.norm2.running_var": "model-00001-of-00002.safetensors",
+    "model.layers.12.router.w1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.router.w2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.13.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.13.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.13.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.13.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.14.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.14.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.14.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.14.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.15.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.15.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.15.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.15.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.16.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.16.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.16.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.16.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.router.norm1.num_batches_tracked": "model-00001-of-00002.safetensors",
+    "model.layers.16.router.norm1.running_mean": "model-00001-of-00002.safetensors",
+    "model.layers.16.router.norm1.running_var": "model-00001-of-00002.safetensors",
+    "model.layers.16.router.norm2.num_batches_tracked": "model-00001-of-00002.safetensors",
+    "model.layers.16.router.norm2.running_mean": "model-00001-of-00002.safetensors",
+    "model.layers.16.router.norm2.running_var": "model-00001-of-00002.safetensors",
+    "model.layers.16.router.w1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.router.w2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.17.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.17.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.17.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.17.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.18.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.18.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.18.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.18.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.moe.b1": "model-00002-of-00002.safetensors",
+    "model.layers.19.moe.b2": "model-00002-of-00002.safetensors",
+    "model.layers.19.moe.u1.bias": "model-00002-of-00002.safetensors",
+    "model.layers.19.moe.u1.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.moe.u2.bias": "model-00002-of-00002.safetensors",
+    "model.layers.19.moe.u2.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.moe.v11.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.moe.v12.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.moe.v21.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.moe.v22.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.19.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.2.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.2.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.2.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.2.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.moe.b1": "model-00002-of-00002.safetensors",
+    "model.layers.20.moe.b2": "model-00002-of-00002.safetensors",
+    "model.layers.20.moe.u1.bias": "model-00002-of-00002.safetensors",
+    "model.layers.20.moe.u1.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.moe.u2.bias": "model-00002-of-00002.safetensors",
+    "model.layers.20.moe.u2.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.moe.v11.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.moe.v12.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.moe.v21.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.moe.v22.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.router.norm1.num_batches_tracked": "model-00002-of-00002.safetensors",
+    "model.layers.20.router.norm1.running_mean": "model-00002-of-00002.safetensors",
+    "model.layers.20.router.norm1.running_var": "model-00002-of-00002.safetensors",
+    "model.layers.20.router.norm2.num_batches_tracked": "model-00002-of-00002.safetensors",
+    "model.layers.20.router.norm2.running_mean": "model-00002-of-00002.safetensors",
+    "model.layers.20.router.norm2.running_var": "model-00002-of-00002.safetensors",
+    "model.layers.20.router.w1.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.router.w2.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.20.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.moe.b1": "model-00002-of-00002.safetensors",
+    "model.layers.21.moe.b2": "model-00002-of-00002.safetensors",
+    "model.layers.21.moe.u1.bias": "model-00002-of-00002.safetensors",
+    "model.layers.21.moe.u1.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.moe.u2.bias": "model-00002-of-00002.safetensors",
+    "model.layers.21.moe.u2.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.moe.v11.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.moe.v12.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.moe.v21.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.moe.v22.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.21.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.moe.b1": "model-00002-of-00002.safetensors",
+    "model.layers.22.moe.b2": "model-00002-of-00002.safetensors",
+    "model.layers.22.moe.u1.bias": "model-00002-of-00002.safetensors",
+    "model.layers.22.moe.u1.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.moe.u2.bias": "model-00002-of-00002.safetensors",
+    "model.layers.22.moe.u2.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.moe.v11.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.moe.v12.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.moe.v21.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.moe.v22.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.22.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.moe.b1": "model-00002-of-00002.safetensors",
+    "model.layers.23.moe.b2": "model-00002-of-00002.safetensors",
+    "model.layers.23.moe.u1.bias": "model-00002-of-00002.safetensors",
+    "model.layers.23.moe.u1.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.moe.u2.bias": "model-00002-of-00002.safetensors",
+    "model.layers.23.moe.u2.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.moe.v11.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.moe.v12.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.moe.v21.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.moe.v22.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.23.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.moe.b1": "model-00002-of-00002.safetensors",
+    "model.layers.24.moe.b2": "model-00002-of-00002.safetensors",
+    "model.layers.24.moe.u1.bias": "model-00002-of-00002.safetensors",
+    "model.layers.24.moe.u1.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.moe.u2.bias": "model-00002-of-00002.safetensors",
+    "model.layers.24.moe.u2.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.moe.v11.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.moe.v12.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.moe.v21.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.moe.v22.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.router.norm1.num_batches_tracked": "model-00002-of-00002.safetensors",
+    "model.layers.24.router.norm1.running_mean": "model-00002-of-00002.safetensors",
+    "model.layers.24.router.norm1.running_var": "model-00002-of-00002.safetensors",
+    "model.layers.24.router.norm2.num_batches_tracked": "model-00002-of-00002.safetensors",
+    "model.layers.24.router.norm2.running_mean": "model-00002-of-00002.safetensors",
+    "model.layers.24.router.norm2.running_var": "model-00002-of-00002.safetensors",
+    "model.layers.24.router.w1.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.router.w2.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.24.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.moe.b1": "model-00002-of-00002.safetensors",
+    "model.layers.25.moe.b2": "model-00002-of-00002.safetensors",
+    "model.layers.25.moe.u1.bias": "model-00002-of-00002.safetensors",
+    "model.layers.25.moe.u1.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.moe.u2.bias": "model-00002-of-00002.safetensors",
+    "model.layers.25.moe.u2.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.moe.v11.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.moe.v12.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.moe.v21.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.moe.v22.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.25.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.moe.b1": "model-00002-of-00002.safetensors",
+    "model.layers.26.moe.b2": "model-00002-of-00002.safetensors",
+    "model.layers.26.moe.u1.bias": "model-00002-of-00002.safetensors",
+    "model.layers.26.moe.u1.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.moe.u2.bias": "model-00002-of-00002.safetensors",
+    "model.layers.26.moe.u2.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.moe.v11.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.moe.v12.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.moe.v21.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.moe.v22.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.26.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.moe.b1": "model-00002-of-00002.safetensors",
+    "model.layers.27.moe.b2": "model-00002-of-00002.safetensors",
+    "model.layers.27.moe.u1.bias": "model-00002-of-00002.safetensors",
+    "model.layers.27.moe.u1.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.moe.u2.bias": "model-00002-of-00002.safetensors",
+    "model.layers.27.moe.u2.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.moe.v11.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.moe.v12.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.moe.v21.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.moe.v22.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.27.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.moe.b1": "model-00002-of-00002.safetensors",
+    "model.layers.28.moe.b2": "model-00002-of-00002.safetensors",
+    "model.layers.28.moe.u1.bias": "model-00002-of-00002.safetensors",
+    "model.layers.28.moe.u1.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.moe.u2.bias": "model-00002-of-00002.safetensors",
+    "model.layers.28.moe.u2.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.moe.v11.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.moe.v12.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.moe.v21.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.moe.v22.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.router.norm1.num_batches_tracked": "model-00002-of-00002.safetensors",
+    "model.layers.28.router.norm1.running_mean": "model-00002-of-00002.safetensors",
+    "model.layers.28.router.norm1.running_var": "model-00002-of-00002.safetensors",
+    "model.layers.28.router.norm2.num_batches_tracked": "model-00002-of-00002.safetensors",
+    "model.layers.28.router.norm2.running_mean": "model-00002-of-00002.safetensors",
+    "model.layers.28.router.norm2.running_var": "model-00002-of-00002.safetensors",
+    "model.layers.28.router.w1.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.router.w2.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.28.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.moe.b1": "model-00002-of-00002.safetensors",
+    "model.layers.29.moe.b2": "model-00002-of-00002.safetensors",
+    "model.layers.29.moe.u1.bias": "model-00002-of-00002.safetensors",
+    "model.layers.29.moe.u1.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.moe.u2.bias": "model-00002-of-00002.safetensors",
+    "model.layers.29.moe.u2.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.moe.v11.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.moe.v12.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.moe.v21.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.moe.v22.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.29.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.3.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.3.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.3.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.3.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.3.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.moe.b1": "model-00002-of-00002.safetensors",
+    "model.layers.30.moe.b2": "model-00002-of-00002.safetensors",
+    "model.layers.30.moe.u1.bias": "model-00002-of-00002.safetensors",
+    "model.layers.30.moe.u1.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.moe.u2.bias": "model-00002-of-00002.safetensors",
+    "model.layers.30.moe.u2.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.moe.v11.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.moe.v12.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.moe.v21.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.moe.v22.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.30.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.moe.b1": "model-00002-of-00002.safetensors",
+    "model.layers.31.moe.b2": "model-00002-of-00002.safetensors",
+    "model.layers.31.moe.u1.bias": "model-00002-of-00002.safetensors",
+    "model.layers.31.moe.u1.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.moe.u2.bias": "model-00002-of-00002.safetensors",
+    "model.layers.31.moe.u2.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.moe.v11.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.moe.v12.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.moe.v21.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.moe.v22.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.31.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.4.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.4.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.4.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.4.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.4.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.router.norm1.num_batches_tracked": "model-00001-of-00002.safetensors",
+    "model.layers.4.router.norm1.running_mean": "model-00001-of-00002.safetensors",
+    "model.layers.4.router.norm1.running_var": "model-00001-of-00002.safetensors",
+    "model.layers.4.router.norm2.num_batches_tracked": "model-00001-of-00002.safetensors",
+    "model.layers.4.router.norm2.running_mean": "model-00001-of-00002.safetensors",
+    "model.layers.4.router.norm2.running_var": "model-00001-of-00002.safetensors",
+    "model.layers.4.router.w1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.router.w2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.5.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.5.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.5.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.5.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.6.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.6.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.6.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.6.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.7.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.7.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.7.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.7.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.8.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.8.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.8.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.8.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.router.norm1.num_batches_tracked": "model-00001-of-00002.safetensors",
+    "model.layers.8.router.norm1.running_mean": "model-00001-of-00002.safetensors",
+    "model.layers.8.router.norm1.running_var": "model-00001-of-00002.safetensors",
+    "model.layers.8.router.norm2.num_batches_tracked": "model-00001-of-00002.safetensors",
+    "model.layers.8.router.norm2.running_mean": "model-00001-of-00002.safetensors",
+    "model.layers.8.router.norm2.running_var": "model-00001-of-00002.safetensors",
+    "model.layers.8.router.w1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.router.w2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.moe.b1": "model-00001-of-00002.safetensors",
+    "model.layers.9.moe.b2": "model-00001-of-00002.safetensors",
+    "model.layers.9.moe.u1.bias": "model-00001-of-00002.safetensors",
+    "model.layers.9.moe.u1.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.moe.u2.bias": "model-00001-of-00002.safetensors",
+    "model.layers.9.moe.u2.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.moe.v11.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.moe.v12.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.moe.v21.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.moe.v22.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
+    "model.norm.weight": "model-00002-of-00002.safetensors"
+  }
+}

modeling_monet.py

@@ -0,0 +1,663 @@
+# fmt: off
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+import torch
+import torch.utils.checkpoint
+from scipy.stats import norm
+from torch import nn
+from torch.nn import CrossEntropyLoss
+from transformers.activations import ACT2FN
+from transformers.cache_utils import Cache, DynamicCache, StaticCache
+from transformers.modeling_attn_mask_utils import AttentionMaskConverter
+from transformers.modeling_utils import PreTrainedModel
+from transformers.models.llama.configuration_llama import LlamaConfig
+from transformers.models.llama.modeling_llama import (
+    LLAMA_ATTENTION_CLASSES,
+    LlamaRMSNorm,
+)
+from transformers.utils import ModelOutput, logging
+
+logger = logging.get_logger(__name__)
+
+
+@dataclass
+class MonetModelOutputWithPast(ModelOutput):
+    last_hidden_state: torch.FloatTensor = None
+    past_key_values: tuple[tuple[torch.FloatTensor]] | None = None
+    hidden_states: tuple[torch.FloatTensor, ...] | None = None
+    attentions: tuple[torch.FloatTensor, ...] | None = None
+    router_probs: tuple[tuple[torch.FloatTensor, ...], ...] | None = None
+
+
+@dataclass
+class MonetCausalLMOutputWithPast(ModelOutput):
+    loss: torch.FloatTensor | None = None
+    aux_loss: torch.FloatTensor | None = None
+    logits: torch.FloatTensor = None
+    past_key_values: tuple[tuple[torch.FloatTensor]] | None = None
+    hidden_states: tuple[torch.FloatTensor, ...] | None = None
+    attentions: tuple[torch.FloatTensor, ...] | None = None
+    router_probs: tuple[tuple[torch.FloatTensor, ...], ...] | None = None
+
+
+class MonetConfig(LlamaConfig):
+    model_type = "monet"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=32000,
+        hidden_size=4096,
+        intermediate_size=None,
+        num_hidden_layers=32,
+        num_attention_heads=32,
+        num_key_value_heads=None,
+        hidden_act="relu2",
+        max_position_embeddings=2048,
+        initializer_range=0.02,
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=1,
+        eos_token_id=2,
+        pretraining_tp=1,
+        tie_word_embeddings=False,
+        rope_theta=10000.0,
+        rope_scaling=None,
+        attention_bias=False,
+        attention_dropout=0.0,
+        mlp_bias=None,
+        moe_dim=8,
+        moe_heads=8,
+        moe_experts=512,
+        moe_topk=32,
+        moe_groups=4,
+        moe_decompose="vertical",
+        output_router_probs=False,
+        **kwargs,
+    ):
+        self.moe_dim = moe_dim
+        self.moe_heads = moe_heads
+        self.moe_experts = moe_experts
+        self.moe_topk = moe_topk
+        self.moe_groups = moe_groups
+        self.moe_decompose = moe_decompose
+        self.output_router_probs = output_router_probs
+
+        super().__init__(
+            vocab_size=vocab_size,
+            hidden_size=hidden_size,
+            intermediate_size=intermediate_size,
+            num_hidden_layers=num_hidden_layers,
+            num_attention_heads=num_attention_heads,
+            num_key_value_heads=num_key_value_heads,
+            hidden_act=hidden_act,
+            max_position_embeddings=max_position_embeddings,
+            initializer_range=initializer_range,
+            rms_norm_eps=rms_norm_eps,
+            use_cache=use_cache,
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            pretraining_tp=pretraining_tp,
+            tie_word_embeddings=tie_word_embeddings,
+            rope_theta=rope_theta,
+            rope_scaling=rope_scaling,
+            attention_bias=attention_bias,
+            attention_dropout=attention_dropout,
+            mlp_bias=mlp_bias,
+            **kwargs,
+        )
+
+
+class MonetRouter(nn.Module):
+    def __init__(self, config: MonetConfig):
+        super().__init__()
+        self.config = config
+        flatten_shape = config.moe_heads * config.moe_experts
+
+        self.w1 = nn.Linear(config.hidden_size, flatten_shape, bias=False)
+        self.w2 = nn.Linear(config.hidden_size, flatten_shape, bias=False)
+        self.norm1 = nn.BatchNorm1d(config.moe_heads, affine=False)
+        self.norm2 = nn.BatchNorm1d(config.moe_heads, affine=False)
+
+    def forward(self, x: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]:
+        g1z = self.w1(x).unflatten(-1, (self.config.moe_heads, -1)).float()
+        g2z = self.w2(x).unflatten(-1, (self.config.moe_heads, -1)).float()
+
+        g1n = self.norm1(g1z.transpose(2, 3).flatten(0, -2))
+        g2n = self.norm2(g2z.transpose(2, 3).flatten(0, -2))
+        g1n = g1n.view(g1z.size(0), g1z.size(1), g1z.size(3), -1).transpose(2, 3)
+        g2n = g2n.view(g2z.size(0), g2z.size(1), g2z.size(3), -1).transpose(2, 3)
+
+        sigma = float(norm.ppf(1 - self.config.moe_topk / self.config.moe_experts))
+        g1s = g1n.amax(-1, keepdim=True).clamp_max_(sigma)
+        g2s = g2n.amax(-1, keepdim=True).clamp_max_(sigma)
+
+        g1 = nn.functional.softmax(torch.where(g1n >= g1s, g1z, -1e10), dim=-1)
+        g2 = nn.functional.softmax(torch.where(g2n >= g2s, g2z, -1e10), dim=-1)
+        return g1, g2
+
+
+class MonetMoVDE(nn.Module):
+    def __init__(self, config: MonetConfig):
+        super().__init__()
+        self.config = config
+        self.act_fn = ACT2FN[config.hidden_act]
+        flatten_shape = config.moe_experts * config.moe_dim // 2
+
+        self.u1 = nn.Linear(config.hidden_size, flatten_shape)
+        self.u2 = nn.Linear(config.hidden_size, flatten_shape)
+
+        self.v11 = nn.Linear(flatten_shape, config.hidden_size // 2, bias=False)
+        self.v12 = nn.Linear(flatten_shape, config.hidden_size // 2, bias=False)
+        self.v21 = nn.Linear(flatten_shape, config.hidden_size // 2, bias=False)
+        self.v22 = nn.Linear(flatten_shape, config.hidden_size // 2, bias=False)
+
+        self.b1 = nn.Parameter(torch.zeros(config.moe_experts, config.hidden_size // 2))
+        self.b2 = nn.Parameter(torch.zeros(config.moe_experts, config.hidden_size // 2))
+
+    def forward(
+        self, x: torch.Tensor, g1: torch.Tensor, g2: torch.Tensor
+    ) -> torch.Tensor:
+        g1, g2 = g1.type_as(x), g2.type_as(x)
+        x1 = self.act_fn(self.u1(x).unflatten(-1, (self.config.moe_experts, -1)))
+        x2 = self.act_fn(self.u2(x).unflatten(-1, (self.config.moe_experts, -1)))
+
+        x11 = self.v11(torch.einsum("btim,bthi->btim", x1, g1).flatten(-2))
+        x12 = self.v12(torch.einsum("btjm,bthj,bthi->btim", x2, g2, g1).flatten(-2))
+        x13 = torch.einsum("bthi,id->btd", g1, self.b1.type_as(x))
+
+        x21 = self.v21(torch.einsum("btim,bthi,bthj->btjm", x1, g1, g2).flatten(-2))
+        x22 = self.v22(torch.einsum("btjm,bthj->btjm", x2, g2).flatten(-2))
+        x23 = torch.einsum("bthj,jd->btd", g2, self.b2.type_as(x))
+
+        return torch.cat((x11 + x12 + x13, x21 + x22 + x23), dim=-1)
+
+
+class MonetMoHDE(nn.Module):
+    def __init__(self, config: MonetConfig):
+        super().__init__()
+        self.config = config
+        self.act_fn = ACT2FN[config.hidden_act]
+        flatten_shape = config.moe_experts * config.moe_dim
+
+        self.u = nn.Linear(config.hidden_size, flatten_shape)
+        self.v = nn.Linear(flatten_shape, config.hidden_size, bias=False)
+        self.b = nn.Parameter(torch.zeros(config.moe_experts, config.hidden_size))
+
+    def forward(
+        self, x: torch.Tensor, g1: torch.Tensor, g2: torch.Tensor
+    ) -> torch.Tensor:
+        g1, g2 = g1.type_as(x), g2.type_as(x)
+        x = self.act_fn(self.u(x).unflatten(-1, (self.config.moe_experts, -1)))
+        x = self.v(torch.einsum("btim,bthi,bthj->btjm", x, g1, g2).flatten(-2))
+        return x + torch.einsum("bthj,jd->btd", g2, self.b)
+
+
+class MonetDecoderLayer(nn.Module):
+    def __init__(self, config: MonetConfig, layer_idx: int):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.self_attn = LLAMA_ATTENTION_CLASSES[config._attn_implementation](
+            config=config, layer_idx=layer_idx
+        )
+        self.input_layernorm = LlamaRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = LlamaRMSNorm(
+            config.hidden_size, eps=config.rms_norm_eps
+        )
+
+        if config.moe_decompose == "vertical":
+            self.moe = MonetMoVDE(config)
+        elif config.moe_decompose == "horizontal":
+            self.moe = MonetMoHDE(config)
+        if layer_idx % config.moe_groups == 0:
+            self.router = MonetRouter(config).requires_grad_(False)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_value: Cache | None = None,
+        previous_router_probs: tuple[torch.Tensor, torch.Tensor] | None = None,
+        output_attentions: bool | None = False,
+        use_cache: bool | None = False,
+        cache_position: torch.LongTensor | None = None,
+        **kwargs,
+    ) -> tuple[torch.FloatTensor, ...]:
+        residual = hidden_states
+
+        hidden_states = self.input_layernorm(hidden_states)
+
+        # Self Attention
+        hidden_states, self_attn_weights, present_key_value = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_value=past_key_value,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+            cache_position=cache_position,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        g1, g2 = (
+            self.router(hidden_states)
+            if hasattr(self, "router")
+            else previous_router_probs
+        )
+        hidden_states = self.moe(hidden_states, g1, g2)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (self_attn_weights,)
+
+        if use_cache:
+            outputs += (present_key_value,)
+
+        return outputs + ((g1, g2) if hasattr(self, "router") else None,)
+
+
+class MonetPreTrainedModel(PreTrainedModel):
+    config_class = MonetConfig
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["MonetDecoderLayer"]
+    _skip_keys_device_placement = ["past_key_values"]
+    _supports_flash_attn_2 = True
+    _supports_sdpa = True
+    _supports_cache_class = True
+    _supports_quantized_cache = 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 MonetModel(MonetPreTrainedModel):
+    def __init__(self, config: MonetConfig):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)  # noqa
+        self.layers = nn.ModuleList([MonetDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)])  # noqa
+        self.norm = LlamaRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.gradient_checkpointing = False
+
+        # 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: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_values: Cache | list[torch.FloatTensor] | None = None,
+        inputs_embeds: torch.FloatTensor | None = None,
+        use_cache: bool | None = None,
+        output_attentions: bool | None = None,
+        output_hidden_states: bool | None = None,
+        output_router_probs: bool | None = None,
+        return_dict: bool | None = None,
+        cache_position: torch.LongTensor | None = None,
+    ) -> tuple[torch.Tensor, ...] | MonetModelOutputWithPast:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions  # noqa
+        output_hidden_states = output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states  # noqa
+        output_router_probs = output_router_probs if output_router_probs is not None else self.config.output_router_probs  # noqa
+        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  # noqa
+
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time, and must specify either one")  # noqa
+
+        if self.gradient_checkpointing and self.training and use_cache:
+            logger.warning_once("`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`.")  # noqa
+            use_cache = False
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+
+        return_legacy_cache = False
+        if use_cache and not isinstance(past_key_values, Cache):  # kept for BC (non `Cache` `past_key_values` inputs)  # noqa
+            return_legacy_cache = True
+            past_key_values = DynamicCache.from_legacy_cache(past_key_values)
+            logger.warning_once(
+                "We detected that you are passing `past_key_values` as a tuple and this is deprecated and will be removed in v4.43. "  # noqa
+                "Please use an appropriate `Cache` class (https://huggingface.co/docs/transformers/v4.41.3/en/internal/generation_utils#transformers.Cache)"  # noqa
+            )
+
+        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)  # noqa
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+        causal_mask = self._update_causal_mask(attention_mask, inputs_embeds, cache_position, past_key_values, output_attentions)  # noqa
+
+        # embed positions
+        hidden_states = inputs_embeds
+
+        # decoder layers
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        all_router_probs = () if output_router_probs else None
+        previous_router_probs, next_decoder_cache = None, None
+
+        for decoder_layer in self.layers:
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+                layer_outputs = self._gradient_checkpointing_func(
+                    decoder_layer.__call__,
+                    hidden_states,
+                    causal_mask,
+                    position_ids,
+                    past_key_values,
+                    previous_router_probs,
+                    output_attentions,
+                    use_cache,
+                    cache_position,
+                )
+            else:
+                layer_outputs = decoder_layer(
+                    hidden_states,
+                    attention_mask=causal_mask,
+                    position_ids=position_ids,
+                    past_key_value=past_key_values,
+                    previous_router_probs=previous_router_probs,
+                    output_attentions=output_attentions,
+                    use_cache=use_cache,
+                    cache_position=cache_position,
+                )
+
+            hidden_states = layer_outputs[0]
+            if use_cache:
+                next_decoder_cache = layer_outputs[2 if output_attentions else 1]
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+            if output_router_probs:
+                all_router_probs += (layer_outputs[-1],)
+            previous_router_probs = (
+                layer_outputs[-1]
+                if layer_outputs[-1] is not None
+                else previous_router_probs
+            )
+
+        hidden_states = self.norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        next_cache = next_decoder_cache if use_cache else None
+        if return_legacy_cache:
+            next_cache = next_cache.to_legacy_cache()
+
+        if not return_dict:
+            return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns, all_router_probs] if v is not None)  # noqa
+        return MonetModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=next_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+            router_probs=all_router_probs,
+        )
+
+    def _update_causal_mask(
+        self,
+        attention_mask: torch.Tensor,
+        input_tensor: torch.Tensor,
+        cache_position: torch.Tensor,
+        past_key_values: Cache,
+        output_attentions: bool,
+    ):
+        if self.config._attn_implementation == "flash_attention_2":
+            if attention_mask is not None and 0.0 in attention_mask:
+                return attention_mask
+            return None
+
+        past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0  # noqa
+        using_static_cache = isinstance(past_key_values, StaticCache)
+
+        if self.config._attn_implementation == "sdpa" and not using_static_cache and not output_attentions:  # noqa
+            if AttentionMaskConverter._ignore_causal_mask_sdpa(
+                attention_mask,
+                inputs_embeds=input_tensor,
+                past_key_values_length=past_seen_tokens,
+                is_training=self.training,
+            ):
+                return None
+
+        dtype, device = input_tensor.dtype, input_tensor.device
+        min_dtype = torch.finfo(dtype).min
+        sequence_length = input_tensor.shape[1]
+        if using_static_cache:
+            target_length = past_key_values.get_max_length()
+        else:
+            target_length = (
+                attention_mask.shape[-1]
+                if isinstance(attention_mask, torch.Tensor)
+                else past_seen_tokens + sequence_length + 1
+            )
+
+        if attention_mask is not None and attention_mask.dim() == 4:
+            if attention_mask.max() != 0:
+                raise ValueError("Custom 4D attention mask should be passed in inverted form with max==0`")  # noqa
+            causal_mask = attention_mask
+        else:
+            causal_mask = torch.full(
+                (sequence_length, target_length), fill_value=min_dtype, dtype=dtype, device=device  # noqa
+            )
+            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)  # noqa
+            causal_mask = causal_mask[None, None, :, :].expand(input_tensor.shape[0], 1, -1, -1)  # noqa
+            if attention_mask is not None:
+                causal_mask = causal_mask.clone()  # copy to contiguous memory for in-place edit  # noqa
+                mask_length = attention_mask.shape[-1]
+                padding_mask = causal_mask[:, :, :, :mask_length] + attention_mask[:, None, None, :]  # noqa
+                padding_mask = padding_mask == 0
+                causal_mask[:, :, :, :mask_length] = causal_mask[:, :, :, :mask_length].masked_fill(padding_mask, min_dtype)  # noqa
+        if (
+            self.config._attn_implementation == "sdpa"
+            and attention_mask is not None
+            and attention_mask.device.type == "cuda"
+            and not output_attentions
+        ):
+            causal_mask = AttentionMaskConverter._unmask_unattended(causal_mask, min_dtype)  # noqa
+
+        return causal_mask
+
+
+class MonetForCausalLM(MonetPreTrainedModel):
+    _tied_weights_keys = ["lm_head.weight"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = MonetModel(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: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_values: Cache | list[torch.FloatTensor] | None = None,
+        inputs_embeds: torch.FloatTensor | None = None,
+        labels: torch.LongTensor | None = None,
+        use_cache: bool | None = None,
+        output_attentions: bool | None = None,
+        output_hidden_states: bool | None = None,
+        output_router_probs: bool | None = None,
+        return_dict: bool | None = None,
+        cache_position: torch.LongTensor | None = None,
+    ) -> tuple[torch.Tensor, ...] | MonetCausalLMOutputWithPast:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions  # noqa
+        output_hidden_states = output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states  # noqa
+        output_router_probs = output_router_probs if output_router_probs is not None else self.config.output_router_probs  # noqa
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict  # noqa
+
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs = self.model(
+            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,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            output_router_probs=output_router_probs,
+            return_dict=return_dict,
+            cache_position=cache_position,
+        )
+
+        hidden_states = outputs[0]
+        logits = self.lm_head(hidden_states)
+        logits = logits.float()
+
+        loss = None
+        if labels is not None:
+            # Shift so that tokens < n predict n
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = CrossEntropyLoss()
+            shift_logits = shift_logits.view(-1, self.config.vocab_size)
+            shift_labels = shift_labels.view(-1)
+            # Enable model parallelism
+            shift_labels = shift_labels.to(shift_logits.device)
+            loss = loss_fct(shift_logits, shift_labels)
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return MonetCausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+            router_probs=outputs.router_probs,
+        )
+
+    def prepare_inputs_for_generation(
+        self,
+        input_ids,
+        past_key_values=None,
+        attention_mask=None,
+        inputs_embeds=None,
+        cache_position=None,
+        use_cache=True,
+        **kwargs,
+    ):
+        past_length = 0
+        if past_key_values is not None:
+            past_length = cache_position[0] if cache_position is not None else past_key_values.get_seq_length()  # noqa
+            max_cache_length = (
+                torch.tensor(past_key_values.get_max_length(), device=input_ids.device)
+                if past_key_values.get_max_length() is not None
+                else None
+            )
+            cache_length = past_length if max_cache_length is None else torch.min(max_cache_length, past_length)  # noqa
+
+            # Keep only the unprocessed tokens:
+            if attention_mask is not None and attention_mask.shape[1] > input_ids.shape[1]:  # noqa
+                input_ids = input_ids[:, -(attention_mask.shape[1] - past_length) :]
+            # input_ids based on the past_length.
+            elif past_length < input_ids.shape[1]:
+                input_ids = input_ids[:, past_length:]
+
+            if (
+                max_cache_length is not None
+                and attention_mask is not None
+                and cache_length + input_ids.shape[1] > max_cache_length
+            ):
+                attention_mask = attention_mask[:, -max_cache_length:]
+
+        position_ids = kwargs.get("position_ids", None)
+        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] :]
+
+        if inputs_embeds is not None and past_length == 0:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {"input_ids": input_ids.contiguous()}
+
+        input_length = position_ids.shape[-1] if position_ids is not None else input_ids.shape[-1]  # noqa
+        if cache_position is None:
+            cache_position = torch.arange(past_length, past_length + input_length, device=input_ids.device)  # noqa
+        elif use_cache:
+            cache_position = cache_position[-input_length:]
+
+        model_inputs.update(
+            {
+                "position_ids": position_ids,
+                "cache_position": cache_position,
+                "past_key_values": past_key_values,
+                "use_cache": use_cache,
+                "attention_mask": attention_mask,
+            }
+        )
+        return model_inputs
+
+    @staticmethod
+    def _reorder_cache(past_key_values, beam_idx):
+        reordered_past = ()
+        for layer_past in past_key_values:
+            reordered_past += (
+                tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past),  # noqa
+            )
+        return reordered_past

special_tokens_map.json

@@ -0,0 +1,23 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723

tokenizer_config.json

@@ -0,0 +1,43 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": true,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "legacy": false,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": null,
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}