Upload LlamaForCausalLM

model-00001-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:200fdcf95a438c7e3851f61708b36a36c458a9426f64128e5579754ace45af2a
-size 7961422720
+oid sha256:a39ce343bee15ce2b9ab9bdbb0e1131ba169f5e3cbacaf3f90011448bd73e717
+size 6995733923

model-00002-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8b5b95a75cc97edd13b7e1f1841e3caca4c7e248fa8bf433c05094dc25df7c90
-size 7983128936
+oid sha256:76fa77cd165a1fbf2512f87622d211c853bf5c4531b52d2d147c3cbeed9fe67d
+size 6934102407

model-00003-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3179e4e250a57a9a1ed1758f3a2b5dacf270e8b1bfe017b9bde5d88e32caeb94
-size 3159458205
+oid sha256:f5991f2a8cb1f83db70ad7e2a11f8766caebf541c881cc144536e3cd74517f1a
+size 5174173555

model.safetensors.index.json

@@ -621,38 +621,38 @@
     "model.layers.20.self_attn.v_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
     "model.layers.20.self_attn.v_proj.weight.quant_map": "model-00001-of-00003.safetensors",
     "model.layers.20.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.21.input_layernorm.weight": "model-00001-of-00003.safetensors",
-    "model.layers.21.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.21.mlp.down_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.21.mlp.down_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.21.mlp.down_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.21.mlp.down_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.21.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.21.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.21.mlp.gate_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.21.mlp.gate_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.21.mlp.gate_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.21.mlp.gate_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.21.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.21.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.21.mlp.up_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.21.mlp.up_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.21.mlp.up_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.21.mlp.up_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.21.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.21.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
+    "model.layers.21.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.layers.21.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.21.mlp.down_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.21.mlp.down_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.21.mlp.down_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.21.mlp.down_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.21.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.21.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.21.mlp.gate_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.21.mlp.gate_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.21.mlp.gate_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.21.mlp.gate_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.21.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.21.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.21.mlp.up_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.21.mlp.up_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.21.mlp.up_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.21.mlp.up_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.21.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.21.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
     "model.layers.21.self_attn.k_proj.weight": "model-00001-of-00003.safetensors",
     "model.layers.21.self_attn.k_proj.weight.absmax": "model-00001-of-00003.safetensors",
     "model.layers.21.self_attn.k_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
     "model.layers.21.self_attn.k_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
     "model.layers.21.self_attn.k_proj.weight.quant_map": "model-00001-of-00003.safetensors",
     "model.layers.21.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.21.self_attn.o_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.21.self_attn.o_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.21.self_attn.o_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.21.self_attn.o_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.21.self_attn.o_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.21.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
+    "model.layers.21.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.21.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.21.self_attn.o_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.21.self_attn.o_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.21.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.21.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
     "model.layers.21.self_attn.q_proj.weight": "model-00001-of-00003.safetensors",
     "model.layers.21.self_attn.q_proj.weight.absmax": "model-00001-of-00003.safetensors",
     "model.layers.21.self_attn.q_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
@@ -665,94 +665,94 @@
     "model.layers.21.self_attn.v_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
     "model.layers.21.self_attn.v_proj.weight.quant_map": "model-00001-of-00003.safetensors",
     "model.layers.21.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.22.input_layernorm.weight": "model-00001-of-00003.safetensors",
-    "model.layers.22.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.22.mlp.down_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.22.mlp.down_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.22.mlp.down_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.22.mlp.down_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.22.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.22.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.22.mlp.gate_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.22.mlp.gate_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.22.mlp.gate_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.22.mlp.gate_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.22.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.22.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.22.mlp.up_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.22.mlp.up_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.22.mlp.up_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.22.mlp.up_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.22.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.22.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.k_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.k_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.k_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.k_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.k_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.o_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.o_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.o_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.o_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.o_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.q_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.q_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.q_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.q_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.q_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.v_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.v_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.v_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.v_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.v_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.22.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.23.input_layernorm.weight": "model-00001-of-00003.safetensors",
-    "model.layers.23.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.23.mlp.down_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.23.mlp.down_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.23.mlp.down_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.23.mlp.down_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.23.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.23.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.23.mlp.gate_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.23.mlp.gate_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.23.mlp.gate_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.23.mlp.gate_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.23.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.23.mlp.up_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.23.mlp.up_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.23.mlp.up_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.23.mlp.up_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.23.mlp.up_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.23.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.23.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.k_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.k_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.k_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.k_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.k_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.o_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.o_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.o_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.o_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.o_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.q_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.q_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.q_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.q_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.q_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.v_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.v_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.v_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.v_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.v_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.23.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
+    "model.layers.22.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.layers.22.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.22.mlp.down_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.22.mlp.down_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.22.mlp.down_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.22.mlp.down_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.22.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.22.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.22.mlp.gate_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.22.mlp.gate_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.22.mlp.gate_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.22.mlp.gate_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.22.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.22.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.22.mlp.up_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.22.mlp.up_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.22.mlp.up_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.22.mlp.up_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.22.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.22.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.k_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.k_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.o_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.o_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.q_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.q_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.v_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.v_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.22.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.23.input_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.layers.23.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.23.mlp.down_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.23.mlp.down_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.23.mlp.down_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.23.mlp.down_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.23.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.23.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.23.mlp.gate_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.23.mlp.gate_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.23.mlp.gate_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.23.mlp.gate_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.23.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.23.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.23.mlp.up_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.23.mlp.up_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.23.mlp.up_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.23.mlp.up_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.23.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.23.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.k_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.k_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.o_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.o_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.q_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.q_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.v_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.v_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.23.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
     "model.layers.24.input_layernorm.weight": "model-00002-of-00003.safetensors",
     "model.layers.24.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
     "model.layers.24.mlp.down_proj.weight.absmax": "model-00002-of-00003.safetensors",
@@ -760,12 +760,12 @@
     "model.layers.24.mlp.down_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
     "model.layers.24.mlp.down_proj.weight.quant_map": "model-00002-of-00003.safetensors",
     "model.layers.24.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.24.mlp.gate_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.24.mlp.gate_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.24.mlp.gate_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.24.mlp.gate_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.24.mlp.gate_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.24.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
+    "model.layers.24.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.24.mlp.gate_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.24.mlp.gate_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.24.mlp.gate_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.24.mlp.gate_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.24.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
     "model.layers.24.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
     "model.layers.24.mlp.up_proj.weight.absmax": "model-00002-of-00003.safetensors",
     "model.layers.24.mlp.up_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
@@ -773,30 +773,30 @@
     "model.layers.24.mlp.up_proj.weight.quant_map": "model-00002-of-00003.safetensors",
     "model.layers.24.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
     "model.layers.24.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
-    "model.layers.24.self_attn.k_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.k_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.k_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.k_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.k_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.o_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.o_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.o_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.o_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.o_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.q_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.q_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.q_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.q_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.q_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.v_proj.weight": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.v_proj.weight.absmax": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.v_proj.weight.nested_absmax": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.v_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.v_proj.weight.quant_map": "model-00001-of-00003.safetensors",
-    "model.layers.24.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
+    "model.layers.24.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.k_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.k_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.o_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.o_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.q_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.q_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.v_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.v_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors",
+    "model.layers.24.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
     "model.layers.25.input_layernorm.weight": "model-00002-of-00003.safetensors",
     "model.layers.25.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
     "model.layers.25.mlp.down_proj.weight.absmax": "model-00002-of-00003.safetensors",
@@ -1765,26 +1765,26 @@
     "model.layers.44.self_attn.v_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
     "model.layers.44.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors",
     "model.layers.44.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.45.input_layernorm.weight": "model-00002-of-00003.safetensors",
-    "model.layers.45.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.45.mlp.down_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.45.mlp.down_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.45.mlp.down_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.45.mlp.down_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.45.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.45.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.45.mlp.gate_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.45.mlp.gate_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.45.mlp.gate_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.45.mlp.gate_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.45.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.45.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.45.mlp.up_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.45.mlp.up_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.45.mlp.up_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.45.mlp.up_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.45.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.45.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
+    "model.layers.45.input_layernorm.weight": "model-00003-of-00003.safetensors",
+    "model.layers.45.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.45.mlp.down_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.45.mlp.down_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.45.mlp.down_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.45.mlp.down_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.45.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.45.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.45.mlp.gate_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.45.mlp.gate_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.45.mlp.gate_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.45.mlp.gate_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.45.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.45.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.45.mlp.up_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.45.mlp.up_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.45.mlp.up_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.45.mlp.up_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.45.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.45.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
     "model.layers.45.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
     "model.layers.45.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors",
     "model.layers.45.self_attn.k_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
@@ -1809,182 +1809,182 @@
     "model.layers.45.self_attn.v_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
     "model.layers.45.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors",
     "model.layers.45.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.46.input_layernorm.weight": "model-00002-of-00003.safetensors",
-    "model.layers.46.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.46.mlp.down_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.46.mlp.down_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.46.mlp.down_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.46.mlp.down_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.46.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.46.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.46.mlp.gate_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.46.mlp.gate_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.46.mlp.gate_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.46.mlp.gate_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.46.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.46.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.46.mlp.up_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.46.mlp.up_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.46.mlp.up_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.46.mlp.up_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.46.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.46.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.k_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.k_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.o_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.o_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.q_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.q_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.v_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.v_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.46.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.47.input_layernorm.weight": "model-00002-of-00003.safetensors",
-    "model.layers.47.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.47.mlp.down_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.47.mlp.down_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.47.mlp.down_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.47.mlp.down_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.47.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.47.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.47.mlp.gate_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.47.mlp.gate_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.47.mlp.gate_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.47.mlp.gate_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.47.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.47.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.47.mlp.up_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.47.mlp.up_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.47.mlp.up_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.47.mlp.up_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.47.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.47.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.k_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.k_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.o_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.o_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.q_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.q_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.v_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.v_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.47.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.48.input_layernorm.weight": "model-00002-of-00003.safetensors",
-    "model.layers.48.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.48.mlp.down_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.48.mlp.down_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.48.mlp.down_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.48.mlp.down_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.48.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.48.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.48.mlp.gate_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.48.mlp.gate_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.48.mlp.gate_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.48.mlp.gate_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.48.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.48.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.48.mlp.up_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.48.mlp.up_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.48.mlp.up_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.48.mlp.up_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.48.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.48.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.k_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.k_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.o_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.o_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.q_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.q_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.v_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.v_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.48.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.49.input_layernorm.weight": "model-00002-of-00003.safetensors",
-    "model.layers.49.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.49.mlp.down_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.49.mlp.down_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.49.mlp.down_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.49.mlp.down_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.49.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.49.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.49.mlp.gate_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.49.mlp.gate_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.49.mlp.gate_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.49.mlp.gate_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.49.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.49.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.49.mlp.up_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.49.mlp.up_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.49.mlp.up_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.49.mlp.up_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.49.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.49.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.k_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.k_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.o_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.o_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.q_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.q_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.v_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.v_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.49.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.46.input_layernorm.weight": "model-00003-of-00003.safetensors",
+    "model.layers.46.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.46.mlp.down_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.46.mlp.down_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.46.mlp.down_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.46.mlp.down_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.46.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.46.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.46.mlp.gate_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.46.mlp.gate_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.46.mlp.gate_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.46.mlp.gate_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.46.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.46.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.46.mlp.up_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.46.mlp.up_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.46.mlp.up_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.46.mlp.up_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.46.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.46.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.k_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.k_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.k_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.k_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.k_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.o_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.o_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.o_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.o_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.o_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.q_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.q_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.q_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.q_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.q_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.v_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.v_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.v_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.v_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.v_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.46.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.47.input_layernorm.weight": "model-00003-of-00003.safetensors",
+    "model.layers.47.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.47.mlp.down_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.47.mlp.down_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.47.mlp.down_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.47.mlp.down_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.47.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.47.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.47.mlp.gate_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.47.mlp.gate_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.47.mlp.gate_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.47.mlp.gate_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.47.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.47.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.47.mlp.up_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.47.mlp.up_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.47.mlp.up_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.47.mlp.up_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.47.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.47.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.k_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.k_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.k_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.k_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.k_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.o_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.o_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.o_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.o_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.o_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.q_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.q_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.q_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.q_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.q_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.v_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.v_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.v_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.v_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.v_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.47.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.48.input_layernorm.weight": "model-00003-of-00003.safetensors",
+    "model.layers.48.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.48.mlp.down_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.48.mlp.down_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.48.mlp.down_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.48.mlp.down_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.48.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.48.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.48.mlp.gate_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.48.mlp.gate_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.48.mlp.gate_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.48.mlp.gate_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.48.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.48.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.48.mlp.up_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.48.mlp.up_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.48.mlp.up_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.48.mlp.up_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.48.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.48.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.k_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.k_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.k_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.k_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.k_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.o_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.o_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.o_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.o_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.o_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.q_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.q_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.q_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.q_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.q_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.v_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.v_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.v_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.v_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.v_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.48.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.49.input_layernorm.weight": "model-00003-of-00003.safetensors",
+    "model.layers.49.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.49.mlp.down_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.49.mlp.down_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.49.mlp.down_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.49.mlp.down_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.49.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.49.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.49.mlp.gate_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.49.mlp.gate_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.49.mlp.gate_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.49.mlp.gate_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.49.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.49.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.49.mlp.up_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.49.mlp.up_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.49.mlp.up_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.49.mlp.up_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.49.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.49.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.k_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.k_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.k_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.k_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.k_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.o_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.o_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.o_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.o_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.o_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.q_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.q_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.q_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.q_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.q_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.v_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.v_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.v_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.v_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.v_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.49.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
     "model.layers.5.input_layernorm.weight": "model-00001-of-00003.safetensors",
     "model.layers.5.mlp.down_proj.weight": "model-00001-of-00003.safetensors",
     "model.layers.5.mlp.down_proj.weight.absmax": "model-00001-of-00003.safetensors",
@@ -2029,94 +2029,94 @@
     "model.layers.5.self_attn.v_proj.weight.nested_quant_map": "model-00001-of-00003.safetensors",
     "model.layers.5.self_attn.v_proj.weight.quant_map": "model-00001-of-00003.safetensors",
     "model.layers.5.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00001-of-00003.safetensors",
-    "model.layers.50.input_layernorm.weight": "model-00002-of-00003.safetensors",
-    "model.layers.50.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.50.mlp.down_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.50.mlp.down_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.50.mlp.down_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.50.mlp.down_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.50.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.50.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.50.mlp.gate_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.50.mlp.gate_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.50.mlp.gate_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.50.mlp.gate_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.50.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.50.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.50.mlp.up_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.50.mlp.up_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.50.mlp.up_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.50.mlp.up_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.50.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.50.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.k_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.k_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.o_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.o_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.q_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.q_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.v_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.v_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.50.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.51.input_layernorm.weight": "model-00002-of-00003.safetensors",
-    "model.layers.51.mlp.down_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.51.mlp.down_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.51.mlp.down_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.51.mlp.down_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.51.mlp.down_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.51.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.51.mlp.gate_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.51.mlp.gate_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.51.mlp.gate_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.51.mlp.gate_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.51.mlp.gate_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.51.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.51.mlp.up_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.51.mlp.up_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.51.mlp.up_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.51.mlp.up_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.51.mlp.up_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.51.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.51.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.k_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.k_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.o_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.o_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.q_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.q_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.v_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.v_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.51.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.50.input_layernorm.weight": "model-00003-of-00003.safetensors",
+    "model.layers.50.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.50.mlp.down_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.50.mlp.down_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.50.mlp.down_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.50.mlp.down_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.50.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.50.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.50.mlp.gate_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.50.mlp.gate_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.50.mlp.gate_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.50.mlp.gate_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.50.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.50.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.50.mlp.up_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.50.mlp.up_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.50.mlp.up_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.50.mlp.up_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.50.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.50.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.k_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.k_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.k_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.k_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.k_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.o_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.o_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.o_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.o_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.o_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.q_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.q_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.q_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.q_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.q_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.v_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.v_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.v_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.v_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.v_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.50.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.51.input_layernorm.weight": "model-00003-of-00003.safetensors",
+    "model.layers.51.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.51.mlp.down_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.51.mlp.down_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.51.mlp.down_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.51.mlp.down_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.51.mlp.down_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.51.mlp.gate_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.51.mlp.gate_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.51.mlp.gate_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.51.mlp.gate_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.51.mlp.gate_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.51.mlp.gate_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.51.mlp.up_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.51.mlp.up_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.51.mlp.up_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.51.mlp.up_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.51.mlp.up_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.51.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.51.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.k_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.k_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.k_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.k_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.k_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.o_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.o_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.o_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.o_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.o_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.q_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.q_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.q_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.q_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.q_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.v_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.v_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.v_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.v_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.v_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.51.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
     "model.layers.52.input_layernorm.weight": "model-00003-of-00003.safetensors",
     "model.layers.52.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
     "model.layers.52.mlp.down_proj.weight.absmax": "model-00003-of-00003.safetensors",
@@ -2137,30 +2137,30 @@
     "model.layers.52.mlp.up_proj.weight.quant_map": "model-00003-of-00003.safetensors",
     "model.layers.52.mlp.up_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
     "model.layers.52.post_attention_layernorm.weight": "model-00003-of-00003.safetensors",
-    "model.layers.52.self_attn.k_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.k_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.k_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.k_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.k_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.o_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.o_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.o_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.o_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.o_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.q_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.q_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.q_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.q_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.q_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.v_proj.weight": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.v_proj.weight.absmax": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.v_proj.weight.nested_absmax": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.v_proj.weight.nested_quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.v_proj.weight.quant_map": "model-00002-of-00003.safetensors",
-    "model.layers.52.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00002-of-00003.safetensors",
+    "model.layers.52.self_attn.k_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.k_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.k_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.k_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.k_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.k_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.o_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.o_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.o_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.o_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.o_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.o_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.q_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.q_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.q_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.q_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.q_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.q_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.v_proj.weight": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.v_proj.weight.absmax": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.v_proj.weight.nested_absmax": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.v_proj.weight.nested_quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.v_proj.weight.quant_map": "model-00003-of-00003.safetensors",
+    "model.layers.52.self_attn.v_proj.weight.quant_state.bitsandbytes__nf4": "model-00003-of-00003.safetensors",
     "model.layers.53.input_layernorm.weight": "model-00003-of-00003.safetensors",
     "model.layers.53.mlp.down_proj.weight": "model-00003-of-00003.safetensors",
     "model.layers.53.mlp.down_proj.weight.absmax": "model-00003-of-00003.safetensors",