Upload 13 files

config.json

@@ -0,0 +1,36 @@
+{
+  "architectures": [
+    "MambaForCausalLM"
+  ],
+  "bos_token_id": 0,
+  "conv_kernel": 4,
+  "eos_token_id": 0,
+  "expand": 2,
+  "fused_add_norm": true,
+  "hidden_act": "silu",
+  "hidden_size": 2560,
+  "initializer_range": 0.1,
+  "intermediate_size": 5120,
+  "layer_norm_epsilon": 1e-05,
+  "model_type": "mamba",
+  "n_layer": 48,
+  "num_hidden_layers": 48,
+  "pad_token_id": 0,
+  "pad_vocab_size_multiple": 8,
+  "rescale_prenorm_residual": false,
+  "residual_in_fp32": true,
+  "rms_norm": true,
+  "state_size": 16,
+  "time_step_floor": 0.0001,
+  "time_step_init_scheme": "random",
+  "time_step_max": 0.1,
+  "time_step_min": 0.001,
+  "time_step_rank": 160,
+  "time_step_scale": 1.0,
+  "torch_dtype": "float32",
+  "transformers_version": "4.40.2",
+  "use_bias": false,
+  "use_cache": true,
+  "use_conv_bias": true,
+  "vocab_size": 50280
+}

generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 0,
+  "eos_token_id": 0,
+  "pad_token_id": 0,
+  "transformers_version": "4.40.2"
+}

optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:844f090d057ca19f33d80a04c419d96920e7da75303a92ffccd8046918d88c52
+size 16867928192

pytorch_model-00001-of-00002.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:360ac36f33cb18b09aeb41e6d7e95100bbe1b07e97ba6f467e191857dc67268e
+size 4969794150

pytorch_model-00002-of-00002.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c93b140c1bc4a0bf90d45cb9e238b1b74b98d842bb016d8c5d31036605252c1
+size 3464130832

pytorch_model.bin.index.json

@@ -0,0 +1,490 @@
+{
+  "metadata": {
+    "total_size": 8433756160
+  },
+  "weight_map": {
+    "backbone.embeddings.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.0.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.0.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.0.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.0.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.0.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.0.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.0.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.0.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.0.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.0.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.1.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.1.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.1.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.1.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.1.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.1.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.1.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.1.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.1.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.1.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.10.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.10.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.10.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.10.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.10.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.10.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.10.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.10.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.10.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.10.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.11.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.11.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.11.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.11.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.11.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.11.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.11.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.11.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.11.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.11.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.12.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.12.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.12.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.12.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.12.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.12.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.12.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.12.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.12.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.12.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.13.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.13.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.13.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.13.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.13.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.13.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.13.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.13.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.13.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.13.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.14.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.14.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.14.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.14.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.14.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.14.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.14.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.14.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.14.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.14.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.15.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.15.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.15.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.15.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.15.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.15.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.15.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.15.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.15.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.15.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.16.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.16.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.16.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.16.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.16.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.16.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.16.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.16.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.16.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.16.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.17.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.17.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.17.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.17.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.17.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.17.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.17.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.17.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.17.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.17.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.18.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.18.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.18.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.18.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.18.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.18.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.18.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.18.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.18.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.18.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.19.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.19.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.19.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.19.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.19.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.19.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.19.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.19.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.19.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.19.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.2.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.2.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.2.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.2.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.2.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.2.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.2.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.2.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.2.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.2.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.20.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.20.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.20.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.20.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.20.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.20.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.20.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.20.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.20.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.20.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.21.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.21.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.21.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.21.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.21.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.21.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.21.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.21.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.21.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.21.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.22.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.22.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.22.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.22.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.22.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.22.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.22.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.22.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.22.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.22.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.23.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.23.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.23.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.23.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.23.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.23.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.23.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.23.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.23.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.23.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.24.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.24.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.24.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.24.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.24.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.24.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.24.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.24.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.24.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.24.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.25.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.25.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.25.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.25.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.25.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.25.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.25.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.25.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.25.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.25.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.26.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.26.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.26.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.26.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.26.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.26.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.26.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.26.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.26.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.26.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.27.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.27.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.27.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.27.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.27.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.27.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.27.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.27.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.27.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.27.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.28.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.28.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.28.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.28.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.28.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.28.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.28.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.28.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.28.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.28.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.29.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.29.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.29.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.29.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.29.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.29.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.29.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.29.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.29.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.29.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.3.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.3.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.3.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.3.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.3.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.3.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.3.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.3.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.3.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.3.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.30.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.30.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.30.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.30.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.30.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.30.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.30.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.30.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.30.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.30.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.31.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.31.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.31.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.31.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.31.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.31.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.31.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.31.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.31.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.31.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.32.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.32.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.32.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.32.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.32.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.32.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.32.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.32.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.32.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.32.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.33.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.33.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.33.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.33.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.33.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.33.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.33.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.33.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.33.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.33.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.34.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.34.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.34.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.34.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.34.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.34.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.34.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.34.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.34.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.34.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.35.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.35.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.35.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.35.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.35.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.35.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.35.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.35.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.35.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.35.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.36.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.36.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.36.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.36.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.36.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.36.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.36.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.36.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.36.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.36.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.37.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.37.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.37.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.37.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.37.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.37.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.37.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.37.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.37.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.37.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.38.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.38.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.38.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.38.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.38.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.38.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.38.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.38.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.38.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.38.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.39.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.39.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.39.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.39.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.39.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.39.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.39.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.39.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.39.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.39.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.4.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.4.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.4.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.4.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.4.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.4.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.4.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.4.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.4.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.4.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.40.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.40.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.40.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.40.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.40.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.40.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.40.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.40.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.40.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.40.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.41.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.41.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.41.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.41.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.41.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.41.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.41.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.41.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.41.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.41.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.42.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.42.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.42.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.42.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.42.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.42.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.42.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.42.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.42.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.42.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.43.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.43.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.43.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.43.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.43.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.43.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.43.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.43.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.43.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.43.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.44.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.44.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.44.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.44.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.44.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.44.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.44.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.44.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.44.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.44.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.45.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.45.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.45.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.45.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.45.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.45.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.45.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.45.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.45.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.45.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.46.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.46.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.46.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.46.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.46.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.46.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.46.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.46.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.46.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.46.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.47.mixer.A_log": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.47.mixer.D": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.47.mixer.conv1d.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.47.mixer.conv1d.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.47.mixer.dt_proj.bias": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.47.mixer.dt_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.47.mixer.in_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.47.mixer.out_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.47.mixer.x_proj.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.47.norm.weight": "pytorch_model-00002-of-00002.bin",
+    "backbone.layers.5.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.5.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.5.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.5.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.5.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.5.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.5.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.5.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.5.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.5.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.6.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.6.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.6.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.6.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.6.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.6.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.6.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.6.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.6.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.6.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.7.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.7.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.7.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.7.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.7.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.7.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.7.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.7.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.7.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.7.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.8.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.8.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.8.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.8.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.8.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.8.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.8.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.8.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.8.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.8.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.9.mixer.A_log": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.9.mixer.D": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.9.mixer.conv1d.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.9.mixer.conv1d.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.9.mixer.dt_proj.bias": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.9.mixer.dt_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.9.mixer.in_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.9.mixer.out_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.9.mixer.x_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.layers.9.norm.weight": "pytorch_model-00001-of-00002.bin",
+    "backbone.norm_f.weight": "pytorch_model-00002-of-00002.bin",
+    "lm_head.weight": "pytorch_model-00001-of-00002.bin"
+  }
+}

rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:386fcc8cc1089aade9450d86fb239ea3483f455fd2d78d8378645feecfec9d69
+size 14244

scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f346a59ff32f65420182867153c8c6a00a1a74b1c2971b280578bc6d6be760ed
+size 1064

special_tokens_map.json

@@ -0,0 +1,30 @@
+{
+  "bos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer_config.json

@@ -0,0 +1,214 @@
+{
+  "add_bos_token": false,
+  "add_eos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<|padding|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "50254": {
+      "content": "                        ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50255": {
+      "content": "                       ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50256": {
+      "content": "                      ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50257": {
+      "content": "                     ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50258": {
+      "content": "                    ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50259": {
+      "content": "                   ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50260": {
+      "content": "                  ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50261": {
+      "content": "                 ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50262": {
+      "content": "                ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50263": {
+      "content": "               ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50264": {
+      "content": "              ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50265": {
+      "content": "             ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50266": {
+      "content": "            ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50267": {
+      "content": "           ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50268": {
+      "content": "          ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50269": {
+      "content": "         ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50270": {
+      "content": "        ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50271": {
+      "content": "       ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50272": {
+      "content": "      ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50273": {
+      "content": "     ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50274": {
+      "content": "    ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50275": {
+      "content": "   ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50276": {
+      "content": "  ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "bos_token": "<|endoftext|>",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "<|endoftext|>",
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "<|endoftext|>",
+  "tokenizer_class": "GPTNeoXTokenizer",
+  "unk_token": "<|endoftext|>"
+}

trainer_state.json

@@ -0,0 +1,3150 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.10921169025567863,
+  "eval_steps": 100000000,
+  "global_step": 13400,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 8.15012613848348e-06,
+      "grad_norm": 43.29102325439453,
+      "learning_rate": 1.0000000000000001e-07,
+      "loss": 81.0533,
+      "step": 1
+    },
+    {
+      "epoch": 0.0002445037841545044,
+      "grad_norm": 42.03493881225586,
+      "learning_rate": 3e-06,
+      "loss": 70.3325,
+      "step": 30
+    },
+    {
+      "epoch": 0.0004890075683090088,
+      "grad_norm": 10.497812271118164,
+      "learning_rate": 6e-06,
+      "loss": 23.8844,
+      "step": 60
+    },
+    {
+      "epoch": 0.0007335113524635132,
+      "grad_norm": 6.692718029022217,
+      "learning_rate": 9e-06,
+      "loss": 11.7198,
+      "step": 90
+    },
+    {
+      "epoch": 0.0009780151366180175,
+      "grad_norm": 9.11446475982666,
+      "learning_rate": 1.2e-05,
+      "loss": 10.3199,
+      "step": 120
+    },
+    {
+      "epoch": 0.001222518920772522,
+      "grad_norm": 16.92243766784668,
+      "learning_rate": 1.5e-05,
+      "loss": 9.5338,
+      "step": 150
+    },
+    {
+      "epoch": 0.0014670227049270264,
+      "grad_norm": 12.929216384887695,
+      "learning_rate": 1.8e-05,
+      "loss": 8.9114,
+      "step": 180
+    },
+    {
+      "epoch": 0.0017115264890815308,
+      "grad_norm": 8.830116271972656,
+      "learning_rate": 2.1e-05,
+      "loss": 8.3867,
+      "step": 210
+    },
+    {
+      "epoch": 0.001956030273236035,
+      "grad_norm": 7.348124980926514,
+      "learning_rate": 2.4e-05,
+      "loss": 8.0113,
+      "step": 240
+    },
+    {
+      "epoch": 0.0022005340573905395,
+      "grad_norm": 12.751787185668945,
+      "learning_rate": 2.7000000000000002e-05,
+      "loss": 7.6093,
+      "step": 270
+    },
+    {
+      "epoch": 0.002445037841545044,
+      "grad_norm": 15.10444164276123,
+      "learning_rate": 3e-05,
+      "loss": 7.3484,
+      "step": 300
+    },
+    {
+      "epoch": 0.0026895416256995483,
+      "grad_norm": 6.653383731842041,
+      "learning_rate": 3.3e-05,
+      "loss": 7.2105,
+      "step": 330
+    },
+    {
+      "epoch": 0.0029340454098540527,
+      "grad_norm": 6.986039161682129,
+      "learning_rate": 3.6e-05,
+      "loss": 7.0228,
+      "step": 360
+    },
+    {
+      "epoch": 0.003178549194008557,
+      "grad_norm": 6.230088710784912,
+      "learning_rate": 3.9000000000000006e-05,
+      "loss": 6.8948,
+      "step": 390
+    },
+    {
+      "epoch": 0.0034230529781630616,
+      "grad_norm": 8.170981407165527,
+      "learning_rate": 4.2e-05,
+      "loss": 6.6965,
+      "step": 420
+    },
+    {
+      "epoch": 0.0036675567623175656,
+      "grad_norm": 5.6345930099487305,
+      "learning_rate": 4.5e-05,
+      "loss": 6.5988,
+      "step": 450
+    },
+    {
+      "epoch": 0.00391206054647207,
+      "grad_norm": 5.156513214111328,
+      "learning_rate": 4.8e-05,
+      "loss": 6.4795,
+      "step": 480
+    },
+    {
+      "epoch": 0.0041565643306265745,
+      "grad_norm": 5.4964189529418945,
+      "learning_rate": 4.999999990869806e-05,
+      "loss": 6.333,
+      "step": 510
+    },
+    {
+      "epoch": 0.004401068114781079,
+      "grad_norm": 3.67378306388855,
+      "learning_rate": 4.999999853916893e-05,
+      "loss": 6.2208,
+      "step": 540
+    },
+    {
+      "epoch": 0.004645571898935583,
+      "grad_norm": 8.507222175598145,
+      "learning_rate": 4.9999995526204936e-05,
+      "loss": 6.1097,
+      "step": 570
+    },
+    {
+      "epoch": 0.004890075683090088,
+      "grad_norm": 3.756618022918701,
+      "learning_rate": 4.999999086980628e-05,
+      "loss": 5.9886,
+      "step": 600
+    },
+    {
+      "epoch": 0.005134579467244592,
+      "grad_norm": 3.8149781227111816,
+      "learning_rate": 4.999998456997326e-05,
+      "loss": 5.8779,
+      "step": 630
+    },
+    {
+      "epoch": 0.005379083251399097,
+      "grad_norm": 3.840543270111084,
+      "learning_rate": 4.999997662670628e-05,
+      "loss": 5.805,
+      "step": 660
+    },
+    {
+      "epoch": 0.005623587035553601,
+      "grad_norm": 3.4208931922912598,
+      "learning_rate": 4.999996704000589e-05,
+      "loss": 5.6992,
+      "step": 690
+    },
+    {
+      "epoch": 0.0058680908197081055,
+      "grad_norm": 3.2975683212280273,
+      "learning_rate": 4.99999558098727e-05,
+      "loss": 5.6531,
+      "step": 720
+    },
+    {
+      "epoch": 0.00611259460386261,
+      "grad_norm": 4.05631160736084,
+      "learning_rate": 4.9999942936307445e-05,
+      "loss": 5.554,
+      "step": 750
+    },
+    {
+      "epoch": 0.006357098388017114,
+      "grad_norm": 3.1539864540100098,
+      "learning_rate": 4.9999928419310994e-05,
+      "loss": 5.4931,
+      "step": 780
+    },
+    {
+      "epoch": 0.006601602172171619,
+      "grad_norm": 4.811732292175293,
+      "learning_rate": 4.999991225888427e-05,
+      "loss": 5.4204,
+      "step": 810
+    },
+    {
+      "epoch": 0.006846105956326123,
+      "grad_norm": 2.9593210220336914,
+      "learning_rate": 4.999989445502837e-05,
+      "loss": 5.3687,
+      "step": 840
+    },
+    {
+      "epoch": 0.007090609740480627,
+      "grad_norm": 3.942239284515381,
+      "learning_rate": 4.9999875007744436e-05,
+      "loss": 5.3238,
+      "step": 870
+    },
+    {
+      "epoch": 0.007335113524635131,
+      "grad_norm": 2.29752254486084,
+      "learning_rate": 4.9999853917033756e-05,
+      "loss": 5.2423,
+      "step": 900
+    },
+    {
+      "epoch": 0.007579617308789636,
+      "grad_norm": 2.243770122528076,
+      "learning_rate": 4.999983118289773e-05,
+      "loss": 5.2384,
+      "step": 930
+    },
+    {
+      "epoch": 0.00782412109294414,
+      "grad_norm": 2.5572686195373535,
+      "learning_rate": 4.999980680533782e-05,
+      "loss": 5.1761,
+      "step": 960
+    },
+    {
+      "epoch": 0.008068624877098645,
+      "grad_norm": 2.4739913940429688,
+      "learning_rate": 4.999978078435567e-05,
+      "loss": 5.1215,
+      "step": 990
+    },
+    {
+      "epoch": 0.008313128661253149,
+      "grad_norm": 1.6927807331085205,
+      "learning_rate": 4.999975311995295e-05,
+      "loss": 5.1264,
+      "step": 1020
+    },
+    {
+      "epoch": 0.008557632445407654,
+      "grad_norm": 2.3166885375976562,
+      "learning_rate": 4.99997238121315e-05,
+      "loss": 5.0469,
+      "step": 1050
+    },
+    {
+      "epoch": 0.008802136229562158,
+      "grad_norm": 1.69430410861969,
+      "learning_rate": 4.999969286089325e-05,
+      "loss": 5.005,
+      "step": 1080
+    },
+    {
+      "epoch": 0.009046640013716663,
+      "grad_norm": 2.0859012603759766,
+      "learning_rate": 4.9999660266240235e-05,
+      "loss": 4.9693,
+      "step": 1110
+    },
+    {
+      "epoch": 0.009291143797871167,
+      "grad_norm": 1.7291619777679443,
+      "learning_rate": 4.9999626028174585e-05,
+      "loss": 4.9352,
+      "step": 1140
+    },
+    {
+      "epoch": 0.00953564758202567,
+      "grad_norm": 1.5966330766677856,
+      "learning_rate": 4.999959014669856e-05,
+      "loss": 4.87,
+      "step": 1170
+    },
+    {
+      "epoch": 0.009780151366180176,
+      "grad_norm": 1.9011143445968628,
+      "learning_rate": 4.9999552621814513e-05,
+      "loss": 4.8772,
+      "step": 1200
+    },
+    {
+      "epoch": 0.010024655150334679,
+      "grad_norm": 1.701004981994629,
+      "learning_rate": 4.9999513453524917e-05,
+      "loss": 4.8484,
+      "step": 1230
+    },
+    {
+      "epoch": 0.010269158934489184,
+      "grad_norm": 2.171808958053589,
+      "learning_rate": 4.9999472641832336e-05,
+      "loss": 4.8233,
+      "step": 1260
+    },
+    {
+      "epoch": 0.010513662718643688,
+      "grad_norm": 1.4896094799041748,
+      "learning_rate": 4.999943018673946e-05,
+      "loss": 4.7921,
+      "step": 1290
+    },
+    {
+      "epoch": 0.010758166502798193,
+      "grad_norm": 2.112971544265747,
+      "learning_rate": 4.999938608824909e-05,
+      "loss": 4.7908,
+      "step": 1320
+    },
+    {
+      "epoch": 0.011002670286952697,
+      "grad_norm": 1.7113062143325806,
+      "learning_rate": 4.999934034636411e-05,
+      "loss": 4.7186,
+      "step": 1350
+    },
+    {
+      "epoch": 0.011247174071107202,
+      "grad_norm": 2.099663019180298,
+      "learning_rate": 4.999929296108753e-05,
+      "loss": 4.7112,
+      "step": 1380
+    },
+    {
+      "epoch": 0.011491677855261706,
+      "grad_norm": 1.40205979347229,
+      "learning_rate": 4.9999243932422466e-05,
+      "loss": 4.6701,
+      "step": 1410
+    },
+    {
+      "epoch": 0.011736181639416211,
+      "grad_norm": 1.53098726272583,
+      "learning_rate": 4.999919326037215e-05,
+      "loss": 4.6786,
+      "step": 1440
+    },
+    {
+      "epoch": 0.011980685423570715,
+      "grad_norm": 1.7461609840393066,
+      "learning_rate": 4.99991409449399e-05,
+      "loss": 4.6614,
+      "step": 1470
+    },
+    {
+      "epoch": 0.01222518920772522,
+      "grad_norm": 1.6045819520950317,
+      "learning_rate": 4.999908698612916e-05,
+      "loss": 4.6233,
+      "step": 1500
+    },
+    {
+      "epoch": 0.012469692991879723,
+      "grad_norm": 1.351077675819397,
+      "learning_rate": 4.9999031383943486e-05,
+      "loss": 4.626,
+      "step": 1530
+    },
+    {
+      "epoch": 0.012714196776034229,
+      "grad_norm": 1.4363055229187012,
+      "learning_rate": 4.999897413838651e-05,
+      "loss": 4.5781,
+      "step": 1560
+    },
+    {
+      "epoch": 0.012958700560188732,
+      "grad_norm": 1.1321336030960083,
+      "learning_rate": 4.999891524946202e-05,
+      "loss": 4.5492,
+      "step": 1590
+    },
+    {
+      "epoch": 0.013203204344343238,
+      "grad_norm": 1.2755215167999268,
+      "learning_rate": 4.999885471717387e-05,
+      "loss": 4.5444,
+      "step": 1620
+    },
+    {
+      "epoch": 0.013447708128497741,
+      "grad_norm": 1.471504807472229,
+      "learning_rate": 4.999879254152605e-05,
+      "loss": 4.5247,
+      "step": 1650
+    },
+    {
+      "epoch": 0.013692211912652246,
+      "grad_norm": 1.3664900064468384,
+      "learning_rate": 4.999872872252265e-05,
+      "loss": 4.5158,
+      "step": 1680
+    },
+    {
+      "epoch": 0.01393671569680675,
+      "grad_norm": 1.360611915588379,
+      "learning_rate": 4.999866326016785e-05,
+      "loss": 4.498,
+      "step": 1710
+    },
+    {
+      "epoch": 0.014181219480961254,
+      "grad_norm": 1.43533456325531,
+      "learning_rate": 4.999859615446596e-05,
+      "loss": 4.5035,
+      "step": 1740
+    },
+    {
+      "epoch": 0.014425723265115759,
+      "grad_norm": 1.4639947414398193,
+      "learning_rate": 4.99985274054214e-05,
+      "loss": 4.4533,
+      "step": 1770
+    },
+    {
+      "epoch": 0.014670227049270262,
+      "grad_norm": 1.438585638999939,
+      "learning_rate": 4.999845701303868e-05,
+      "loss": 4.4484,
+      "step": 1800
+    },
+    {
+      "epoch": 0.014914730833424768,
+      "grad_norm": 1.1868066787719727,
+      "learning_rate": 4.999838497732243e-05,
+      "loss": 4.4516,
+      "step": 1830
+    },
+    {
+      "epoch": 0.015159234617579271,
+      "grad_norm": 1.1804672479629517,
+      "learning_rate": 4.999831129827739e-05,
+      "loss": 4.4378,
+      "step": 1860
+    },
+    {
+      "epoch": 0.015403738401733777,
+      "grad_norm": 1.127652645111084,
+      "learning_rate": 4.9998235975908394e-05,
+      "loss": 4.4,
+      "step": 1890
+    },
+    {
+      "epoch": 0.01564824218588828,
+      "grad_norm": 1.2617137432098389,
+      "learning_rate": 4.99981590102204e-05,
+      "loss": 4.3802,
+      "step": 1920
+    },
+    {
+      "epoch": 0.015892745970042785,
+      "grad_norm": 1.4249097108840942,
+      "learning_rate": 4.9998080401218464e-05,
+      "loss": 4.3689,
+      "step": 1950
+    },
+    {
+      "epoch": 0.01613724975419729,
+      "grad_norm": 1.0717036724090576,
+      "learning_rate": 4.999800014890777e-05,
+      "loss": 4.3645,
+      "step": 1980
+    },
+    {
+      "epoch": 0.016381753538351793,
+      "grad_norm": 1.414404273033142,
+      "learning_rate": 4.9997918253293555e-05,
+      "loss": 4.343,
+      "step": 2010
+    },
+    {
+      "epoch": 0.016626257322506298,
+      "grad_norm": 1.1941649913787842,
+      "learning_rate": 4.999783471438124e-05,
+      "loss": 4.3271,
+      "step": 2040
+    },
+    {
+      "epoch": 0.016870761106660803,
+      "grad_norm": 1.0737369060516357,
+      "learning_rate": 4.999774953217631e-05,
+      "loss": 4.3223,
+      "step": 2070
+    },
+    {
+      "epoch": 0.01711526489081531,
+      "grad_norm": 1.2367453575134277,
+      "learning_rate": 4.9997662706684345e-05,
+      "loss": 4.3106,
+      "step": 2100
+    },
+    {
+      "epoch": 0.01735976867496981,
+      "grad_norm": 1.0919674634933472,
+      "learning_rate": 4.999757423791107e-05,
+      "loss": 4.2916,
+      "step": 2130
+    },
+    {
+      "epoch": 0.017604272459124316,
+      "grad_norm": 1.2348527908325195,
+      "learning_rate": 4.9997484125862306e-05,
+      "loss": 4.2935,
+      "step": 2160
+    },
+    {
+      "epoch": 0.01784877624327882,
+      "grad_norm": 1.173093318939209,
+      "learning_rate": 4.999739237054395e-05,
+      "loss": 4.2761,
+      "step": 2190
+    },
+    {
+      "epoch": 0.018093280027433326,
+      "grad_norm": 1.1982988119125366,
+      "learning_rate": 4.9997298971962065e-05,
+      "loss": 4.2619,
+      "step": 2220
+    },
+    {
+      "epoch": 0.018337783811587828,
+      "grad_norm": 1.4401408433914185,
+      "learning_rate": 4.999720393012277e-05,
+      "loss": 4.2672,
+      "step": 2250
+    },
+    {
+      "epoch": 0.018582287595742333,
+      "grad_norm": 1.0311753749847412,
+      "learning_rate": 4.999710724503233e-05,
+      "loss": 4.2272,
+      "step": 2280
+    },
+    {
+      "epoch": 0.01882679137989684,
+      "grad_norm": 1.4242304563522339,
+      "learning_rate": 4.9997008916697075e-05,
+      "loss": 4.2415,
+      "step": 2310
+    },
+    {
+      "epoch": 0.01907129516405134,
+      "grad_norm": 1.039548397064209,
+      "learning_rate": 4.999690894512349e-05,
+      "loss": 4.2086,
+      "step": 2340
+    },
+    {
+      "epoch": 0.019315798948205846,
+      "grad_norm": 1.1213486194610596,
+      "learning_rate": 4.999680733031814e-05,
+      "loss": 4.2043,
+      "step": 2370
+    },
+    {
+      "epoch": 0.01956030273236035,
+      "grad_norm": 1.0591011047363281,
+      "learning_rate": 4.9996704072287716e-05,
+      "loss": 4.1894,
+      "step": 2400
+    },
+    {
+      "epoch": 0.019804806516514856,
+      "grad_norm": 1.0503829717636108,
+      "learning_rate": 4.9996599171038984e-05,
+      "loss": 4.184,
+      "step": 2430
+    },
+    {
+      "epoch": 0.020049310300669358,
+      "grad_norm": 1.068516492843628,
+      "learning_rate": 4.999649262657886e-05,
+      "loss": 4.1918,
+      "step": 2460
+    },
+    {
+      "epoch": 0.020293814084823864,
+      "grad_norm": 0.9620047211647034,
+      "learning_rate": 4.999638443891434e-05,
+      "loss": 4.182,
+      "step": 2490
+    },
+    {
+      "epoch": 0.02053831786897837,
+      "grad_norm": 0.9728343486785889,
+      "learning_rate": 4.999627460805253e-05,
+      "loss": 4.1746,
+      "step": 2520
+    },
+    {
+      "epoch": 0.020782821653132874,
+      "grad_norm": 0.990840494632721,
+      "learning_rate": 4.999616313400066e-05,
+      "loss": 4.1339,
+      "step": 2550
+    },
+    {
+      "epoch": 0.021027325437287376,
+      "grad_norm": 1.0678601264953613,
+      "learning_rate": 4.999605001676605e-05,
+      "loss": 4.1548,
+      "step": 2580
+    },
+    {
+      "epoch": 0.02127182922144188,
+      "grad_norm": 1.1837236881256104,
+      "learning_rate": 4.9995935256356144e-05,
+      "loss": 4.1508,
+      "step": 2610
+    },
+    {
+      "epoch": 0.021516333005596387,
+      "grad_norm": 1.0381749868392944,
+      "learning_rate": 4.9995818852778476e-05,
+      "loss": 4.0977,
+      "step": 2640
+    },
+    {
+      "epoch": 0.021760836789750892,
+      "grad_norm": 0.9649094343185425,
+      "learning_rate": 4.999570080604071e-05,
+      "loss": 4.1405,
+      "step": 2670
+    },
+    {
+      "epoch": 0.022005340573905394,
+      "grad_norm": 1.1858196258544922,
+      "learning_rate": 4.99955811161506e-05,
+      "loss": 4.1242,
+      "step": 2700
+    },
+    {
+      "epoch": 0.0222498443580599,
+      "grad_norm": 1.035134196281433,
+      "learning_rate": 4.9995459783116004e-05,
+      "loss": 4.1042,
+      "step": 2730
+    },
+    {
+      "epoch": 0.022494348142214404,
+      "grad_norm": 1.0874470472335815,
+      "learning_rate": 4.999533680694493e-05,
+      "loss": 4.0951,
+      "step": 2760
+    },
+    {
+      "epoch": 0.02273885192636891,
+      "grad_norm": 1.5923190116882324,
+      "learning_rate": 4.9995212187645416e-05,
+      "loss": 4.1031,
+      "step": 2790
+    },
+    {
+      "epoch": 0.02298335571052341,
+      "grad_norm": 0.935354471206665,
+      "learning_rate": 4.9995085925225693e-05,
+      "loss": 4.0804,
+      "step": 2820
+    },
+    {
+      "epoch": 0.023227859494677917,
+      "grad_norm": 1.6059459447860718,
+      "learning_rate": 4.999495801969404e-05,
+      "loss": 4.0703,
+      "step": 2850
+    },
+    {
+      "epoch": 0.023472363278832422,
+      "grad_norm": 1.1116045713424683,
+      "learning_rate": 4.9994828471058876e-05,
+      "loss": 4.0579,
+      "step": 2880
+    },
+    {
+      "epoch": 0.023716867062986924,
+      "grad_norm": 1.0068392753601074,
+      "learning_rate": 4.9994697279328714e-05,
+      "loss": 4.0635,
+      "step": 2910
+    },
+    {
+      "epoch": 0.02396137084714143,
+      "grad_norm": 0.9715657234191895,
+      "learning_rate": 4.9994564444512176e-05,
+      "loss": 4.068,
+      "step": 2940
+    },
+    {
+      "epoch": 0.024205874631295934,
+      "grad_norm": 1.1550418138504028,
+      "learning_rate": 4.9994429966618e-05,
+      "loss": 4.0416,
+      "step": 2970
+    },
+    {
+      "epoch": 0.02445037841545044,
+      "grad_norm": 1.1463241577148438,
+      "learning_rate": 4.999429384565502e-05,
+      "loss": 4.0224,
+      "step": 3000
+    },
+    {
+      "epoch": 0.02469488219960494,
+      "grad_norm": 0.9614786505699158,
+      "learning_rate": 4.999415608163217e-05,
+      "loss": 4.0092,
+      "step": 3030
+    },
+    {
+      "epoch": 0.024939385983759447,
+      "grad_norm": 1.0928614139556885,
+      "learning_rate": 4.999401667455854e-05,
+      "loss": 4.0083,
+      "step": 3060
+    },
+    {
+      "epoch": 0.025183889767913952,
+      "grad_norm": 1.0884851217269897,
+      "learning_rate": 4.9993875624443274e-05,
+      "loss": 4.0251,
+      "step": 3090
+    },
+    {
+      "epoch": 0.025428393552068457,
+      "grad_norm": 1.0158635377883911,
+      "learning_rate": 4.9993732931295646e-05,
+      "loss": 3.9629,
+      "step": 3120
+    },
+    {
+      "epoch": 0.02567289733622296,
+      "grad_norm": 1.0032926797866821,
+      "learning_rate": 4.999358859512503e-05,
+      "loss": 4.0015,
+      "step": 3150
+    },
+    {
+      "epoch": 0.025917401120377465,
+      "grad_norm": 0.9303460121154785,
+      "learning_rate": 4.9993442615940936e-05,
+      "loss": 4.0066,
+      "step": 3180
+    },
+    {
+      "epoch": 0.02616190490453197,
+      "grad_norm": 1.0750280618667603,
+      "learning_rate": 4.999329499375292e-05,
+      "loss": 3.9937,
+      "step": 3210
+    },
+    {
+      "epoch": 0.026406408688686475,
+      "grad_norm": 1.2123857736587524,
+      "learning_rate": 4.999314572857074e-05,
+      "loss": 3.9891,
+      "step": 3240
+    },
+    {
+      "epoch": 0.026650912472840977,
+      "grad_norm": 0.988030731678009,
+      "learning_rate": 4.9992994820404174e-05,
+      "loss": 3.9779,
+      "step": 3270
+    },
+    {
+      "epoch": 0.026895416256995482,
+      "grad_norm": 1.0025092363357544,
+      "learning_rate": 4.999284226926314e-05,
+      "loss": 3.9684,
+      "step": 3300
+    },
+    {
+      "epoch": 0.027139920041149988,
+      "grad_norm": 0.9497402310371399,
+      "learning_rate": 4.999268807515768e-05,
+      "loss": 3.9542,
+      "step": 3330
+    },
+    {
+      "epoch": 0.027384423825304493,
+      "grad_norm": 0.9898300766944885,
+      "learning_rate": 4.999253223809792e-05,
+      "loss": 3.9462,
+      "step": 3360
+    },
+    {
+      "epoch": 0.027628927609458995,
+      "grad_norm": 0.8951054215431213,
+      "learning_rate": 4.999237475809411e-05,
+      "loss": 3.9535,
+      "step": 3390
+    },
+    {
+      "epoch": 0.0278734313936135,
+      "grad_norm": 1.1410372257232666,
+      "learning_rate": 4.99922156351566e-05,
+      "loss": 3.9701,
+      "step": 3420
+    },
+    {
+      "epoch": 0.028117935177768005,
+      "grad_norm": 0.9642576575279236,
+      "learning_rate": 4.999205486929586e-05,
+      "loss": 3.9532,
+      "step": 3450
+    },
+    {
+      "epoch": 0.028362438961922507,
+      "grad_norm": 0.9304039478302002,
+      "learning_rate": 4.999189246052245e-05,
+      "loss": 3.9335,
+      "step": 3480
+    },
+    {
+      "epoch": 0.028606942746077012,
+      "grad_norm": 0.9684301614761353,
+      "learning_rate": 4.999172840884704e-05,
+      "loss": 3.9289,
+      "step": 3510
+    },
+    {
+      "epoch": 0.028851446530231518,
+      "grad_norm": 0.9566449522972107,
+      "learning_rate": 4.999156271428043e-05,
+      "loss": 3.92,
+      "step": 3540
+    },
+    {
+      "epoch": 0.029095950314386023,
+      "grad_norm": 0.9796332120895386,
+      "learning_rate": 4.9991395376833496e-05,
+      "loss": 3.9173,
+      "step": 3570
+    },
+    {
+      "epoch": 0.029340454098540525,
+      "grad_norm": 0.939052164554596,
+      "learning_rate": 4.999122639651725e-05,
+      "loss": 3.8938,
+      "step": 3600
+    },
+    {
+      "epoch": 0.02958495788269503,
+      "grad_norm": 0.8773409724235535,
+      "learning_rate": 4.9991055773342795e-05,
+      "loss": 3.9225,
+      "step": 3630
+    },
+    {
+      "epoch": 0.029829461666849535,
+      "grad_norm": 0.9443736672401428,
+      "learning_rate": 4.9990883507321354e-05,
+      "loss": 3.8889,
+      "step": 3660
+    },
+    {
+      "epoch": 0.03007396545100404,
+      "grad_norm": 0.9129781723022461,
+      "learning_rate": 4.999070959846424e-05,
+      "loss": 3.894,
+      "step": 3690
+    },
+    {
+      "epoch": 0.030318469235158543,
+      "grad_norm": 0.988441526889801,
+      "learning_rate": 4.999053404678289e-05,
+      "loss": 3.8968,
+      "step": 3720
+    },
+    {
+      "epoch": 0.030562973019313048,
+      "grad_norm": 1.0134508609771729,
+      "learning_rate": 4.999035685228884e-05,
+      "loss": 3.873,
+      "step": 3750
+    },
+    {
+      "epoch": 0.030807476803467553,
+      "grad_norm": 0.9853084087371826,
+      "learning_rate": 4.999017801499375e-05,
+      "loss": 3.8921,
+      "step": 3780
+    },
+    {
+      "epoch": 0.03105198058762206,
+      "grad_norm": 1.0263316631317139,
+      "learning_rate": 4.998999753490937e-05,
+      "loss": 3.8691,
+      "step": 3810
+    },
+    {
+      "epoch": 0.03129648437177656,
+      "grad_norm": 1.2496927976608276,
+      "learning_rate": 4.998981541204757e-05,
+      "loss": 3.8631,
+      "step": 3840
+    },
+    {
+      "epoch": 0.031540988155931066,
+      "grad_norm": 1.2647733688354492,
+      "learning_rate": 4.998963164642031e-05,
+      "loss": 3.8676,
+      "step": 3870
+    },
+    {
+      "epoch": 0.03178549194008557,
+      "grad_norm": 0.8532811403274536,
+      "learning_rate": 4.9989446238039676e-05,
+      "loss": 3.8196,
+      "step": 3900
+    },
+    {
+      "epoch": 0.032029995724240076,
+      "grad_norm": 0.9394932985305786,
+      "learning_rate": 4.998925918691786e-05,
+      "loss": 3.8464,
+      "step": 3930
+    },
+    {
+      "epoch": 0.03227449950839458,
+      "grad_norm": 0.9216808676719666,
+      "learning_rate": 4.998907049306715e-05,
+      "loss": 3.8294,
+      "step": 3960
+    },
+    {
+      "epoch": 0.03251900329254909,
+      "grad_norm": 0.9735797047615051,
+      "learning_rate": 4.998888015649996e-05,
+      "loss": 3.8412,
+      "step": 3990
+    },
+    {
+      "epoch": 0.032763507076703585,
+      "grad_norm": 0.9803655743598938,
+      "learning_rate": 4.99886881772288e-05,
+      "loss": 3.834,
+      "step": 4020
+    },
+    {
+      "epoch": 0.03300801086085809,
+      "grad_norm": 0.8677938580513,
+      "learning_rate": 4.998849455526628e-05,
+      "loss": 3.8325,
+      "step": 4050
+    },
+    {
+      "epoch": 0.033252514645012596,
+      "grad_norm": 1.0530720949172974,
+      "learning_rate": 4.998829929062515e-05,
+      "loss": 3.8125,
+      "step": 4080
+    },
+    {
+      "epoch": 0.0334970184291671,
+      "grad_norm": 0.9517835378646851,
+      "learning_rate": 4.998810238331822e-05,
+      "loss": 3.8079,
+      "step": 4110
+    },
+    {
+      "epoch": 0.033741522213321606,
+      "grad_norm": 0.9056320190429688,
+      "learning_rate": 4.998790383335845e-05,
+      "loss": 3.8023,
+      "step": 4140
+    },
+    {
+      "epoch": 0.03398602599747611,
+      "grad_norm": 1.042631983757019,
+      "learning_rate": 4.9987703640758894e-05,
+      "loss": 3.8089,
+      "step": 4170
+    },
+    {
+      "epoch": 0.03423052978163062,
+      "grad_norm": 1.003548502922058,
+      "learning_rate": 4.99875018055327e-05,
+      "loss": 3.8088,
+      "step": 4200
+    },
+    {
+      "epoch": 0.034475033565785115,
+      "grad_norm": 0.9754064083099365,
+      "learning_rate": 4.998729832769315e-05,
+      "loss": 3.7969,
+      "step": 4230
+    },
+    {
+      "epoch": 0.03471953734993962,
+      "grad_norm": 0.9784180521965027,
+      "learning_rate": 4.998709320725361e-05,
+      "loss": 3.8066,
+      "step": 4260
+    },
+    {
+      "epoch": 0.034964041134094126,
+      "grad_norm": 0.9522731900215149,
+      "learning_rate": 4.998688644422756e-05,
+      "loss": 3.8062,
+      "step": 4290
+    },
+    {
+      "epoch": 0.03520854491824863,
+      "grad_norm": 1.0362435579299927,
+      "learning_rate": 4.998667803862861e-05,
+      "loss": 3.8007,
+      "step": 4320
+    },
+    {
+      "epoch": 0.03545304870240314,
+      "grad_norm": 0.899608850479126,
+      "learning_rate": 4.9986467990470445e-05,
+      "loss": 3.8013,
+      "step": 4350
+    },
+    {
+      "epoch": 0.03569755248655764,
+      "grad_norm": 0.9725902676582336,
+      "learning_rate": 4.998625629976688e-05,
+      "loss": 3.7688,
+      "step": 4380
+    },
+    {
+      "epoch": 0.03594205627071215,
+      "grad_norm": 0.9904030561447144,
+      "learning_rate": 4.998604296653182e-05,
+      "loss": 3.7503,
+      "step": 4410
+    },
+    {
+      "epoch": 0.03618656005486665,
+      "grad_norm": 0.9177742004394531,
+      "learning_rate": 4.99858279907793e-05,
+      "loss": 3.7664,
+      "step": 4440
+    },
+    {
+      "epoch": 0.03643106383902115,
+      "grad_norm": 0.964356005191803,
+      "learning_rate": 4.998561137252346e-05,
+      "loss": 3.7574,
+      "step": 4470
+    },
+    {
+      "epoch": 0.036675567623175656,
+      "grad_norm": 1.0006535053253174,
+      "learning_rate": 4.9985393111778525e-05,
+      "loss": 3.7616,
+      "step": 4500
+    },
+    {
+      "epoch": 0.03692007140733016,
+      "grad_norm": 0.9334198236465454,
+      "learning_rate": 4.998517320855884e-05,
+      "loss": 3.776,
+      "step": 4530
+    },
+    {
+      "epoch": 0.03716457519148467,
+      "grad_norm": 0.9192038774490356,
+      "learning_rate": 4.998495166287887e-05,
+      "loss": 3.7552,
+      "step": 4560
+    },
+    {
+      "epoch": 0.03740907897563917,
+      "grad_norm": 0.9278284311294556,
+      "learning_rate": 4.998472847475318e-05,
+      "loss": 3.7508,
+      "step": 4590
+    },
+    {
+      "epoch": 0.03765358275979368,
+      "grad_norm": 0.870951771736145,
+      "learning_rate": 4.998450364419643e-05,
+      "loss": 3.7419,
+      "step": 4620
+    },
+    {
+      "epoch": 0.03789808654394818,
+      "grad_norm": 0.8986937403678894,
+      "learning_rate": 4.998427717122342e-05,
+      "loss": 3.7285,
+      "step": 4650
+    },
+    {
+      "epoch": 0.03814259032810268,
+      "grad_norm": 0.8922073841094971,
+      "learning_rate": 4.9984049055849024e-05,
+      "loss": 3.7252,
+      "step": 4680
+    },
+    {
+      "epoch": 0.038387094112257186,
+      "grad_norm": 0.8716844916343689,
+      "learning_rate": 4.9983819298088234e-05,
+      "loss": 3.7243,
+      "step": 4710
+    },
+    {
+      "epoch": 0.03863159789641169,
+      "grad_norm": 0.9074581861495972,
+      "learning_rate": 4.9983587897956166e-05,
+      "loss": 3.7254,
+      "step": 4740
+    },
+    {
+      "epoch": 0.0388761016805662,
+      "grad_norm": 0.8903953433036804,
+      "learning_rate": 4.998335485546802e-05,
+      "loss": 3.7363,
+      "step": 4770
+    },
+    {
+      "epoch": 0.0391206054647207,
+      "grad_norm": 1.0858160257339478,
+      "learning_rate": 4.998312017063912e-05,
+      "loss": 3.708,
+      "step": 4800
+    },
+    {
+      "epoch": 0.03936510924887521,
+      "grad_norm": 1.1008403301239014,
+      "learning_rate": 4.9982883843484895e-05,
+      "loss": 3.727,
+      "step": 4830
+    },
+    {
+      "epoch": 0.03960961303302971,
+      "grad_norm": 0.9895632863044739,
+      "learning_rate": 4.998264587402088e-05,
+      "loss": 3.7383,
+      "step": 4860
+    },
+    {
+      "epoch": 0.03985411681718422,
+      "grad_norm": 1.0174016952514648,
+      "learning_rate": 4.9982406262262726e-05,
+      "loss": 3.6967,
+      "step": 4890
+    },
+    {
+      "epoch": 0.040098620601338716,
+      "grad_norm": 0.9164692163467407,
+      "learning_rate": 4.9982165008226175e-05,
+      "loss": 3.7038,
+      "step": 4920
+    },
+    {
+      "epoch": 0.04034312438549322,
+      "grad_norm": 0.8442469835281372,
+      "learning_rate": 4.998192211192708e-05,
+      "loss": 3.7045,
+      "step": 4950
+    },
+    {
+      "epoch": 0.04058762816964773,
+      "grad_norm": 0.9134944677352905,
+      "learning_rate": 4.9981677573381423e-05,
+      "loss": 3.6909,
+      "step": 4980
+    },
+    {
+      "epoch": 0.04083213195380223,
+      "grad_norm": 0.9397178292274475,
+      "learning_rate": 4.9981431392605274e-05,
+      "loss": 3.7087,
+      "step": 5010
+    },
+    {
+      "epoch": 0.04107663573795674,
+      "grad_norm": 1.1442530155181885,
+      "learning_rate": 4.998118356961481e-05,
+      "loss": 3.6804,
+      "step": 5040
+    },
+    {
+      "epoch": 0.04132113952211124,
+      "grad_norm": 0.8651236891746521,
+      "learning_rate": 4.998093410442632e-05,
+      "loss": 3.668,
+      "step": 5070
+    },
+    {
+      "epoch": 0.04156564330626575,
+      "grad_norm": 1.0404279232025146,
+      "learning_rate": 4.998068299705623e-05,
+      "loss": 3.673,
+      "step": 5100
+    },
+    {
+      "epoch": 0.041810147090420254,
+      "grad_norm": 0.9545896053314209,
+      "learning_rate": 4.9980430247521014e-05,
+      "loss": 3.6635,
+      "step": 5130
+    },
+    {
+      "epoch": 0.04205465087457475,
+      "grad_norm": 0.9344103932380676,
+      "learning_rate": 4.998017585583731e-05,
+      "loss": 3.6834,
+      "step": 5160
+    },
+    {
+      "epoch": 0.04229915465872926,
+      "grad_norm": 1.0414049625396729,
+      "learning_rate": 4.9979919822021824e-05,
+      "loss": 3.6868,
+      "step": 5190
+    },
+    {
+      "epoch": 0.04254365844288376,
+      "grad_norm": 0.9391636252403259,
+      "learning_rate": 4.99796621460914e-05,
+      "loss": 3.6671,
+      "step": 5220
+    },
+    {
+      "epoch": 0.04278816222703827,
+      "grad_norm": 0.8789961934089661,
+      "learning_rate": 4.9979402828062963e-05,
+      "loss": 3.6575,
+      "step": 5250
+    },
+    {
+      "epoch": 0.04303266601119277,
+      "grad_norm": 0.8953958749771118,
+      "learning_rate": 4.997914186795358e-05,
+      "loss": 3.6741,
+      "step": 5280
+    },
+    {
+      "epoch": 0.04327716979534728,
+      "grad_norm": 0.8470034003257751,
+      "learning_rate": 4.9978879265780385e-05,
+      "loss": 3.6517,
+      "step": 5310
+    },
+    {
+      "epoch": 0.043521673579501784,
+      "grad_norm": 0.8863801956176758,
+      "learning_rate": 4.997861502156066e-05,
+      "loss": 3.6667,
+      "step": 5340
+    },
+    {
+      "epoch": 0.04376617736365628,
+      "grad_norm": 0.9392006993293762,
+      "learning_rate": 4.997834913531176e-05,
+      "loss": 3.6278,
+      "step": 5370
+    },
+    {
+      "epoch": 0.04401068114781079,
+      "grad_norm": 0.8935637474060059,
+      "learning_rate": 4.9978081607051176e-05,
+      "loss": 3.6656,
+      "step": 5400
+    },
+    {
+      "epoch": 0.04425518493196529,
+      "grad_norm": 0.9051403403282166,
+      "learning_rate": 4.997781243679648e-05,
+      "loss": 3.6439,
+      "step": 5430
+    },
+    {
+      "epoch": 0.0444996887161198,
+      "grad_norm": 0.9147741794586182,
+      "learning_rate": 4.9977541624565374e-05,
+      "loss": 3.6362,
+      "step": 5460
+    },
+    {
+      "epoch": 0.0447441925002743,
+      "grad_norm": 0.8926962614059448,
+      "learning_rate": 4.9977269170375665e-05,
+      "loss": 3.6179,
+      "step": 5490
+    },
+    {
+      "epoch": 0.04498869628442881,
+      "grad_norm": 0.9078648686408997,
+      "learning_rate": 4.997699507424526e-05,
+      "loss": 3.6323,
+      "step": 5520
+    },
+    {
+      "epoch": 0.045233200068583314,
+      "grad_norm": 0.8844118118286133,
+      "learning_rate": 4.997671933619218e-05,
+      "loss": 3.6219,
+      "step": 5550
+    },
+    {
+      "epoch": 0.04547770385273782,
+      "grad_norm": 0.9098038077354431,
+      "learning_rate": 4.9976441956234546e-05,
+      "loss": 3.6341,
+      "step": 5580
+    },
+    {
+      "epoch": 0.04572220763689232,
+      "grad_norm": 0.8704332709312439,
+      "learning_rate": 4.99761629343906e-05,
+      "loss": 3.6038,
+      "step": 5610
+    },
+    {
+      "epoch": 0.04596671142104682,
+      "grad_norm": 0.9353320002555847,
+      "learning_rate": 4.9975882270678676e-05,
+      "loss": 3.6171,
+      "step": 5640
+    },
+    {
+      "epoch": 0.04621121520520133,
+      "grad_norm": 1.0025008916854858,
+      "learning_rate": 4.997559996511723e-05,
+      "loss": 3.6084,
+      "step": 5670
+    },
+    {
+      "epoch": 0.04645571898935583,
+      "grad_norm": 0.8339236974716187,
+      "learning_rate": 4.997531601772481e-05,
+      "loss": 3.6271,
+      "step": 5700
+    },
+    {
+      "epoch": 0.04670022277351034,
+      "grad_norm": 0.8991410136222839,
+      "learning_rate": 4.9975030428520084e-05,
+      "loss": 3.5897,
+      "step": 5730
+    },
+    {
+      "epoch": 0.046944726557664844,
+      "grad_norm": 0.8854517936706543,
+      "learning_rate": 4.997474319752184e-05,
+      "loss": 3.603,
+      "step": 5760
+    },
+    {
+      "epoch": 0.04718923034181935,
+      "grad_norm": 0.9445385932922363,
+      "learning_rate": 4.997445432474895e-05,
+      "loss": 3.6084,
+      "step": 5790
+    },
+    {
+      "epoch": 0.04743373412597385,
+      "grad_norm": 0.871096134185791,
+      "learning_rate": 4.9974163810220406e-05,
+      "loss": 3.5892,
+      "step": 5820
+    },
+    {
+      "epoch": 0.04767823791012835,
+      "grad_norm": 0.8828051686286926,
+      "learning_rate": 4.99738716539553e-05,
+      "loss": 3.6075,
+      "step": 5850
+    },
+    {
+      "epoch": 0.04792274169428286,
+      "grad_norm": 0.8362672328948975,
+      "learning_rate": 4.997357785597284e-05,
+      "loss": 3.5772,
+      "step": 5880
+    },
+    {
+      "epoch": 0.048167245478437364,
+      "grad_norm": 0.9075796008110046,
+      "learning_rate": 4.997328241629234e-05,
+      "loss": 3.5881,
+      "step": 5910
+    },
+    {
+      "epoch": 0.04841174926259187,
+      "grad_norm": 0.9475303888320923,
+      "learning_rate": 4.997298533493323e-05,
+      "loss": 3.5963,
+      "step": 5940
+    },
+    {
+      "epoch": 0.048656253046746374,
+      "grad_norm": 0.807908296585083,
+      "learning_rate": 4.997268661191503e-05,
+      "loss": 3.5776,
+      "step": 5970
+    },
+    {
+      "epoch": 0.04890075683090088,
+      "grad_norm": 0.9100547432899475,
+      "learning_rate": 4.9972386247257385e-05,
+      "loss": 3.5924,
+      "step": 6000
+    },
+    {
+      "epoch": 0.049145260615055385,
+      "grad_norm": 0.79295814037323,
+      "learning_rate": 4.9972084240980025e-05,
+      "loss": 3.5603,
+      "step": 6030
+    },
+    {
+      "epoch": 0.04938976439920988,
+      "grad_norm": 0.8548495173454285,
+      "learning_rate": 4.997178059310281e-05,
+      "loss": 3.5624,
+      "step": 6060
+    },
+    {
+      "epoch": 0.04963426818336439,
+      "grad_norm": 0.8862835168838501,
+      "learning_rate": 4.997147530364571e-05,
+      "loss": 3.5738,
+      "step": 6090
+    },
+    {
+      "epoch": 0.049878771967518894,
+      "grad_norm": 1.0974136590957642,
+      "learning_rate": 4.9971168372628793e-05,
+      "loss": 3.5549,
+      "step": 6120
+    },
+    {
+      "epoch": 0.0501232757516734,
+      "grad_norm": 0.9224910140037537,
+      "learning_rate": 4.997085980007222e-05,
+      "loss": 3.5632,
+      "step": 6150
+    },
+    {
+      "epoch": 0.050367779535827904,
+      "grad_norm": 0.8416465520858765,
+      "learning_rate": 4.99705495859963e-05,
+      "loss": 3.5605,
+      "step": 6180
+    },
+    {
+      "epoch": 0.05061228331998241,
+      "grad_norm": 0.9750398993492126,
+      "learning_rate": 4.99702377304214e-05,
+      "loss": 3.5652,
+      "step": 6210
+    },
+    {
+      "epoch": 0.050856787104136915,
+      "grad_norm": 1.0585671663284302,
+      "learning_rate": 4.9969924233368036e-05,
+      "loss": 3.5605,
+      "step": 6240
+    },
+    {
+      "epoch": 0.05110129088829142,
+      "grad_norm": 1.0180178880691528,
+      "learning_rate": 4.996960909485681e-05,
+      "loss": 3.5543,
+      "step": 6270
+    },
+    {
+      "epoch": 0.05134579467244592,
+      "grad_norm": 0.8081057667732239,
+      "learning_rate": 4.9969292314908446e-05,
+      "loss": 3.5551,
+      "step": 6300
+    },
+    {
+      "epoch": 0.051590298456600424,
+      "grad_norm": 0.8196061253547668,
+      "learning_rate": 4.996897389354376e-05,
+      "loss": 3.5566,
+      "step": 6330
+    },
+    {
+      "epoch": 0.05183480224075493,
+      "grad_norm": 0.9229000806808472,
+      "learning_rate": 4.9968653830783684e-05,
+      "loss": 3.5439,
+      "step": 6360
+    },
+    {
+      "epoch": 0.052079306024909434,
+      "grad_norm": 0.8422220945358276,
+      "learning_rate": 4.996833212664927e-05,
+      "loss": 3.5413,
+      "step": 6390
+    },
+    {
+      "epoch": 0.05232380980906394,
+      "grad_norm": 0.9290680885314941,
+      "learning_rate": 4.996800878116166e-05,
+      "loss": 3.5245,
+      "step": 6420
+    },
+    {
+      "epoch": 0.052568313593218445,
+      "grad_norm": 0.7990279793739319,
+      "learning_rate": 4.99676837943421e-05,
+      "loss": 3.5519,
+      "step": 6450
+    },
+    {
+      "epoch": 0.05281281737737295,
+      "grad_norm": 0.8968884348869324,
+      "learning_rate": 4.996735716621196e-05,
+      "loss": 3.5306,
+      "step": 6480
+    },
+    {
+      "epoch": 0.05305732116152745,
+      "grad_norm": 0.8280816674232483,
+      "learning_rate": 4.996702889679272e-05,
+      "loss": 3.5407,
+      "step": 6510
+    },
+    {
+      "epoch": 0.053301824945681954,
+      "grad_norm": 0.9637565612792969,
+      "learning_rate": 4.996669898610595e-05,
+      "loss": 3.5361,
+      "step": 6540
+    },
+    {
+      "epoch": 0.05354632872983646,
+      "grad_norm": 0.8355570435523987,
+      "learning_rate": 4.996636743417334e-05,
+      "loss": 3.5215,
+      "step": 6570
+    },
+    {
+      "epoch": 0.053790832513990965,
+      "grad_norm": 0.7987068295478821,
+      "learning_rate": 4.996603424101669e-05,
+      "loss": 3.5135,
+      "step": 6600
+    },
+    {
+      "epoch": 0.05403533629814547,
+      "grad_norm": 0.8417894840240479,
+      "learning_rate": 4.996569940665789e-05,
+      "loss": 3.5254,
+      "step": 6630
+    },
+    {
+      "epoch": 0.054279840082299975,
+      "grad_norm": 0.8814358115196228,
+      "learning_rate": 4.996536293111896e-05,
+      "loss": 3.5091,
+      "step": 6660
+    },
+    {
+      "epoch": 0.05452434386645448,
+      "grad_norm": 0.8625611662864685,
+      "learning_rate": 4.996502481442202e-05,
+      "loss": 3.5207,
+      "step": 6690
+    },
+    {
+      "epoch": 0.054768847650608986,
+      "grad_norm": 0.8265979886054993,
+      "learning_rate": 4.9964685056589314e-05,
+      "loss": 3.5011,
+      "step": 6720
+    },
+    {
+      "epoch": 0.055013351434763484,
+      "grad_norm": 0.8092629313468933,
+      "learning_rate": 4.996434365764314e-05,
+      "loss": 3.4897,
+      "step": 6750
+    },
+    {
+      "epoch": 0.05525785521891799,
+      "grad_norm": 0.917871356010437,
+      "learning_rate": 4.996400061760597e-05,
+      "loss": 3.5188,
+      "step": 6780
+    },
+    {
+      "epoch": 0.055502359003072495,
+      "grad_norm": 0.8477138876914978,
+      "learning_rate": 4.996365593650033e-05,
+      "loss": 3.5142,
+      "step": 6810
+    },
+    {
+      "epoch": 0.055746862787227,
+      "grad_norm": 0.7970532178878784,
+      "learning_rate": 4.99633096143489e-05,
+      "loss": 3.5112,
+      "step": 6840
+    },
+    {
+      "epoch": 0.055991366571381505,
+      "grad_norm": 0.8973976969718933,
+      "learning_rate": 4.9962961651174436e-05,
+      "loss": 3.5156,
+      "step": 6870
+    },
+    {
+      "epoch": 0.05623587035553601,
+      "grad_norm": 0.962066113948822,
+      "learning_rate": 4.9962612046999827e-05,
+      "loss": 3.4848,
+      "step": 6900
+    },
+    {
+      "epoch": 0.056480374139690516,
+      "grad_norm": 0.9020524621009827,
+      "learning_rate": 4.996226080184803e-05,
+      "loss": 3.503,
+      "step": 6930
+    },
+    {
+      "epoch": 0.056724877923845014,
+      "grad_norm": 0.8064045906066895,
+      "learning_rate": 4.996190791574215e-05,
+      "loss": 3.4982,
+      "step": 6960
+    },
+    {
+      "epoch": 0.05696938170799952,
+      "grad_norm": 0.8507282733917236,
+      "learning_rate": 4.996155338870538e-05,
+      "loss": 3.4713,
+      "step": 6990
+    },
+    {
+      "epoch": 0.057213885492154025,
+      "grad_norm": 0.8574703931808472,
+      "learning_rate": 4.9961197220761035e-05,
+      "loss": 3.4924,
+      "step": 7020
+    },
+    {
+      "epoch": 0.05745838927630853,
+      "grad_norm": 0.8214017152786255,
+      "learning_rate": 4.996083941193252e-05,
+      "loss": 3.4773,
+      "step": 7050
+    },
+    {
+      "epoch": 0.057702893060463036,
+      "grad_norm": 0.8267714977264404,
+      "learning_rate": 4.9960479962243367e-05,
+      "loss": 3.4836,
+      "step": 7080
+    },
+    {
+      "epoch": 0.05794739684461754,
+      "grad_norm": 0.8415769934654236,
+      "learning_rate": 4.996011887171719e-05,
+      "loss": 3.4871,
+      "step": 7110
+    },
+    {
+      "epoch": 0.058191900628772046,
+      "grad_norm": 0.9093459248542786,
+      "learning_rate": 4.995975614037773e-05,
+      "loss": 3.4777,
+      "step": 7140
+    },
+    {
+      "epoch": 0.05843640441292655,
+      "grad_norm": 0.865168571472168,
+      "learning_rate": 4.995939176824883e-05,
+      "loss": 3.4798,
+      "step": 7170
+    },
+    {
+      "epoch": 0.05868090819708105,
+      "grad_norm": 0.8812352418899536,
+      "learning_rate": 4.995902575535446e-05,
+      "loss": 3.4634,
+      "step": 7200
+    },
+    {
+      "epoch": 0.058925411981235555,
+      "grad_norm": 0.8318623900413513,
+      "learning_rate": 4.995865810171866e-05,
+      "loss": 3.4691,
+      "step": 7230
+    },
+    {
+      "epoch": 0.05916991576539006,
+      "grad_norm": 0.8475897312164307,
+      "learning_rate": 4.995828880736561e-05,
+      "loss": 3.4894,
+      "step": 7260
+    },
+    {
+      "epoch": 0.059414419549544566,
+      "grad_norm": 0.8317278027534485,
+      "learning_rate": 4.995791787231958e-05,
+      "loss": 3.4597,
+      "step": 7290
+    },
+    {
+      "epoch": 0.05965892333369907,
+      "grad_norm": 0.8189125657081604,
+      "learning_rate": 4.9957545296604965e-05,
+      "loss": 3.4495,
+      "step": 7320
+    },
+    {
+      "epoch": 0.059903427117853576,
+      "grad_norm": 0.8047211766242981,
+      "learning_rate": 4.9957171080246245e-05,
+      "loss": 3.4865,
+      "step": 7350
+    },
+    {
+      "epoch": 0.06014793090200808,
+      "grad_norm": 0.8353013396263123,
+      "learning_rate": 4.995679522326803e-05,
+      "loss": 3.4671,
+      "step": 7380
+    },
+    {
+      "epoch": 0.06039243468616259,
+      "grad_norm": 0.7747114300727844,
+      "learning_rate": 4.995641772569502e-05,
+      "loss": 3.4423,
+      "step": 7410
+    },
+    {
+      "epoch": 0.060636938470317085,
+      "grad_norm": 0.792785108089447,
+      "learning_rate": 4.995603858755203e-05,
+      "loss": 3.448,
+      "step": 7440
+    },
+    {
+      "epoch": 0.06088144225447159,
+      "grad_norm": 0.9009000658988953,
+      "learning_rate": 4.9955657808863985e-05,
+      "loss": 3.4471,
+      "step": 7470
+    },
+    {
+      "epoch": 0.061125946038626096,
+      "grad_norm": 0.8719607591629028,
+      "learning_rate": 4.995527538965593e-05,
+      "loss": 3.4512,
+      "step": 7500
+    },
+    {
+      "epoch": 0.0613704498227806,
+      "grad_norm": 0.840308666229248,
+      "learning_rate": 4.995489132995298e-05,
+      "loss": 3.4576,
+      "step": 7530
+    },
+    {
+      "epoch": 0.061614953606935106,
+      "grad_norm": 0.8216726779937744,
+      "learning_rate": 4.99545056297804e-05,
+      "loss": 3.4369,
+      "step": 7560
+    },
+    {
+      "epoch": 0.06185945739108961,
+      "grad_norm": 0.7841954827308655,
+      "learning_rate": 4.995411828916354e-05,
+      "loss": 3.4338,
+      "step": 7590
+    },
+    {
+      "epoch": 0.06210396117524412,
+      "grad_norm": 0.7832449674606323,
+      "learning_rate": 4.9953729308127874e-05,
+      "loss": 3.431,
+      "step": 7620
+    },
+    {
+      "epoch": 0.062348464959398615,
+      "grad_norm": 0.821207582950592,
+      "learning_rate": 4.995333868669895e-05,
+      "loss": 3.44,
+      "step": 7650
+    },
+    {
+      "epoch": 0.06259296874355312,
+      "grad_norm": 0.7705381512641907,
+      "learning_rate": 4.995294642490246e-05,
+      "loss": 3.4479,
+      "step": 7680
+    },
+    {
+      "epoch": 0.06283747252770763,
+      "grad_norm": 0.7850989699363708,
+      "learning_rate": 4.995255252276418e-05,
+      "loss": 3.4276,
+      "step": 7710
+    },
+    {
+      "epoch": 0.06308197631186213,
+      "grad_norm": 0.7798038721084595,
+      "learning_rate": 4.9952156980310016e-05,
+      "loss": 3.4376,
+      "step": 7740
+    },
+    {
+      "epoch": 0.06332648009601663,
+      "grad_norm": 0.8357362151145935,
+      "learning_rate": 4.9951759797565965e-05,
+      "loss": 3.4408,
+      "step": 7770
+    },
+    {
+      "epoch": 0.06357098388017114,
+      "grad_norm": 0.8627969622612,
+      "learning_rate": 4.995136097455815e-05,
+      "loss": 3.4328,
+      "step": 7800
+    },
+    {
+      "epoch": 0.06381548766432564,
+      "grad_norm": 0.7546987533569336,
+      "learning_rate": 4.995096051131276e-05,
+      "loss": 3.4168,
+      "step": 7830
+    },
+    {
+      "epoch": 0.06405999144848015,
+      "grad_norm": 0.8405272960662842,
+      "learning_rate": 4.995055840785614e-05,
+      "loss": 3.402,
+      "step": 7860
+    },
+    {
+      "epoch": 0.06430449523263465,
+      "grad_norm": 0.8333033919334412,
+      "learning_rate": 4.995015466421473e-05,
+      "loss": 3.4333,
+      "step": 7890
+    },
+    {
+      "epoch": 0.06454899901678916,
+      "grad_norm": 0.787031352519989,
+      "learning_rate": 4.9949749280415056e-05,
+      "loss": 3.4065,
+      "step": 7920
+    },
+    {
+      "epoch": 0.06479350280094366,
+      "grad_norm": 0.7898181676864624,
+      "learning_rate": 4.9949342256483766e-05,
+      "loss": 3.4023,
+      "step": 7950
+    },
+    {
+      "epoch": 0.06503800658509817,
+      "grad_norm": 0.7997327446937561,
+      "learning_rate": 4.9948933592447636e-05,
+      "loss": 3.4268,
+      "step": 7980
+    },
+    {
+      "epoch": 0.06528251036925267,
+      "grad_norm": 0.8030378818511963,
+      "learning_rate": 4.9948523288333506e-05,
+      "loss": 3.4106,
+      "step": 8010
+    },
+    {
+      "epoch": 0.06552701415340717,
+      "grad_norm": 0.8793920278549194,
+      "learning_rate": 4.994811134416836e-05,
+      "loss": 3.4075,
+      "step": 8040
+    },
+    {
+      "epoch": 0.06577151793756168,
+      "grad_norm": 0.810756504535675,
+      "learning_rate": 4.994769775997927e-05,
+      "loss": 3.4177,
+      "step": 8070
+    },
+    {
+      "epoch": 0.06601602172171618,
+      "grad_norm": 0.8193936347961426,
+      "learning_rate": 4.994728253579345e-05,
+      "loss": 3.4195,
+      "step": 8100
+    },
+    {
+      "epoch": 0.0662605255058707,
+      "grad_norm": 0.8704981803894043,
+      "learning_rate": 4.9946865671638166e-05,
+      "loss": 3.3981,
+      "step": 8130
+    },
+    {
+      "epoch": 0.06650502929002519,
+      "grad_norm": 0.9237441420555115,
+      "learning_rate": 4.9946447167540835e-05,
+      "loss": 3.4177,
+      "step": 8160
+    },
+    {
+      "epoch": 0.0667495330741797,
+      "grad_norm": 0.7941862344741821,
+      "learning_rate": 4.994602702352896e-05,
+      "loss": 3.4252,
+      "step": 8190
+    },
+    {
+      "epoch": 0.0669940368583342,
+      "grad_norm": 0.8230422735214233,
+      "learning_rate": 4.994560523963018e-05,
+      "loss": 3.3915,
+      "step": 8220
+    },
+    {
+      "epoch": 0.0672385406424887,
+      "grad_norm": 0.7928892970085144,
+      "learning_rate": 4.9945181815872196e-05,
+      "loss": 3.3738,
+      "step": 8250
+    },
+    {
+      "epoch": 0.06748304442664321,
+      "grad_norm": 0.798755407333374,
+      "learning_rate": 4.9944756752282855e-05,
+      "loss": 3.3991,
+      "step": 8280
+    },
+    {
+      "epoch": 0.06772754821079771,
+      "grad_norm": 0.744189977645874,
+      "learning_rate": 4.994433004889011e-05,
+      "loss": 3.3874,
+      "step": 8310
+    },
+    {
+      "epoch": 0.06797205199495222,
+      "grad_norm": 0.7570254802703857,
+      "learning_rate": 4.994390170572199e-05,
+      "loss": 3.3991,
+      "step": 8340
+    },
+    {
+      "epoch": 0.06821655577910672,
+      "grad_norm": 0.7630516886711121,
+      "learning_rate": 4.994347172280667e-05,
+      "loss": 3.38,
+      "step": 8370
+    },
+    {
+      "epoch": 0.06846105956326123,
+      "grad_norm": 0.8707906603813171,
+      "learning_rate": 4.994304010017241e-05,
+      "loss": 3.3794,
+      "step": 8400
+    },
+    {
+      "epoch": 0.06870556334741573,
+      "grad_norm": 0.7646734714508057,
+      "learning_rate": 4.994260683784758e-05,
+      "loss": 3.4037,
+      "step": 8430
+    },
+    {
+      "epoch": 0.06895006713157023,
+      "grad_norm": 0.8288954496383667,
+      "learning_rate": 4.9942171935860674e-05,
+      "loss": 3.3847,
+      "step": 8460
+    },
+    {
+      "epoch": 0.06919457091572474,
+      "grad_norm": 0.7819532752037048,
+      "learning_rate": 4.994173539424026e-05,
+      "loss": 3.3854,
+      "step": 8490
+    },
+    {
+      "epoch": 0.06943907469987924,
+      "grad_norm": 0.7958394289016724,
+      "learning_rate": 4.994129721301506e-05,
+      "loss": 3.3937,
+      "step": 8520
+    },
+    {
+      "epoch": 0.06968357848403375,
+      "grad_norm": 0.8431640267372131,
+      "learning_rate": 4.994085739221386e-05,
+      "loss": 3.4072,
+      "step": 8550
+    },
+    {
+      "epoch": 0.06992808226818825,
+      "grad_norm": 0.7959804534912109,
+      "learning_rate": 4.994041593186558e-05,
+      "loss": 3.3662,
+      "step": 8580
+    },
+    {
+      "epoch": 0.07017258605234276,
+      "grad_norm": 0.7684959769248962,
+      "learning_rate": 4.993997283199924e-05,
+      "loss": 3.395,
+      "step": 8610
+    },
+    {
+      "epoch": 0.07041708983649726,
+      "grad_norm": 0.7869526147842407,
+      "learning_rate": 4.993952809264397e-05,
+      "loss": 3.3903,
+      "step": 8640
+    },
+    {
+      "epoch": 0.07066159362065176,
+      "grad_norm": 0.8075766563415527,
+      "learning_rate": 4.9939081713829006e-05,
+      "loss": 3.3732,
+      "step": 8670
+    },
+    {
+      "epoch": 0.07090609740480627,
+      "grad_norm": 0.7754388451576233,
+      "learning_rate": 4.993863369558369e-05,
+      "loss": 3.3836,
+      "step": 8700
+    },
+    {
+      "epoch": 0.07115060118896077,
+      "grad_norm": 0.7476597428321838,
+      "learning_rate": 4.9938184037937466e-05,
+      "loss": 3.3767,
+      "step": 8730
+    },
+    {
+      "epoch": 0.07139510497311528,
+      "grad_norm": 0.7789965271949768,
+      "learning_rate": 4.993773274091991e-05,
+      "loss": 3.3518,
+      "step": 8760
+    },
+    {
+      "epoch": 0.07163960875726978,
+      "grad_norm": 0.8229824900627136,
+      "learning_rate": 4.993727980456067e-05,
+      "loss": 3.3629,
+      "step": 8790
+    },
+    {
+      "epoch": 0.0718841125414243,
+      "grad_norm": 0.762668251991272,
+      "learning_rate": 4.993682522888954e-05,
+      "loss": 3.375,
+      "step": 8820
+    },
+    {
+      "epoch": 0.07212861632557879,
+      "grad_norm": 0.8019202947616577,
+      "learning_rate": 4.993636901393639e-05,
+      "loss": 3.3635,
+      "step": 8850
+    },
+    {
+      "epoch": 0.0723731201097333,
+      "grad_norm": 0.9177828431129456,
+      "learning_rate": 4.993591115973121e-05,
+      "loss": 3.37,
+      "step": 8880
+    },
+    {
+      "epoch": 0.0726176238938878,
+      "grad_norm": 0.8312643766403198,
+      "learning_rate": 4.9935451666304105e-05,
+      "loss": 3.3763,
+      "step": 8910
+    },
+    {
+      "epoch": 0.0728621276780423,
+      "grad_norm": 0.828384518623352,
+      "learning_rate": 4.993499053368528e-05,
+      "loss": 3.368,
+      "step": 8940
+    },
+    {
+      "epoch": 0.07310663146219681,
+      "grad_norm": 1.0234721899032593,
+      "learning_rate": 4.993452776190504e-05,
+      "loss": 3.3534,
+      "step": 8970
+    },
+    {
+      "epoch": 0.07335113524635131,
+      "grad_norm": 0.8042975068092346,
+      "learning_rate": 4.993406335099382e-05,
+      "loss": 3.3677,
+      "step": 9000
+    },
+    {
+      "epoch": 0.07359563903050582,
+      "grad_norm": 0.753710150718689,
+      "learning_rate": 4.993359730098214e-05,
+      "loss": 3.3652,
+      "step": 9030
+    },
+    {
+      "epoch": 0.07384014281466032,
+      "grad_norm": 0.8520479798316956,
+      "learning_rate": 4.993312961190064e-05,
+      "loss": 3.3511,
+      "step": 9060
+    },
+    {
+      "epoch": 0.07408464659881484,
+      "grad_norm": 0.7559372186660767,
+      "learning_rate": 4.993266028378006e-05,
+      "loss": 3.3438,
+      "step": 9090
+    },
+    {
+      "epoch": 0.07432915038296933,
+      "grad_norm": 0.7419788837432861,
+      "learning_rate": 4.993218931665126e-05,
+      "loss": 3.3453,
+      "step": 9120
+    },
+    {
+      "epoch": 0.07457365416712383,
+      "grad_norm": 0.8156657218933105,
+      "learning_rate": 4.993171671054519e-05,
+      "loss": 3.3654,
+      "step": 9150
+    },
+    {
+      "epoch": 0.07481815795127834,
+      "grad_norm": 0.8189958930015564,
+      "learning_rate": 4.993124246549293e-05,
+      "loss": 3.3534,
+      "step": 9180
+    },
+    {
+      "epoch": 0.07506266173543284,
+      "grad_norm": 0.7935733795166016,
+      "learning_rate": 4.9930766581525645e-05,
+      "loss": 3.3472,
+      "step": 9210
+    },
+    {
+      "epoch": 0.07530716551958735,
+      "grad_norm": 0.823512852191925,
+      "learning_rate": 4.993028905867463e-05,
+      "loss": 3.3476,
+      "step": 9240
+    },
+    {
+      "epoch": 0.07555166930374185,
+      "grad_norm": 0.8196141719818115,
+      "learning_rate": 4.992980989697126e-05,
+      "loss": 3.3479,
+      "step": 9270
+    },
+    {
+      "epoch": 0.07579617308789637,
+      "grad_norm": 0.7773461937904358,
+      "learning_rate": 4.992932909644705e-05,
+      "loss": 3.344,
+      "step": 9300
+    },
+    {
+      "epoch": 0.07604067687205086,
+      "grad_norm": 0.7703831791877747,
+      "learning_rate": 4.9928846657133596e-05,
+      "loss": 3.3495,
+      "step": 9330
+    },
+    {
+      "epoch": 0.07628518065620536,
+      "grad_norm": 0.7672068476676941,
+      "learning_rate": 4.992836257906262e-05,
+      "loss": 3.3607,
+      "step": 9360
+    },
+    {
+      "epoch": 0.07652968444035987,
+      "grad_norm": 0.7359826564788818,
+      "learning_rate": 4.9927876862265935e-05,
+      "loss": 3.3213,
+      "step": 9390
+    },
+    {
+      "epoch": 0.07677418822451437,
+      "grad_norm": 0.8856120705604553,
+      "learning_rate": 4.992738950677548e-05,
+      "loss": 3.3132,
+      "step": 9420
+    },
+    {
+      "epoch": 0.07701869200866888,
+      "grad_norm": 0.7555311918258667,
+      "learning_rate": 4.992690051262329e-05,
+      "loss": 3.3171,
+      "step": 9450
+    },
+    {
+      "epoch": 0.07726319579282338,
+      "grad_norm": 0.8101349472999573,
+      "learning_rate": 4.99264098798415e-05,
+      "loss": 3.3356,
+      "step": 9480
+    },
+    {
+      "epoch": 0.0775076995769779,
+      "grad_norm": 0.7647647857666016,
+      "learning_rate": 4.992591760846238e-05,
+      "loss": 3.318,
+      "step": 9510
+    },
+    {
+      "epoch": 0.0777522033611324,
+      "grad_norm": 0.7218281030654907,
+      "learning_rate": 4.9925423698518274e-05,
+      "loss": 3.3136,
+      "step": 9540
+    },
+    {
+      "epoch": 0.0779967071452869,
+      "grad_norm": 0.8410701155662537,
+      "learning_rate": 4.992492815004166e-05,
+      "loss": 3.3436,
+      "step": 9570
+    },
+    {
+      "epoch": 0.0782412109294414,
+      "grad_norm": 0.7226998209953308,
+      "learning_rate": 4.992443096306512e-05,
+      "loss": 3.3171,
+      "step": 9600
+    },
+    {
+      "epoch": 0.0784857147135959,
+      "grad_norm": 0.7702776193618774,
+      "learning_rate": 4.992393213762132e-05,
+      "loss": 3.3512,
+      "step": 9630
+    },
+    {
+      "epoch": 0.07873021849775041,
+      "grad_norm": 0.816752016544342,
+      "learning_rate": 4.992343167374307e-05,
+      "loss": 3.3189,
+      "step": 9660
+    },
+    {
+      "epoch": 0.07897472228190491,
+      "grad_norm": 0.854523241519928,
+      "learning_rate": 4.992292957146326e-05,
+      "loss": 3.315,
+      "step": 9690
+    },
+    {
+      "epoch": 0.07921922606605943,
+      "grad_norm": 0.7951149940490723,
+      "learning_rate": 4.992242583081489e-05,
+      "loss": 3.314,
+      "step": 9720
+    },
+    {
+      "epoch": 0.07946372985021392,
+      "grad_norm": 0.7649703621864319,
+      "learning_rate": 4.992192045183109e-05,
+      "loss": 3.2931,
+      "step": 9750
+    },
+    {
+      "epoch": 0.07970823363436844,
+      "grad_norm": 0.8076454401016235,
+      "learning_rate": 4.9921413434545075e-05,
+      "loss": 3.2975,
+      "step": 9780
+    },
+    {
+      "epoch": 0.07995273741852293,
+      "grad_norm": 0.7618278861045837,
+      "learning_rate": 4.992090477899018e-05,
+      "loss": 3.2909,
+      "step": 9810
+    },
+    {
+      "epoch": 0.08019724120267743,
+      "grad_norm": 0.748193085193634,
+      "learning_rate": 4.992039448519982e-05,
+      "loss": 3.2894,
+      "step": 9840
+    },
+    {
+      "epoch": 0.08044174498683195,
+      "grad_norm": 0.7488676309585571,
+      "learning_rate": 4.9919882553207566e-05,
+      "loss": 3.2905,
+      "step": 9870
+    },
+    {
+      "epoch": 0.08068624877098644,
+      "grad_norm": 0.7544564604759216,
+      "learning_rate": 4.9919368983047066e-05,
+      "loss": 3.2983,
+      "step": 9900
+    },
+    {
+      "epoch": 0.08093075255514096,
+      "grad_norm": 0.7570173144340515,
+      "learning_rate": 4.9918853774752074e-05,
+      "loss": 3.3151,
+      "step": 9930
+    },
+    {
+      "epoch": 0.08117525633929545,
+      "grad_norm": 0.7475308775901794,
+      "learning_rate": 4.991833692835646e-05,
+      "loss": 3.2878,
+      "step": 9960
+    },
+    {
+      "epoch": 0.08141976012344997,
+      "grad_norm": 0.858024001121521,
+      "learning_rate": 4.9917818443894203e-05,
+      "loss": 3.298,
+      "step": 9990
+    },
+    {
+      "epoch": 0.08166426390760446,
+      "grad_norm": 0.9019065499305725,
+      "learning_rate": 4.991729832139939e-05,
+      "loss": 3.2961,
+      "step": 10020
+    },
+    {
+      "epoch": 0.08190876769175896,
+      "grad_norm": 0.757036030292511,
+      "learning_rate": 4.991677656090621e-05,
+      "loss": 3.2939,
+      "step": 10050
+    },
+    {
+      "epoch": 0.08215327147591348,
+      "grad_norm": 0.7498080730438232,
+      "learning_rate": 4.991625316244896e-05,
+      "loss": 3.3187,
+      "step": 10080
+    },
+    {
+      "epoch": 0.08239777526006797,
+      "grad_norm": 0.7562547922134399,
+      "learning_rate": 4.991572812606205e-05,
+      "loss": 3.3113,
+      "step": 10110
+    },
+    {
+      "epoch": 0.08264227904422249,
+      "grad_norm": 0.7378952503204346,
+      "learning_rate": 4.991520145177998e-05,
+      "loss": 3.3021,
+      "step": 10140
+    },
+    {
+      "epoch": 0.08288678282837698,
+      "grad_norm": 0.7791491150856018,
+      "learning_rate": 4.991467313963739e-05,
+      "loss": 3.2856,
+      "step": 10170
+    },
+    {
+      "epoch": 0.0831312866125315,
+      "grad_norm": 0.7295765280723572,
+      "learning_rate": 4.991414318966901e-05,
+      "loss": 3.2815,
+      "step": 10200
+    },
+    {
+      "epoch": 0.083375790396686,
+      "grad_norm": 0.7607465982437134,
+      "learning_rate": 4.991361160190966e-05,
+      "loss": 3.2709,
+      "step": 10230
+    },
+    {
+      "epoch": 0.08362029418084051,
+      "grad_norm": 0.7715272903442383,
+      "learning_rate": 4.9913078376394304e-05,
+      "loss": 3.3011,
+      "step": 10260
+    },
+    {
+      "epoch": 0.083864797964995,
+      "grad_norm": 0.7366231679916382,
+      "learning_rate": 4.991254351315799e-05,
+      "loss": 3.2798,
+      "step": 10290
+    },
+    {
+      "epoch": 0.0841093017491495,
+      "grad_norm": 0.7520325183868408,
+      "learning_rate": 4.991200701223587e-05,
+      "loss": 3.2925,
+      "step": 10320
+    },
+    {
+      "epoch": 0.08435380553330402,
+      "grad_norm": 0.7658284902572632,
+      "learning_rate": 4.991146887366323e-05,
+      "loss": 3.2954,
+      "step": 10350
+    },
+    {
+      "epoch": 0.08459830931745851,
+      "grad_norm": 0.7013542056083679,
+      "learning_rate": 4.991092909747542e-05,
+      "loss": 3.2897,
+      "step": 10380
+    },
+    {
+      "epoch": 0.08484281310161303,
+      "grad_norm": 0.8035088181495667,
+      "learning_rate": 4.9910387683707946e-05,
+      "loss": 3.2876,
+      "step": 10410
+    },
+    {
+      "epoch": 0.08508731688576752,
+      "grad_norm": 0.7619040608406067,
+      "learning_rate": 4.9909844632396386e-05,
+      "loss": 3.278,
+      "step": 10440
+    },
+    {
+      "epoch": 0.08533182066992204,
+      "grad_norm": 0.7567503452301025,
+      "learning_rate": 4.9909299943576445e-05,
+      "loss": 3.2774,
+      "step": 10470
+    },
+    {
+      "epoch": 0.08557632445407654,
+      "grad_norm": 0.7545759677886963,
+      "learning_rate": 4.990875361728393e-05,
+      "loss": 3.2824,
+      "step": 10500
+    },
+    {
+      "epoch": 0.08582082823823103,
+      "grad_norm": 0.7955304980278015,
+      "learning_rate": 4.990820565355475e-05,
+      "loss": 3.2809,
+      "step": 10530
+    },
+    {
+      "epoch": 0.08606533202238555,
+      "grad_norm": 0.8105009198188782,
+      "learning_rate": 4.990765605242493e-05,
+      "loss": 3.2745,
+      "step": 10560
+    },
+    {
+      "epoch": 0.08630983580654004,
+      "grad_norm": 0.758289098739624,
+      "learning_rate": 4.990710481393061e-05,
+      "loss": 3.2549,
+      "step": 10590
+    },
+    {
+      "epoch": 0.08655433959069456,
+      "grad_norm": 0.7958084344863892,
+      "learning_rate": 4.9906551938108003e-05,
+      "loss": 3.3004,
+      "step": 10620
+    },
+    {
+      "epoch": 0.08679884337484906,
+      "grad_norm": 0.7208577394485474,
+      "learning_rate": 4.990599742499347e-05,
+      "loss": 3.291,
+      "step": 10650
+    },
+    {
+      "epoch": 0.08704334715900357,
+      "grad_norm": 0.7224754691123962,
+      "learning_rate": 4.990544127462346e-05,
+      "loss": 3.2676,
+      "step": 10680
+    },
+    {
+      "epoch": 0.08728785094315807,
+      "grad_norm": 0.7175268530845642,
+      "learning_rate": 4.9904883487034537e-05,
+      "loss": 3.2753,
+      "step": 10710
+    },
+    {
+      "epoch": 0.08753235472731256,
+      "grad_norm": 0.75492262840271,
+      "learning_rate": 4.990432406226336e-05,
+      "loss": 3.2707,
+      "step": 10740
+    },
+    {
+      "epoch": 0.08777685851146708,
+      "grad_norm": 0.7345853447914124,
+      "learning_rate": 4.990376300034671e-05,
+      "loss": 3.2593,
+      "step": 10770
+    },
+    {
+      "epoch": 0.08802136229562157,
+      "grad_norm": 0.719987154006958,
+      "learning_rate": 4.990320030132147e-05,
+      "loss": 3.2689,
+      "step": 10800
+    },
+    {
+      "epoch": 0.08826586607977609,
+      "grad_norm": 0.7458385229110718,
+      "learning_rate": 4.990263596522462e-05,
+      "loss": 3.2822,
+      "step": 10830
+    },
+    {
+      "epoch": 0.08851036986393059,
+      "grad_norm": 0.7266592383384705,
+      "learning_rate": 4.9902069992093275e-05,
+      "loss": 3.2818,
+      "step": 10860
+    },
+    {
+      "epoch": 0.0887548736480851,
+      "grad_norm": 0.6844702363014221,
+      "learning_rate": 4.990150238196463e-05,
+      "loss": 3.2711,
+      "step": 10890
+    },
+    {
+      "epoch": 0.0889993774322396,
+      "grad_norm": 0.7185301184654236,
+      "learning_rate": 4.9900933134876e-05,
+      "loss": 3.2682,
+      "step": 10920
+    },
+    {
+      "epoch": 0.0892438812163941,
+      "grad_norm": 0.7460073232650757,
+      "learning_rate": 4.990036225086481e-05,
+      "loss": 3.2419,
+      "step": 10950
+    },
+    {
+      "epoch": 0.0894883850005486,
+      "grad_norm": 0.7382895946502686,
+      "learning_rate": 4.9899789729968585e-05,
+      "loss": 3.2612,
+      "step": 10980
+    },
+    {
+      "epoch": 0.0897328887847031,
+      "grad_norm": 0.7204514741897583,
+      "learning_rate": 4.989921557222496e-05,
+      "loss": 3.2525,
+      "step": 11010
+    },
+    {
+      "epoch": 0.08997739256885762,
+      "grad_norm": 0.7512257695198059,
+      "learning_rate": 4.989863977767167e-05,
+      "loss": 3.2656,
+      "step": 11040
+    },
+    {
+      "epoch": 0.09022189635301212,
+      "grad_norm": 0.7579988837242126,
+      "learning_rate": 4.989806234634659e-05,
+      "loss": 3.2545,
+      "step": 11070
+    },
+    {
+      "epoch": 0.09046640013716663,
+      "grad_norm": 0.7989354729652405,
+      "learning_rate": 4.989748327828765e-05,
+      "loss": 3.2588,
+      "step": 11100
+    },
+    {
+      "epoch": 0.09071090392132113,
+      "grad_norm": 0.7362430095672607,
+      "learning_rate": 4.989690257353294e-05,
+      "loss": 3.2205,
+      "step": 11130
+    },
+    {
+      "epoch": 0.09095540770547564,
+      "grad_norm": 0.6724596619606018,
+      "learning_rate": 4.989632023212062e-05,
+      "loss": 3.252,
+      "step": 11160
+    },
+    {
+      "epoch": 0.09119991148963014,
+      "grad_norm": 0.804040789604187,
+      "learning_rate": 4.9895736254088975e-05,
+      "loss": 3.2554,
+      "step": 11190
+    },
+    {
+      "epoch": 0.09144441527378463,
+      "grad_norm": 0.7371083498001099,
+      "learning_rate": 4.989515063947641e-05,
+      "loss": 3.2551,
+      "step": 11220
+    },
+    {
+      "epoch": 0.09168891905793915,
+      "grad_norm": 0.8144158124923706,
+      "learning_rate": 4.9894563388321395e-05,
+      "loss": 3.2396,
+      "step": 11250
+    },
+    {
+      "epoch": 0.09193342284209365,
+      "grad_norm": 0.7379621863365173,
+      "learning_rate": 4.989397450066254e-05,
+      "loss": 3.2588,
+      "step": 11280
+    },
+    {
+      "epoch": 0.09217792662624816,
+      "grad_norm": 0.7908419370651245,
+      "learning_rate": 4.989338397653858e-05,
+      "loss": 3.2316,
+      "step": 11310
+    },
+    {
+      "epoch": 0.09242243041040266,
+      "grad_norm": 0.763767421245575,
+      "learning_rate": 4.98927918159883e-05,
+      "loss": 3.2585,
+      "step": 11340
+    },
+    {
+      "epoch": 0.09266693419455717,
+      "grad_norm": 0.7726889252662659,
+      "learning_rate": 4.989219801905066e-05,
+      "loss": 3.2481,
+      "step": 11370
+    },
+    {
+      "epoch": 0.09291143797871167,
+      "grad_norm": 0.7199206352233887,
+      "learning_rate": 4.989160258576469e-05,
+      "loss": 3.2468,
+      "step": 11400
+    },
+    {
+      "epoch": 0.09315594176286617,
+      "grad_norm": 0.7201483249664307,
+      "learning_rate": 4.98910055161695e-05,
+      "loss": 3.2226,
+      "step": 11430
+    },
+    {
+      "epoch": 0.09340044554702068,
+      "grad_norm": 0.724740743637085,
+      "learning_rate": 4.989040681030437e-05,
+      "loss": 3.2546,
+      "step": 11460
+    },
+    {
+      "epoch": 0.09364494933117518,
+      "grad_norm": 0.7120088338851929,
+      "learning_rate": 4.988980646820865e-05,
+      "loss": 3.2397,
+      "step": 11490
+    },
+    {
+      "epoch": 0.09388945311532969,
+      "grad_norm": 0.7296812534332275,
+      "learning_rate": 4.9889204489921804e-05,
+      "loss": 3.2201,
+      "step": 11520
+    },
+    {
+      "epoch": 0.09413395689948419,
+      "grad_norm": 0.7710677981376648,
+      "learning_rate": 4.9888600875483404e-05,
+      "loss": 3.2475,
+      "step": 11550
+    },
+    {
+      "epoch": 0.0943784606836387,
+      "grad_norm": 0.7161293029785156,
+      "learning_rate": 4.9887995624933137e-05,
+      "loss": 3.2289,
+      "step": 11580
+    },
+    {
+      "epoch": 0.0946229644677932,
+      "grad_norm": 0.6972730755805969,
+      "learning_rate": 4.988738873831078e-05,
+      "loss": 3.2247,
+      "step": 11610
+    },
+    {
+      "epoch": 0.0948674682519477,
+      "grad_norm": 0.7108049988746643,
+      "learning_rate": 4.988678021565623e-05,
+      "loss": 3.2227,
+      "step": 11640
+    },
+    {
+      "epoch": 0.09511197203610221,
+      "grad_norm": 0.729718029499054,
+      "learning_rate": 4.988617005700949e-05,
+      "loss": 3.2396,
+      "step": 11670
+    },
+    {
+      "epoch": 0.0953564758202567,
+      "grad_norm": 0.6947882771492004,
+      "learning_rate": 4.988555826241068e-05,
+      "loss": 3.238,
+      "step": 11700
+    },
+    {
+      "epoch": 0.09560097960441122,
+      "grad_norm": 0.7448226809501648,
+      "learning_rate": 4.988494483190001e-05,
+      "loss": 3.2274,
+      "step": 11730
+    },
+    {
+      "epoch": 0.09584548338856572,
+      "grad_norm": 0.6681617498397827,
+      "learning_rate": 4.98843297655178e-05,
+      "loss": 3.2504,
+      "step": 11760
+    },
+    {
+      "epoch": 0.09608998717272023,
+      "grad_norm": 0.7603419423103333,
+      "learning_rate": 4.988371306330449e-05,
+      "loss": 3.2206,
+      "step": 11790
+    },
+    {
+      "epoch": 0.09633449095687473,
+      "grad_norm": 0.7187109589576721,
+      "learning_rate": 4.9883094725300625e-05,
+      "loss": 3.2313,
+      "step": 11820
+    },
+    {
+      "epoch": 0.09657899474102924,
+      "grad_norm": 0.7375375628471375,
+      "learning_rate": 4.9882474751546846e-05,
+      "loss": 3.214,
+      "step": 11850
+    },
+    {
+      "epoch": 0.09682349852518374,
+      "grad_norm": 0.7366687059402466,
+      "learning_rate": 4.98818531420839e-05,
+      "loss": 3.2191,
+      "step": 11880
+    },
+    {
+      "epoch": 0.09706800230933824,
+      "grad_norm": 0.690647304058075,
+      "learning_rate": 4.988122989695267e-05,
+      "loss": 3.2378,
+      "step": 11910
+    },
+    {
+      "epoch": 0.09731250609349275,
+      "grad_norm": 0.7315085530281067,
+      "learning_rate": 4.9880605016194116e-05,
+      "loss": 3.2121,
+      "step": 11940
+    },
+    {
+      "epoch": 0.09755700987764725,
+      "grad_norm": 0.7334670424461365,
+      "learning_rate": 4.9879978499849304e-05,
+      "loss": 3.2194,
+      "step": 11970
+    },
+    {
+      "epoch": 0.09780151366180176,
+      "grad_norm": 0.7092124223709106,
+      "learning_rate": 4.987935034795944e-05,
+      "loss": 3.2162,
+      "step": 12000
+    },
+    {
+      "epoch": 0.09804601744595626,
+      "grad_norm": 0.7557071447372437,
+      "learning_rate": 4.9878720560565803e-05,
+      "loss": 3.21,
+      "step": 12030
+    },
+    {
+      "epoch": 0.09829052123011077,
+      "grad_norm": 0.7767185568809509,
+      "learning_rate": 4.9878089137709806e-05,
+      "loss": 3.2221,
+      "step": 12060
+    },
+    {
+      "epoch": 0.09853502501426527,
+      "grad_norm": 0.7004647850990295,
+      "learning_rate": 4.987745607943295e-05,
+      "loss": 3.2333,
+      "step": 12090
+    },
+    {
+      "epoch": 0.09877952879841977,
+      "grad_norm": 0.6746474504470825,
+      "learning_rate": 4.9876821385776854e-05,
+      "loss": 3.2185,
+      "step": 12120
+    },
+    {
+      "epoch": 0.09902403258257428,
+      "grad_norm": 0.7061980962753296,
+      "learning_rate": 4.9876185056783226e-05,
+      "loss": 3.2159,
+      "step": 12150
+    },
+    {
+      "epoch": 0.09926853636672878,
+      "grad_norm": 0.7409046292304993,
+      "learning_rate": 4.987554709249391e-05,
+      "loss": 3.2214,
+      "step": 12180
+    },
+    {
+      "epoch": 0.09951304015088329,
+      "grad_norm": 0.6951879858970642,
+      "learning_rate": 4.987490749295085e-05,
+      "loss": 3.2124,
+      "step": 12210
+    },
+    {
+      "epoch": 0.09975754393503779,
+      "grad_norm": 0.7459700107574463,
+      "learning_rate": 4.9874266258196084e-05,
+      "loss": 3.2074,
+      "step": 12240
+    },
+    {
+      "epoch": 0.1000020477191923,
+      "grad_norm": 0.654021680355072,
+      "learning_rate": 4.9873623388271764e-05,
+      "loss": 3.2199,
+      "step": 12270
+    },
+    {
+      "epoch": 0.1002465515033468,
+      "grad_norm": 0.7051617503166199,
+      "learning_rate": 4.987297888322015e-05,
+      "loss": 3.1998,
+      "step": 12300
+    },
+    {
+      "epoch": 0.1004910552875013,
+      "grad_norm": 0.7212511301040649,
+      "learning_rate": 4.9872332743083615e-05,
+      "loss": 3.2144,
+      "step": 12330
+    },
+    {
+      "epoch": 0.10073555907165581,
+      "grad_norm": 0.6988245844841003,
+      "learning_rate": 4.987168496790463e-05,
+      "loss": 3.2036,
+      "step": 12360
+    },
+    {
+      "epoch": 0.10098006285581031,
+      "grad_norm": 0.7479727864265442,
+      "learning_rate": 4.9871035557725774e-05,
+      "loss": 3.2035,
+      "step": 12390
+    },
+    {
+      "epoch": 0.10122456663996482,
+      "grad_norm": 0.6703733801841736,
+      "learning_rate": 4.987038451258975e-05,
+      "loss": 3.2119,
+      "step": 12420
+    },
+    {
+      "epoch": 0.10146907042411932,
+      "grad_norm": 0.7142158150672913,
+      "learning_rate": 4.9869731832539346e-05,
+      "loss": 3.1866,
+      "step": 12450
+    },
+    {
+      "epoch": 0.10171357420827383,
+      "grad_norm": 0.6590874195098877,
+      "learning_rate": 4.9869077517617474e-05,
+      "loss": 3.1854,
+      "step": 12480
+    },
+    {
+      "epoch": 0.10195807799242833,
+      "grad_norm": 0.7379269599914551,
+      "learning_rate": 4.986842156786714e-05,
+      "loss": 3.1755,
+      "step": 12510
+    },
+    {
+      "epoch": 0.10220258177658284,
+      "grad_norm": 0.7689487934112549,
+      "learning_rate": 4.9867763983331474e-05,
+      "loss": 3.1838,
+      "step": 12540
+    },
+    {
+      "epoch": 0.10244708556073734,
+      "grad_norm": 0.6958329677581787,
+      "learning_rate": 4.9867104764053697e-05,
+      "loss": 3.191,
+      "step": 12570
+    },
+    {
+      "epoch": 0.10269158934489184,
+      "grad_norm": 0.7052045464515686,
+      "learning_rate": 4.986644391007714e-05,
+      "loss": 3.1773,
+      "step": 12600
+    },
+    {
+      "epoch": 0.10293609312904635,
+      "grad_norm": 0.6961473226547241,
+      "learning_rate": 4.986578142144525e-05,
+      "loss": 3.1982,
+      "step": 12630
+    },
+    {
+      "epoch": 0.10318059691320085,
+      "grad_norm": 0.6776019930839539,
+      "learning_rate": 4.986511729820158e-05,
+      "loss": 3.1932,
+      "step": 12660
+    },
+    {
+      "epoch": 0.10342510069735536,
+      "grad_norm": 0.7005933523178101,
+      "learning_rate": 4.986445154038979e-05,
+      "loss": 3.1892,
+      "step": 12690
+    },
+    {
+      "epoch": 0.10366960448150986,
+      "grad_norm": 0.6863512992858887,
+      "learning_rate": 4.986378414805364e-05,
+      "loss": 3.199,
+      "step": 12720
+    },
+    {
+      "epoch": 0.10391410826566437,
+      "grad_norm": 0.7182413339614868,
+      "learning_rate": 4.9863115121237006e-05,
+      "loss": 3.178,
+      "step": 12750
+    },
+    {
+      "epoch": 0.10415861204981887,
+      "grad_norm": 0.7226789593696594,
+      "learning_rate": 4.986244445998386e-05,
+      "loss": 3.197,
+      "step": 12780
+    },
+    {
+      "epoch": 0.10440311583397337,
+      "grad_norm": 0.7314379215240479,
+      "learning_rate": 4.9861772164338304e-05,
+      "loss": 3.184,
+      "step": 12810
+    },
+    {
+      "epoch": 0.10464761961812788,
+      "grad_norm": 0.7286015152931213,
+      "learning_rate": 4.986109823434452e-05,
+      "loss": 3.1846,
+      "step": 12840
+    },
+    {
+      "epoch": 0.10489212340228238,
+      "grad_norm": 0.6875571608543396,
+      "learning_rate": 4.986042267004681e-05,
+      "loss": 3.1965,
+      "step": 12870
+    },
+    {
+      "epoch": 0.10513662718643689,
+      "grad_norm": 0.7378973364830017,
+      "learning_rate": 4.98597454714896e-05,
+      "loss": 3.1946,
+      "step": 12900
+    },
+    {
+      "epoch": 0.10538113097059139,
+      "grad_norm": 0.7709534764289856,
+      "learning_rate": 4.985906663871739e-05,
+      "loss": 3.195,
+      "step": 12930
+    },
+    {
+      "epoch": 0.1056256347547459,
+      "grad_norm": 0.6385577917098999,
+      "learning_rate": 4.9858386171774804e-05,
+      "loss": 3.1958,
+      "step": 12960
+    },
+    {
+      "epoch": 0.1058701385389004,
+      "grad_norm": 0.7177209258079529,
+      "learning_rate": 4.985770407070659e-05,
+      "loss": 3.1669,
+      "step": 12990
+    },
+    {
+      "epoch": 0.1061146423230549,
+      "grad_norm": 0.6712928414344788,
+      "learning_rate": 4.985702033555757e-05,
+      "loss": 3.193,
+      "step": 13020
+    },
+    {
+      "epoch": 0.10635914610720941,
+      "grad_norm": 0.670972466468811,
+      "learning_rate": 4.9856334966372705e-05,
+      "loss": 3.1788,
+      "step": 13050
+    },
+    {
+      "epoch": 0.10660364989136391,
+      "grad_norm": 0.7206739187240601,
+      "learning_rate": 4.985564796319705e-05,
+      "loss": 3.1928,
+      "step": 13080
+    },
+    {
+      "epoch": 0.10684815367551842,
+      "grad_norm": 0.7591316103935242,
+      "learning_rate": 4.985495932607576e-05,
+      "loss": 3.196,
+      "step": 13110
+    },
+    {
+      "epoch": 0.10709265745967292,
+      "grad_norm": 0.6864051222801208,
+      "learning_rate": 4.9854269055054096e-05,
+      "loss": 3.1929,
+      "step": 13140
+    },
+    {
+      "epoch": 0.10733716124382743,
+      "grad_norm": 0.6852788329124451,
+      "learning_rate": 4.985357715017744e-05,
+      "loss": 3.1824,
+      "step": 13170
+    },
+    {
+      "epoch": 0.10758166502798193,
+      "grad_norm": 0.6889312863349915,
+      "learning_rate": 4.985288361149129e-05,
+      "loss": 3.1918,
+      "step": 13200
+    },
+    {
+      "epoch": 0.10782616881213643,
+      "grad_norm": 0.724328875541687,
+      "learning_rate": 4.985218843904122e-05,
+      "loss": 3.1944,
+      "step": 13230
+    },
+    {
+      "epoch": 0.10807067259629094,
+      "grad_norm": 0.7326011061668396,
+      "learning_rate": 4.985149163287294e-05,
+      "loss": 3.1815,
+      "step": 13260
+    },
+    {
+      "epoch": 0.10831517638044544,
+      "grad_norm": 0.7652591466903687,
+      "learning_rate": 4.985079319303225e-05,
+      "loss": 3.1747,
+      "step": 13290
+    },
+    {
+      "epoch": 0.10855968016459995,
+      "grad_norm": 0.7179089188575745,
+      "learning_rate": 4.985009311956507e-05,
+      "loss": 3.1729,
+      "step": 13320
+    },
+    {
+      "epoch": 0.10880418394875445,
+      "grad_norm": 0.6816122531890869,
+      "learning_rate": 4.984939141251741e-05,
+      "loss": 3.1877,
+      "step": 13350
+    },
+    {
+      "epoch": 0.10904868773290896,
+      "grad_norm": 0.660858690738678,
+      "learning_rate": 4.9848688071935415e-05,
+      "loss": 3.1658,
+      "step": 13380
+    }
+  ],
+  "logging_steps": 30,
+  "max_steps": 368091,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 100,
+  "total_flos": 1.303917921042432e+19,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd09cf647966b7aa9734641f3c430c3c948068505081b7e0375830487e6a3632
+size 4984