Add files using upload-large-folder tool

config.json

@@ -0,0 +1,35 @@
+{
+  "_name_or_path": "/leonardo_work/EUHPC_A02_045/models/ltg_norbert3-base",
+  "architectures": [
+    "NorbertForSequenceClassification"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "auto_map": {
+    "AutoConfig": "configuration_norbert.NorbertConfig",
+    "AutoModel": "ltg/norbert3-base--modeling_norbert.NorbertModel",
+    "AutoModelForMaskedLM": "ltg/norbert3-base--modeling_norbert.NorbertForMaskedLM",
+    "AutoModelForMultipleChoice": "ltg/norbert3-base--modeling_norbert.NorbertForMultipleChoice",
+    "AutoModelForQuestionAnswering": "ltg/norbert3-base--modeling_norbert.NorbertForQuestionAnswering",
+    "AutoModelForSequenceClassification": "modeling_norbert.NorbertForSequenceClassification",
+    "AutoModelForTokenClassification": "ltg/norbert3-base--modeling_norbert.NorbertForTokenClassification"
+  },
+  "hidden_dropout_prob": 0.0,
+  "hidden_size": 768,
+  "id2label": {
+    "0": "LABEL_0"
+  },
+  "intermediate_size": 2048,
+  "label2id": {
+    "LABEL_0": 0
+  },
+  "layer_norm_eps": 1e-07,
+  "max_position_embeddings": 512,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "output_all_encoded_layers": true,
+  "position_bucket_size": 32,
+  "problem_type": "regression",
+  "torch_dtype": "float32",
+  "transformers_version": "4.44.2",
+  "vocab_size": 50000
+}

configuration_norbert.py

@@ -0,0 +1,34 @@
+from transformers.configuration_utils import PretrainedConfig
+
+
+class NorbertConfig(PretrainedConfig):
+    """Configuration class to store the configuration of a `NorbertModel`.
+    """
+    def __init__(
+        self,
+        vocab_size=50000,
+        attention_probs_dropout_prob=0.1,
+        hidden_dropout_prob=0.1,
+        hidden_size=768,
+        intermediate_size=2048,
+        max_position_embeddings=512,
+        position_bucket_size=32,
+        num_attention_heads=12,
+        num_hidden_layers=12,
+        layer_norm_eps=1.0e-7,
+        output_all_encoded_layers=True,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.intermediate_size = intermediate_size
+        self.hidden_dropout_prob = hidden_dropout_prob
+        self.attention_probs_dropout_prob = attention_probs_dropout_prob
+        self.max_position_embeddings = max_position_embeddings
+        self.output_all_encoded_layers = output_all_encoded_layers
+        self.position_bucket_size = position_bucket_size
+        self.layer_norm_eps = layer_norm_eps

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8219c3afc3f9d4c8eff51823d770f2ec916a96a845c9c3596e900dce45f56ed0
+size 496139908

modeling_norbert.py

@@ -0,0 +1,635 @@
+import math
+from typing import List, Optional, Tuple, Union
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from torch.utils import checkpoint
+
+from .configuration_norbert import NorbertConfig
+from transformers.modeling_utils import PreTrainedModel
+from transformers.activations import gelu_new
+from transformers.modeling_outputs import (
+    MaskedLMOutput,
+    MultipleChoiceModelOutput,
+    QuestionAnsweringModelOutput,
+    SequenceClassifierOutput,
+    TokenClassifierOutput,
+    BaseModelOutput
+)
+from transformers.pytorch_utils import softmax_backward_data
+
+
+class Encoder(nn.Module):
+    def __init__(self, config, activation_checkpointing=False):
+        super().__init__()
+        self.layers = nn.ModuleList([EncoderLayer(config) for _ in range(config.num_hidden_layers)])
+
+        for i, layer in enumerate(self.layers):
+            layer.mlp.mlp[1].weight.data *= math.sqrt(1.0 / (2.0 * (1 + i)))
+            layer.mlp.mlp[-2].weight.data *= math.sqrt(1.0 / (2.0 * (1 + i)))
+
+        self.activation_checkpointing = activation_checkpointing
+    
+    def forward(self, hidden_states, attention_mask, relative_embedding):
+        hidden_states, attention_probs = [hidden_states], []
+
+        for layer in self.layers:
+            if self.activation_checkpointing:
+                hidden_state, attention_p = checkpoint.checkpoint(layer, hidden_states[-1], attention_mask, relative_embedding)
+            else:
+                hidden_state, attention_p = layer(hidden_states[-1], attention_mask, relative_embedding)
+
+            hidden_states.append(hidden_state)
+            attention_probs.append(attention_p)
+
+        return hidden_states, attention_probs
+
+
+class MaskClassifier(nn.Module):
+    def __init__(self, config, subword_embedding):
+        super().__init__()
+        self.nonlinearity = nn.Sequential(
+            nn.LayerNorm(config.hidden_size, config.layer_norm_eps, elementwise_affine=False),
+            nn.Linear(config.hidden_size, config.hidden_size),
+            nn.GELU(),
+            nn.LayerNorm(config.hidden_size, config.layer_norm_eps, elementwise_affine=False),
+            nn.Dropout(config.hidden_dropout_prob),
+            nn.Linear(subword_embedding.size(1), subword_embedding.size(0))
+        )
+
+    def forward(self, x, masked_lm_labels=None):
+        if masked_lm_labels is not None:
+            x = torch.index_select(x.flatten(0, 1), 0, torch.nonzero(masked_lm_labels.flatten() != -100).squeeze())
+        x = self.nonlinearity(x)
+        return x
+
+
+class EncoderLayer(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.attention = Attention(config)
+        self.mlp = FeedForward(config)
+
+    def forward(self, x, padding_mask, relative_embedding):
+        attention_output, attention_probs = self.attention(x, padding_mask, relative_embedding)
+        x = x + attention_output
+        x = x + self.mlp(x)
+        return x, attention_probs
+
+
+class GeGLU(nn.Module):
+    def forward(self, x):
+        x, gate = x.chunk(2, dim=-1)
+        x = x * gelu_new(gate)
+        return x
+
+
+class FeedForward(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.mlp = nn.Sequential(
+            nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps, elementwise_affine=False),
+            nn.Linear(config.hidden_size, 2*config.intermediate_size, bias=False),
+            GeGLU(),
+            nn.LayerNorm(config.intermediate_size, eps=config.layer_norm_eps, elementwise_affine=False),
+            nn.Linear(config.intermediate_size, config.hidden_size, bias=False),
+            nn.Dropout(config.hidden_dropout_prob)
+        )
+
+    def forward(self, x):
+        return self.mlp(x)
+
+
+class MaskedSoftmax(torch.autograd.Function):
+    @staticmethod
+    def forward(self, x, mask, dim):
+        self.dim = dim
+        x.masked_fill_(mask, float('-inf'))
+        x = torch.softmax(x, self.dim)
+        x.masked_fill_(mask, 0.0)
+        self.save_for_backward(x)
+        return x
+
+    @staticmethod
+    def backward(self, grad_output):
+        output, = self.saved_tensors
+        input_grad = softmax_backward_data(self, grad_output, output, self.dim, output)
+        return input_grad, None, None
+
+
+class Attention(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+
+        self.config = config
+
+        if config.hidden_size % config.num_attention_heads != 0:
+            raise ValueError(f"The hidden size {config.hidden_size} is not a multiple of the number of attention heads {config.num_attention_heads}")
+
+        self.hidden_size = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_size = config.hidden_size // config.num_attention_heads
+
+        self.in_proj_qk = nn.Linear(config.hidden_size, 2*config.hidden_size, bias=True)
+        self.in_proj_v = nn.Linear(config.hidden_size, config.hidden_size, bias=True)
+        self.out_proj = nn.Linear(config.hidden_size, config.hidden_size, bias=True)
+
+        self.pre_layer_norm = nn.LayerNorm(config.hidden_size, config.layer_norm_eps, elementwise_affine=False)
+        self.post_layer_norm = nn.LayerNorm(config.hidden_size, config.layer_norm_eps, elementwise_affine=True)
+
+        position_indices = torch.arange(config.max_position_embeddings, dtype=torch.long).unsqueeze(1) \
+            - torch.arange(config.max_position_embeddings, dtype=torch.long).unsqueeze(0)
+        position_indices = self.make_log_bucket_position(position_indices, config.position_bucket_size, config.max_position_embeddings)
+        position_indices = config.position_bucket_size - 1 + position_indices
+        self.register_buffer("position_indices", position_indices, persistent=False)
+
+        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
+        self.scale = 1.0 / math.sqrt(3 * self.head_size)
+
+    def make_log_bucket_position(self, relative_pos, bucket_size, max_position):
+        sign = torch.sign(relative_pos)
+        mid = bucket_size // 2
+        abs_pos = torch.where((relative_pos < mid) & (relative_pos > -mid), mid - 1, torch.abs(relative_pos).clamp(max=max_position - 1))
+        log_pos = torch.ceil(torch.log(abs_pos / mid) / math.log((max_position-1) / mid) * (mid - 1)).int() + mid
+        bucket_pos = torch.where(abs_pos <= mid, relative_pos, log_pos * sign).long()
+        return bucket_pos
+
+    def compute_attention_scores(self, hidden_states, relative_embedding):
+        key_len, batch_size, _ = hidden_states.size()
+        query_len = key_len
+
+        if self.position_indices.size(0) < query_len:
+            position_indices = torch.arange(query_len, dtype=torch.long).unsqueeze(1) \
+                - torch.arange(query_len, dtype=torch.long).unsqueeze(0)
+            position_indices = self.make_log_bucket_position(position_indices, self.config.position_bucket_size, 512)
+            position_indices = self.config.position_bucket_size - 1 + position_indices
+            self.position_indices = position_indices.to(hidden_states.device)
+
+        hidden_states = self.pre_layer_norm(hidden_states)
+
+        query, key = self.in_proj_qk(hidden_states).chunk(2, dim=2)  # shape: [T, B, D]
+        value = self.in_proj_v(hidden_states)  # shape: [T, B, D]
+
+        query = query.reshape(query_len, batch_size * self.num_heads, self.head_size).transpose(0, 1)
+        key = key.reshape(key_len, batch_size * self.num_heads, self.head_size).transpose(0, 1)
+        value = value.view(key_len, batch_size * self.num_heads, self.head_size).transpose(0, 1)
+
+        attention_scores = torch.bmm(query, key.transpose(1, 2) * self.scale)
+
+        pos = self.in_proj_qk(self.dropout(relative_embedding))  # shape: [2T-1, 2D]
+        query_pos, key_pos = pos.view(-1, self.num_heads, 2*self.head_size).chunk(2, dim=2)
+        query = query.view(batch_size, self.num_heads, query_len, self.head_size)
+        key = key.view(batch_size, self.num_heads, query_len, self.head_size)
+
+        attention_c_p = torch.einsum("bhqd,khd->bhqk", query, key_pos.squeeze(1) * self.scale)
+        attention_p_c = torch.einsum("bhkd,qhd->bhqk", key * self.scale, query_pos.squeeze(1))
+
+        position_indices = self.position_indices[:query_len, :key_len].expand(batch_size, self.num_heads, -1, -1)
+        attention_c_p = attention_c_p.gather(3, position_indices)
+        attention_p_c = attention_p_c.gather(2, position_indices)
+
+        attention_scores = attention_scores.view(batch_size, self.num_heads, query_len, key_len)
+        attention_scores.add_(attention_c_p)
+        attention_scores.add_(attention_p_c)
+
+        return attention_scores, value
+
+    def compute_output(self, attention_probs, value):
+        attention_probs = self.dropout(attention_probs)
+        context = torch.bmm(attention_probs.flatten(0, 1), value)  # shape: [B*H, Q, D]
+        context = context.transpose(0, 1).reshape(context.size(1), -1, self.hidden_size)  # shape: [Q, B, H*D]
+        context = self.out_proj(context)
+        context = self.post_layer_norm(context)
+        context = self.dropout(context)
+        return context
+
+    def forward(self, hidden_states, attention_mask, relative_embedding):
+        attention_scores, value = self.compute_attention_scores(hidden_states, relative_embedding)
+        attention_probs = MaskedSoftmax.apply(attention_scores, attention_mask, -1)
+        return self.compute_output(attention_probs, value), attention_probs.detach()
+
+
+class Embedding(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+
+        self.word_embedding = nn.Embedding(config.vocab_size, config.hidden_size)
+        self.word_layer_norm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps, elementwise_affine=False)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+        self.relative_embedding = nn.Parameter(torch.empty(2 * config.position_bucket_size - 1, config.hidden_size))
+        self.relative_layer_norm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+
+    def forward(self, input_ids):
+        word_embedding = self.dropout(self.word_layer_norm(self.word_embedding(input_ids)))
+        relative_embeddings = self.relative_layer_norm(self.relative_embedding)
+        return word_embedding, relative_embeddings
+
+
+#
+# HuggingFace wrappers
+#
+
+class NorbertPreTrainedModel(PreTrainedModel):
+    config_class = NorbertConfig
+    base_model_prefix = "norbert3"
+    supports_gradient_checkpointing = True
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, Encoder):
+            module.activation_checkpointing = value
+
+    def _init_weights(self, module):
+        std = math.sqrt(2.0 / (5.0 * self.hidden_size))
+
+        if isinstance(module, nn.Linear):
+            nn.init.trunc_normal_(module.weight.data, mean=0.0, std=std, a=-2*std, b=2*std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            nn.init.trunc_normal_(module.weight.data, mean=0.0, std=std, a=-2*std, b=2*std)
+        elif isinstance(module, nn.LayerNorm):
+            module.bias.data.zero_()
+            module.weight.data.fill_(1.0)
+
+
+class NorbertModel(NorbertPreTrainedModel):
+    def __init__(self, config, add_mlm_layer=False, gradient_checkpointing=False, **kwargs):
+        super().__init__(config, **kwargs)
+        self.config = config
+        self.hidden_size = config.hidden_size
+
+        self.embedding = Embedding(config)
+        self.transformer = Encoder(config, activation_checkpointing=gradient_checkpointing)
+        self.classifier = MaskClassifier(config, self.embedding.word_embedding.weight) if add_mlm_layer else None
+
+    def get_input_embeddings(self):
+        return self.embedding.word_embedding
+
+    def set_input_embeddings(self, value):
+        self.embedding.word_embedding = value
+
+    def get_contextualized_embeddings(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None
+    ) -> List[torch.Tensor]:
+        if input_ids is not None:
+            input_shape = input_ids.size()
+        else:
+            raise ValueError("You have to specify input_ids")
+
+        batch_size, seq_length = input_shape
+        device = input_ids.device
+
+        if attention_mask is None:
+            attention_mask = torch.zeros(batch_size, seq_length, dtype=torch.bool, device=device)
+        else:
+            attention_mask = ~attention_mask.bool()
+        attention_mask = attention_mask.unsqueeze(1).unsqueeze(2)
+ 
+        static_embeddings, relative_embedding = self.embedding(input_ids.t())
+        contextualized_embeddings, attention_probs = self.transformer(static_embeddings, attention_mask, relative_embedding)
+        contextualized_embeddings = [e.transpose(0, 1) for e in contextualized_embeddings]
+        last_layer = contextualized_embeddings[-1]
+        contextualized_embeddings = [contextualized_embeddings[0]] + [
+            contextualized_embeddings[i] - contextualized_embeddings[i - 1]
+            for i in range(1, len(contextualized_embeddings))
+        ]
+        return last_layer, contextualized_embeddings, attention_probs
+
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        output_hidden_states: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        **kwargs
+    ) -> Union[Tuple[torch.Tensor], BaseModelOutput]:
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        sequence_output, contextualized_embeddings, attention_probs = self.get_contextualized_embeddings(input_ids, attention_mask)
+
+        if not return_dict:
+            return (
+                sequence_output,
+                *([contextualized_embeddings] if output_hidden_states else []),
+                *([attention_probs] if output_attentions else [])
+            )
+
+        return BaseModelOutput(
+            last_hidden_state=sequence_output,
+            hidden_states=contextualized_embeddings if output_hidden_states else None,
+            attentions=attention_probs if output_attentions else None
+        )
+
+
+class NorbertForMaskedLM(NorbertModel):
+    _keys_to_ignore_on_load_unexpected = ["head"]
+
+    def __init__(self, config, **kwargs):
+        super().__init__(config, add_mlm_layer=True, **kwargs)
+
+    def get_output_embeddings(self):
+        return self.classifier.nonlinearity[-1].weight
+
+    def set_output_embeddings(self, new_embeddings):
+        self.classifier.nonlinearity[-1].weight = new_embeddings
+
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        output_hidden_states: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        labels: Optional[torch.LongTensor] = None,
+        **kwargs
+    ) -> Union[Tuple[torch.Tensor], MaskedLMOutput]:
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        sequence_output, contextualized_embeddings, attention_probs = self.get_contextualized_embeddings(input_ids, attention_mask)
+        subword_prediction = self.classifier(sequence_output)
+        subword_prediction[:, :, :106+1] = float("-inf")
+
+        masked_lm_loss = None
+        if labels is not None:
+            masked_lm_loss = F.cross_entropy(subword_prediction.flatten(0, 1), labels.flatten())
+
+        if not return_dict:
+            output = (
+                subword_prediction,
+                *([contextualized_embeddings] if output_hidden_states else []),
+                *([attention_probs] if output_attentions else [])
+            )
+            return ((masked_lm_loss,) + output) if masked_lm_loss is not None else output
+
+        return MaskedLMOutput(
+            loss=masked_lm_loss,
+            logits=subword_prediction,
+            hidden_states=contextualized_embeddings if output_hidden_states else None,
+            attentions=attention_probs if output_attentions else None
+        )
+
+
+class Classifier(nn.Module):
+    def __init__(self, config, num_labels: int):
+        super().__init__()
+
+        drop_out = getattr(config, "cls_dropout", None)
+        drop_out = config.hidden_dropout_prob if drop_out is None else drop_out
+
+        self.nonlinearity = nn.Sequential(
+            nn.LayerNorm(config.hidden_size, config.layer_norm_eps, elementwise_affine=False),
+            nn.Linear(config.hidden_size, config.hidden_size),
+            nn.GELU(),
+            nn.LayerNorm(config.hidden_size, config.layer_norm_eps, elementwise_affine=False),
+            nn.Dropout(drop_out),
+            nn.Linear(config.hidden_size, num_labels)
+        )
+
+    def forward(self, x):
+        x = self.nonlinearity(x)
+        return x
+
+
+class NorbertForSequenceClassification(NorbertModel):
+    _keys_to_ignore_on_load_unexpected = ["classifier"]
+    _keys_to_ignore_on_load_missing = ["head"]
+
+    def __init__(self, config, **kwargs):
+        super().__init__(config, add_mlm_layer=False, **kwargs)
+
+        self.num_labels = config.num_labels
+        self.head = Classifier(config, self.num_labels)
+
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        labels: Optional[torch.LongTensor] = None,
+        **kwargs
+    ) -> Union[Tuple[torch.Tensor], SequenceClassifierOutput]:
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        sequence_output, contextualized_embeddings, attention_probs = self.get_contextualized_embeddings(input_ids, attention_mask)
+        logits = self.head(sequence_output[:, 0, :])
+
+        loss = None
+        if labels is not None:
+            if self.config.problem_type is None:
+                if self.num_labels == 1:
+                    self.config.problem_type = "regression"
+                elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
+                    self.config.problem_type = "single_label_classification"
+                else:
+                    self.config.problem_type = "multi_label_classification"
+
+            if self.config.problem_type == "regression":
+                loss_fct = nn.MSELoss()
+                if self.num_labels == 1:
+                    loss = loss_fct(logits.squeeze(), labels.squeeze())
+                else:
+                    loss = loss_fct(logits, labels)
+            elif self.config.problem_type == "single_label_classification":
+                loss_fct = nn.CrossEntropyLoss()
+                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
+            elif self.config.problem_type == "multi_label_classification":
+                loss_fct = nn.BCEWithLogitsLoss()
+                loss = loss_fct(logits, labels)
+
+        if not return_dict:
+            output = (
+                logits,
+                *([contextualized_embeddings] if output_hidden_states else []),
+                *([attention_probs] if output_attentions else [])
+            )
+            return ((loss,) + output) if loss is not None else output
+
+        return SequenceClassifierOutput(
+            loss=loss,
+            logits=logits,
+            hidden_states=contextualized_embeddings if output_hidden_states else None,
+            attentions=attention_probs if output_attentions else None
+        )
+
+
+class NorbertForTokenClassification(NorbertModel):
+    _keys_to_ignore_on_load_unexpected = ["classifier"]
+    _keys_to_ignore_on_load_missing = ["head"]
+
+    def __init__(self, config, **kwargs):
+        super().__init__(config, add_mlm_layer=False, **kwargs)
+
+        self.num_labels = config.num_labels
+        self.head = Classifier(config, self.num_labels)
+
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        labels: Optional[torch.LongTensor] = None,
+        **kwargs
+    ) -> Union[Tuple[torch.Tensor], TokenClassifierOutput]:
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        sequence_output, contextualized_embeddings, attention_probs = self.get_contextualized_embeddings(input_ids, attention_mask)
+        logits = self.head(sequence_output)
+
+        loss = None
+        if labels is not None:
+            loss_fct = nn.CrossEntropyLoss()
+            loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
+
+        if not return_dict:
+            output = (
+                logits,
+                *([contextualized_embeddings] if output_hidden_states else []),
+                *([attention_probs] if output_attentions else [])
+            )
+            return ((loss,) + output) if loss is not None else output
+
+        return TokenClassifierOutput(
+            loss=loss,
+            logits=logits,
+            hidden_states=contextualized_embeddings if output_hidden_states else None,
+            attentions=attention_probs if output_attentions else None
+        )
+
+
+class NorbertForQuestionAnswering(NorbertModel):
+    _keys_to_ignore_on_load_unexpected = ["classifier"]
+    _keys_to_ignore_on_load_missing = ["head"]
+
+    def __init__(self, config, **kwargs):
+        super().__init__(config, add_mlm_layer=False, **kwargs)
+
+        self.num_labels = config.num_labels
+        self.head = Classifier(config, self.num_labels)
+
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        start_positions: Optional[torch.Tensor] = None,
+        end_positions: Optional[torch.Tensor] = None,
+        **kwargs
+    ) -> Union[Tuple[torch.Tensor], QuestionAnsweringModelOutput]:
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        sequence_output, contextualized_embeddings, attention_probs = self.get_contextualized_embeddings(input_ids, attention_mask)
+        logits = self.head(sequence_output)
+
+        start_logits, end_logits = logits.split(1, dim=-1)
+        start_logits = start_logits.squeeze(-1).contiguous()
+        end_logits = end_logits.squeeze(-1).contiguous()
+
+        total_loss = None
+        if start_positions is not None and end_positions is not None:
+            # If we are on multi-GPU, split add a dimension
+            if len(start_positions.size()) > 1:
+                start_positions = start_positions.squeeze(-1)
+            if len(end_positions.size()) > 1:
+                end_positions = end_positions.squeeze(-1)
+
+            # sometimes the start/end positions are outside our model inputs, we ignore these terms
+            ignored_index = start_logits.size(1)
+            start_positions = start_positions.clamp(0, ignored_index)
+            end_positions = end_positions.clamp(0, ignored_index)
+
+            loss_fct = nn.CrossEntropyLoss(ignore_index=ignored_index)
+            start_loss = loss_fct(start_logits, start_positions)
+            end_loss = loss_fct(end_logits, end_positions)
+            total_loss = (start_loss + end_loss) / 2
+
+        if not return_dict:
+            output = (
+                start_logits,
+                end_logits,
+                *([contextualized_embeddings] if output_hidden_states else []),
+                *([attention_probs] if output_attentions else [])
+            )
+            return ((total_loss,) + output) if total_loss is not None else output
+
+        return QuestionAnsweringModelOutput(
+            loss=total_loss,
+            start_logits=start_logits,
+            end_logits=end_logits,
+            hidden_states=contextualized_embeddings if output_hidden_states else None,
+            attentions=attention_probs if output_attentions else None
+        )
+
+
+class NorbertForMultipleChoice(NorbertModel):
+    _keys_to_ignore_on_load_unexpected = ["classifier"]
+    _keys_to_ignore_on_load_missing = ["head"]
+
+    def __init__(self, config, **kwargs):
+        super().__init__(config, add_mlm_layer=False, **kwargs)
+
+        self.num_labels = getattr(config, "num_labels", 2)
+        self.head = Classifier(config, self.num_labels)
+
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        labels: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        **kwargs
+    ) -> Union[Tuple[torch.Tensor], MultipleChoiceModelOutput]:
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        num_choices = input_ids.shape[1]
+
+        flat_input_ids = input_ids.view(-1, input_ids.size(-1))
+        flat_attention_mask = attention_mask.view(-1, attention_mask.size(-1)) if attention_mask is not None else None
+
+        sequence_output, contextualized_embeddings, attention_probs = self.get_contextualized_embeddings(flat_input_ids, flat_attention_mask)
+        logits = self.head(sequence_output)
+        reshaped_logits = logits.view(-1, num_choices)
+
+        loss = None
+        if labels is not None:
+            loss_fct = nn.CrossEntropyLoss()
+            loss = loss_fct(reshaped_logits, labels)
+
+        if not return_dict:
+            output = (
+                reshaped_logits,
+                *([contextualized_embeddings] if output_hidden_states else []),
+                *([attention_probs] if output_attentions else [])
+            )
+            return ((loss,) + output) if loss is not None else output
+
+        return MultipleChoiceModelOutput(
+            loss=loss,
+            logits=reshaped_logits,
+            hidden_states=contextualized_embeddings if output_hidden_states else None,
+            attentions=attention_probs if output_attentions else None
+        )

optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:144912b41c99ce11c9ff41ade8ce11e6432b0eedfae30a55a60152bdfb7a5c5d
+size 4735250

rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e78c2b2dfd67dd86ab81cb66717a46e57aba66c9b9e57f20766946e1e48703d2
+size 14180

scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a20310f80bd8c93b2778c2b9d92d7e31c911403d1bad2087e2913900421f86d1
+size 1064

special_tokens_map.json

@@ -0,0 +1,51 @@
+{
+  "bos_token": {
+    "content": "[BOS]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "cls_token": {
+    "content": "[CLS]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "[EOS]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "mask_token": {
+    "content": "[MASK]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "[PAD]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "sep_token": {
+    "content": "[SEP]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "[UNK]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer_config.json

@@ -0,0 +1,870 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "[UNK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "[CLS]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "[SEP]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "3": {
+      "content": "[PAD]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "4": {
+      "content": "[MASK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "5": {
+      "content": "[BOS]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "6": {
+      "content": "[EOS]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "7": {
+      "content": "[MASK_0]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "8": {
+      "content": "[MASK_1]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "9": {
+      "content": "[MASK_2]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "10": {
+      "content": "[MASK_3]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "11": {
+      "content": "[MASK_4]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "12": {
+      "content": "[MASK_5]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "13": {
+      "content": "[MASK_6]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "14": {
+      "content": "[MASK_7]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "15": {
+      "content": "[MASK_8]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "16": {
+      "content": "[MASK_9]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "17": {
+      "content": "[MASK_10]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "18": {
+      "content": "[MASK_11]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "19": {
+      "content": "[MASK_12]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "20": {
+      "content": "[MASK_13]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "21": {
+      "content": "[MASK_14]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "22": {
+      "content": "[MASK_15]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "23": {
+      "content": "[MASK_16]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "24": {
+      "content": "[MASK_17]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "25": {
+      "content": "[MASK_18]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "26": {
+      "content": "[MASK_19]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "27": {
+      "content": "[MASK_20]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "28": {
+      "content": "[MASK_21]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "29": {
+      "content": "[MASK_22]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "30": {
+      "content": "[MASK_23]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "31": {
+      "content": "[MASK_24]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "32": {
+      "content": "[MASK_25]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "33": {
+      "content": "[MASK_26]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "34": {
+      "content": "[MASK_27]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "35": {
+      "content": "[MASK_28]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "36": {
+      "content": "[MASK_29]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "37": {
+      "content": "[MASK_30]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "38": {
+      "content": "[MASK_31]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "39": {
+      "content": "[MASK_32]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "40": {
+      "content": "[MASK_33]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "41": {
+      "content": "[MASK_34]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "42": {
+      "content": "[MASK_35]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "43": {
+      "content": "[MASK_36]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "44": {
+      "content": "[MASK_37]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "45": {
+      "content": "[MASK_38]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "46": {
+      "content": "[MASK_39]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "47": {
+      "content": "[MASK_40]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "48": {
+      "content": "[MASK_41]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "49": {
+      "content": "[MASK_42]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "50": {
+      "content": "[MASK_43]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "51": {
+      "content": "[MASK_44]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "52": {
+      "content": "[MASK_45]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "53": {
+      "content": "[MASK_46]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "54": {
+      "content": "[MASK_47]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "55": {
+      "content": "[MASK_48]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "56": {
+      "content": "[MASK_49]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "57": {
+      "content": "[MASK_50]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "58": {
+      "content": "[MASK_51]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "59": {
+      "content": "[MASK_52]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "60": {
+      "content": "[MASK_53]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "61": {
+      "content": "[MASK_54]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "62": {
+      "content": "[MASK_55]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "63": {
+      "content": "[MASK_56]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "64": {
+      "content": "[MASK_57]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "65": {
+      "content": "[MASK_58]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "66": {
+      "content": "[MASK_59]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "67": {
+      "content": "[MASK_60]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "68": {
+      "content": "[MASK_61]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "69": {
+      "content": "[MASK_62]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "70": {
+      "content": "[MASK_63]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "71": {
+      "content": "[MASK_64]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "72": {
+      "content": "[MASK_65]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "73": {
+      "content": "[MASK_66]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "74": {
+      "content": "[MASK_67]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "75": {
+      "content": "[MASK_68]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "76": {
+      "content": "[MASK_69]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "77": {
+      "content": "[MASK_70]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "78": {
+      "content": "[MASK_71]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "79": {
+      "content": "[MASK_72]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "80": {
+      "content": "[MASK_73]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "81": {
+      "content": "[MASK_74]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "82": {
+      "content": "[MASK_75]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "83": {
+      "content": "[MASK_76]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "84": {
+      "content": "[MASK_77]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "85": {
+      "content": "[MASK_78]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "86": {
+      "content": "[MASK_79]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "87": {
+      "content": "[MASK_80]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "88": {
+      "content": "[MASK_81]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "89": {
+      "content": "[MASK_82]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "90": {
+      "content": "[MASK_83]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "91": {
+      "content": "[MASK_84]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "92": {
+      "content": "[MASK_85]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "93": {
+      "content": "[MASK_86]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "94": {
+      "content": "[MASK_87]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "95": {
+      "content": "[MASK_88]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "96": {
+      "content": "[MASK_89]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "97": {
+      "content": "[MASK_90]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "98": {
+      "content": "[MASK_91]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "99": {
+      "content": "[MASK_92]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "100": {
+      "content": "[MASK_93]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "101": {
+      "content": "[MASK_94]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "102": {
+      "content": "[MASK_95]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "103": {
+      "content": "[MASK_96]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "104": {
+      "content": "[MASK_97]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "105": {
+      "content": "[MASK_98]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "106": {
+      "content": "[MASK_99]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "[BOS]",
+  "clean_up_tokenization_spaces": true,
+  "cls_token": "[CLS]",
+  "eos_token": "[EOS]",
+  "mask_token": "[MASK]",
+  "model_max_length": 512,
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "tokenizer_class": "PreTrainedTokenizerFast",
+  "unk_token": "[UNK]"
+}

trainer_state.json

@@ -0,0 +1,2833 @@
+{
+  "best_metric": 0.47654790915623974,
+  "best_model_checkpoint": "/leonardo_work/EUHPC_A02_045/scandinavian-lm/robin/fw-classifier-checkpoints-no-70b/checkpoint-33000",
+  "epoch": 19.976498237367803,
+  "eval_steps": 1000,
+  "global_step": 34000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0,
+      "eval_accuracy": 0.6101936630634727,
+      "eval_f1_macro": 0.12631889713244165,
+      "eval_loss": 4.240240097045898,
+      "eval_precision": 0.1016989438439121,
+      "eval_recall": 0.16666666666666666,
+      "eval_runtime": 571.773,
+      "eval_samples_per_second": 84.619,
+      "eval_steps_per_second": 0.661,
+      "step": 0
+    },
+    {
+      "epoch": 0.05875440658049354,
+      "grad_norm": 4.152867794036865,
+      "learning_rate": 0.00029911868390129255,
+      "loss": 0.5817,
+      "step": 100
+    },
+    {
+      "epoch": 0.11750881316098707,
+      "grad_norm": 8.71384334564209,
+      "learning_rate": 0.0002982373678025852,
+      "loss": 0.3984,
+      "step": 200
+    },
+    {
+      "epoch": 0.1762632197414806,
+      "grad_norm": 13.754929542541504,
+      "learning_rate": 0.00029735605170387776,
+      "loss": 0.3821,
+      "step": 300
+    },
+    {
+      "epoch": 0.23501762632197415,
+      "grad_norm": 0.5506241917610168,
+      "learning_rate": 0.0002964747356051704,
+      "loss": 0.3869,
+      "step": 400
+    },
+    {
+      "epoch": 0.2937720329024677,
+      "grad_norm": 7.549045562744141,
+      "learning_rate": 0.00029559341950646296,
+      "loss": 0.346,
+      "step": 500
+    },
+    {
+      "epoch": 0.3525264394829612,
+      "grad_norm": 8.952414512634277,
+      "learning_rate": 0.00029471210340775554,
+      "loss": 0.3535,
+      "step": 600
+    },
+    {
+      "epoch": 0.4112808460634548,
+      "grad_norm": 23.95926856994629,
+      "learning_rate": 0.00029383078730904817,
+      "loss": 0.3638,
+      "step": 700
+    },
+    {
+      "epoch": 0.4700352526439483,
+      "grad_norm": 5.874898910522461,
+      "learning_rate": 0.00029294947121034074,
+      "loss": 0.3458,
+      "step": 800
+    },
+    {
+      "epoch": 0.5287896592244419,
+      "grad_norm": 1.2818429470062256,
+      "learning_rate": 0.0002920681551116333,
+      "loss": 0.3183,
+      "step": 900
+    },
+    {
+      "epoch": 0.5875440658049353,
+      "grad_norm": 11.916783332824707,
+      "learning_rate": 0.00029118683901292595,
+      "loss": 0.3415,
+      "step": 1000
+    },
+    {
+      "epoch": 0.5875440658049353,
+      "eval_accuracy": 0.7214517495814645,
+      "eval_f1_macro": 0.36894750243003943,
+      "eval_loss": 0.28744974732398987,
+      "eval_precision": 0.4977336949727908,
+      "eval_recall": 0.35626936254459146,
+      "eval_runtime": 546.9073,
+      "eval_samples_per_second": 88.467,
+      "eval_steps_per_second": 0.691,
+      "step": 1000
+    },
+    {
+      "epoch": 0.6462984723854289,
+      "grad_norm": 9.043974876403809,
+      "learning_rate": 0.0002903055229142185,
+      "loss": 0.3187,
+      "step": 1100
+    },
+    {
+      "epoch": 0.7050528789659224,
+      "grad_norm": 4.782003879547119,
+      "learning_rate": 0.00028942420681551115,
+      "loss": 0.3466,
+      "step": 1200
+    },
+    {
+      "epoch": 0.763807285546416,
+      "grad_norm": 18.947124481201172,
+      "learning_rate": 0.00028854289071680373,
+      "loss": 0.3365,
+      "step": 1300
+    },
+    {
+      "epoch": 0.8225616921269095,
+      "grad_norm": 3.7607452869415283,
+      "learning_rate": 0.00028766157461809636,
+      "loss": 0.3307,
+      "step": 1400
+    },
+    {
+      "epoch": 0.881316098707403,
+      "grad_norm": 7.354115009307861,
+      "learning_rate": 0.00028678025851938894,
+      "loss": 0.327,
+      "step": 1500
+    },
+    {
+      "epoch": 0.9400705052878966,
+      "grad_norm": 12.593878746032715,
+      "learning_rate": 0.00028589894242068157,
+      "loss": 0.3143,
+      "step": 1600
+    },
+    {
+      "epoch": 0.9988249118683902,
+      "grad_norm": 10.552128791809082,
+      "learning_rate": 0.00028501762632197414,
+      "loss": 0.3082,
+      "step": 1700
+    },
+    {
+      "epoch": 1.0575793184488838,
+      "grad_norm": 16.90986442565918,
+      "learning_rate": 0.0002841363102232667,
+      "loss": 0.3153,
+      "step": 1800
+    },
+    {
+      "epoch": 1.1163337250293772,
+      "grad_norm": 5.408417224884033,
+      "learning_rate": 0.00028325499412455935,
+      "loss": 0.3051,
+      "step": 1900
+    },
+    {
+      "epoch": 1.1750881316098707,
+      "grad_norm": 10.757403373718262,
+      "learning_rate": 0.0002823736780258519,
+      "loss": 0.2999,
+      "step": 2000
+    },
+    {
+      "epoch": 1.1750881316098707,
+      "eval_accuracy": 0.6296013062439286,
+      "eval_f1_macro": 0.39274972019560317,
+      "eval_loss": 0.34983837604522705,
+      "eval_precision": 0.4577861635371156,
+      "eval_recall": 0.40350488729829226,
+      "eval_runtime": 573.3711,
+      "eval_samples_per_second": 84.383,
+      "eval_steps_per_second": 0.659,
+      "step": 2000
+    },
+    {
+      "epoch": 1.2338425381903644,
+      "grad_norm": 1.7697181701660156,
+      "learning_rate": 0.0002814923619271445,
+      "loss": 0.3158,
+      "step": 2100
+    },
+    {
+      "epoch": 1.2925969447708578,
+      "grad_norm": 9.139455795288086,
+      "learning_rate": 0.0002806110458284371,
+      "loss": 0.2964,
+      "step": 2200
+    },
+    {
+      "epoch": 1.3513513513513513,
+      "grad_norm": 1.4484944343566895,
+      "learning_rate": 0.0002797297297297297,
+      "loss": 0.296,
+      "step": 2300
+    },
+    {
+      "epoch": 1.410105757931845,
+      "grad_norm": 1.3425700664520264,
+      "learning_rate": 0.0002788484136310223,
+      "loss": 0.2906,
+      "step": 2400
+    },
+    {
+      "epoch": 1.4688601645123385,
+      "grad_norm": 9.240585327148438,
+      "learning_rate": 0.0002779670975323149,
+      "loss": 0.2901,
+      "step": 2500
+    },
+    {
+      "epoch": 1.527614571092832,
+      "grad_norm": 8.707176208496094,
+      "learning_rate": 0.0002770857814336075,
+      "loss": 0.2903,
+      "step": 2600
+    },
+    {
+      "epoch": 1.5863689776733256,
+      "grad_norm": 3.509387969970703,
+      "learning_rate": 0.0002762044653349001,
+      "loss": 0.2811,
+      "step": 2700
+    },
+    {
+      "epoch": 1.6451233842538189,
+      "grad_norm": 3.0751891136169434,
+      "learning_rate": 0.0002753231492361927,
+      "loss": 0.2759,
+      "step": 2800
+    },
+    {
+      "epoch": 1.7038777908343126,
+      "grad_norm": 4.9634013175964355,
+      "learning_rate": 0.0002744418331374853,
+      "loss": 0.2842,
+      "step": 2900
+    },
+    {
+      "epoch": 1.7626321974148063,
+      "grad_norm": 10.635833740234375,
+      "learning_rate": 0.0002735605170387779,
+      "loss": 0.2909,
+      "step": 3000
+    },
+    {
+      "epoch": 1.7626321974148063,
+      "eval_accuracy": 0.7319306367939152,
+      "eval_f1_macro": 0.3785908786929529,
+      "eval_loss": 0.26183947920799255,
+      "eval_precision": 0.5055778316715768,
+      "eval_recall": 0.3727759846099492,
+      "eval_runtime": 546.7019,
+      "eval_samples_per_second": 88.5,
+      "eval_steps_per_second": 0.691,
+      "step": 3000
+    },
+    {
+      "epoch": 1.8213866039952995,
+      "grad_norm": 10.297226905822754,
+      "learning_rate": 0.0002726792009400705,
+      "loss": 0.2826,
+      "step": 3100
+    },
+    {
+      "epoch": 1.8801410105757932,
+      "grad_norm": 2.603403329849243,
+      "learning_rate": 0.0002717978848413631,
+      "loss": 0.2787,
+      "step": 3200
+    },
+    {
+      "epoch": 1.9388954171562869,
+      "grad_norm": 12.185776710510254,
+      "learning_rate": 0.0002709165687426557,
+      "loss": 0.276,
+      "step": 3300
+    },
+    {
+      "epoch": 1.9976498237367801,
+      "grad_norm": 4.290465354919434,
+      "learning_rate": 0.00027003525264394825,
+      "loss": 0.2846,
+      "step": 3400
+    },
+    {
+      "epoch": 2.056404230317274,
+      "grad_norm": 3.0092501640319824,
+      "learning_rate": 0.00026915393654524083,
+      "loss": 0.2805,
+      "step": 3500
+    },
+    {
+      "epoch": 2.1151586368977675,
+      "grad_norm": 4.7245893478393555,
+      "learning_rate": 0.00026827262044653346,
+      "loss": 0.27,
+      "step": 3600
+    },
+    {
+      "epoch": 2.1739130434782608,
+      "grad_norm": 9.71957778930664,
+      "learning_rate": 0.00026739130434782604,
+      "loss": 0.2722,
+      "step": 3700
+    },
+    {
+      "epoch": 2.2326674500587544,
+      "grad_norm": 2.2740237712860107,
+      "learning_rate": 0.00026650998824911867,
+      "loss": 0.2663,
+      "step": 3800
+    },
+    {
+      "epoch": 2.291421856639248,
+      "grad_norm": 1.9909628629684448,
+      "learning_rate": 0.00026562867215041124,
+      "loss": 0.266,
+      "step": 3900
+    },
+    {
+      "epoch": 2.3501762632197414,
+      "grad_norm": 5.9694366455078125,
+      "learning_rate": 0.00026474735605170387,
+      "loss": 0.2724,
+      "step": 4000
+    },
+    {
+      "epoch": 2.3501762632197414,
+      "eval_accuracy": 0.7637393299299341,
+      "eval_f1_macro": 0.4012224284754011,
+      "eval_loss": 0.24033646285533905,
+      "eval_precision": 0.5199015574271643,
+      "eval_recall": 0.37999768266709827,
+      "eval_runtime": 549.807,
+      "eval_samples_per_second": 88.0,
+      "eval_steps_per_second": 0.688,
+      "step": 4000
+    },
+    {
+      "epoch": 2.408930669800235,
+      "grad_norm": 0.8608851432800293,
+      "learning_rate": 0.00026386603995299645,
+      "loss": 0.2665,
+      "step": 4100
+    },
+    {
+      "epoch": 2.4676850763807288,
+      "grad_norm": 3.54764723777771,
+      "learning_rate": 0.0002629847238542891,
+      "loss": 0.262,
+      "step": 4200
+    },
+    {
+      "epoch": 2.526439482961222,
+      "grad_norm": 4.2886481285095215,
+      "learning_rate": 0.00026210340775558165,
+      "loss": 0.2651,
+      "step": 4300
+    },
+    {
+      "epoch": 2.5851938895417157,
+      "grad_norm": 9.533616065979004,
+      "learning_rate": 0.00026122209165687423,
+      "loss": 0.2705,
+      "step": 4400
+    },
+    {
+      "epoch": 2.6439482961222094,
+      "grad_norm": 1.6619293689727783,
+      "learning_rate": 0.00026034077555816686,
+      "loss": 0.2728,
+      "step": 4500
+    },
+    {
+      "epoch": 2.7027027027027026,
+      "grad_norm": 5.174167633056641,
+      "learning_rate": 0.00025945945945945944,
+      "loss": 0.265,
+      "step": 4600
+    },
+    {
+      "epoch": 2.7614571092831963,
+      "grad_norm": 3.4489777088165283,
+      "learning_rate": 0.000258578143360752,
+      "loss": 0.2608,
+      "step": 4700
+    },
+    {
+      "epoch": 2.82021151586369,
+      "grad_norm": 5.9784111976623535,
+      "learning_rate": 0.00025769682726204464,
+      "loss": 0.268,
+      "step": 4800
+    },
+    {
+      "epoch": 2.8789659224441833,
+      "grad_norm": 1.8021718263626099,
+      "learning_rate": 0.0002568155111633372,
+      "loss": 0.2604,
+      "step": 4900
+    },
+    {
+      "epoch": 2.937720329024677,
+      "grad_norm": 1.6711304187774658,
+      "learning_rate": 0.0002559341950646298,
+      "loss": 0.2673,
+      "step": 5000
+    },
+    {
+      "epoch": 2.937720329024677,
+      "eval_accuracy": 0.7662815451708245,
+      "eval_f1_macro": 0.42256323386502537,
+      "eval_loss": 0.23298443853855133,
+      "eval_precision": 0.4985552431790052,
+      "eval_recall": 0.4017059313674283,
+      "eval_runtime": 572.543,
+      "eval_samples_per_second": 84.505,
+      "eval_steps_per_second": 0.66,
+      "step": 5000
+    },
+    {
+      "epoch": 2.9964747356051706,
+      "grad_norm": 0.876846432685852,
+      "learning_rate": 0.0002550528789659224,
+      "loss": 0.2559,
+      "step": 5100
+    },
+    {
+      "epoch": 3.055229142185664,
+      "grad_norm": 2.291898250579834,
+      "learning_rate": 0.000254171562867215,
+      "loss": 0.2614,
+      "step": 5200
+    },
+    {
+      "epoch": 3.1139835487661576,
+      "grad_norm": 2.2618095874786377,
+      "learning_rate": 0.00025329024676850763,
+      "loss": 0.2578,
+      "step": 5300
+    },
+    {
+      "epoch": 3.172737955346651,
+      "grad_norm": 2.6534600257873535,
+      "learning_rate": 0.0002524089306698002,
+      "loss": 0.2568,
+      "step": 5400
+    },
+    {
+      "epoch": 3.2314923619271445,
+      "grad_norm": 3.1308279037475586,
+      "learning_rate": 0.00025152761457109283,
+      "loss": 0.2531,
+      "step": 5500
+    },
+    {
+      "epoch": 3.290246768507638,
+      "grad_norm": 2.781928300857544,
+      "learning_rate": 0.0002506462984723854,
+      "loss": 0.2526,
+      "step": 5600
+    },
+    {
+      "epoch": 3.3490011750881314,
+      "grad_norm": 3.065544366836548,
+      "learning_rate": 0.00024976498237367804,
+      "loss": 0.2654,
+      "step": 5700
+    },
+    {
+      "epoch": 3.407755581668625,
+      "grad_norm": 1.8798550367355347,
+      "learning_rate": 0.0002488836662749706,
+      "loss": 0.2504,
+      "step": 5800
+    },
+    {
+      "epoch": 3.466509988249119,
+      "grad_norm": 6.618080139160156,
+      "learning_rate": 0.0002480023501762632,
+      "loss": 0.2593,
+      "step": 5900
+    },
+    {
+      "epoch": 3.525264394829612,
+      "grad_norm": 3.1250927448272705,
+      "learning_rate": 0.00024712103407755577,
+      "loss": 0.2592,
+      "step": 6000
+    },
+    {
+      "epoch": 3.525264394829612,
+      "eval_accuracy": 0.7667155819192691,
+      "eval_f1_macro": 0.38853739645942603,
+      "eval_loss": 0.24237428605556488,
+      "eval_precision": 0.5286680026993323,
+      "eval_recall": 0.36739233287894374,
+      "eval_runtime": 551.8425,
+      "eval_samples_per_second": 87.675,
+      "eval_steps_per_second": 0.685,
+      "step": 6000
+    },
+    {
+      "epoch": 3.5840188014101058,
+      "grad_norm": 1.4983100891113281,
+      "learning_rate": 0.0002462397179788484,
+      "loss": 0.2492,
+      "step": 6100
+    },
+    {
+      "epoch": 3.6427732079905994,
+      "grad_norm": 7.058569431304932,
+      "learning_rate": 0.000245358401880141,
+      "loss": 0.2594,
+      "step": 6200
+    },
+    {
+      "epoch": 3.7015276145710927,
+      "grad_norm": 1.7073079347610474,
+      "learning_rate": 0.0002444770857814336,
+      "loss": 0.259,
+      "step": 6300
+    },
+    {
+      "epoch": 3.7602820211515864,
+      "grad_norm": 3.5612850189208984,
+      "learning_rate": 0.00024359576968272618,
+      "loss": 0.2499,
+      "step": 6400
+    },
+    {
+      "epoch": 3.8190364277320796,
+      "grad_norm": 3.4439921379089355,
+      "learning_rate": 0.00024271445358401875,
+      "loss": 0.2478,
+      "step": 6500
+    },
+    {
+      "epoch": 3.8777908343125733,
+      "grad_norm": 1.5288629531860352,
+      "learning_rate": 0.00024183313748531138,
+      "loss": 0.251,
+      "step": 6600
+    },
+    {
+      "epoch": 3.936545240893067,
+      "grad_norm": 4.820594787597656,
+      "learning_rate": 0.00024095182138660396,
+      "loss": 0.2556,
+      "step": 6700
+    },
+    {
+      "epoch": 3.9952996474735603,
+      "grad_norm": 2.041408061981201,
+      "learning_rate": 0.00024007050528789656,
+      "loss": 0.2559,
+      "step": 6800
+    },
+    {
+      "epoch": 4.054054054054054,
+      "grad_norm": 7.40335750579834,
+      "learning_rate": 0.00023918918918918917,
+      "loss": 0.252,
+      "step": 6900
+    },
+    {
+      "epoch": 4.112808460634548,
+      "grad_norm": 2.86159086227417,
+      "learning_rate": 0.00023830787309048177,
+      "loss": 0.2504,
+      "step": 7000
+    },
+    {
+      "epoch": 4.112808460634548,
+      "eval_accuracy": 0.7713866440691979,
+      "eval_f1_macro": 0.4170037789256203,
+      "eval_loss": 0.23320935666561127,
+      "eval_precision": 0.520598590788087,
+      "eval_recall": 0.39262170670987384,
+      "eval_runtime": 570.908,
+      "eval_samples_per_second": 84.747,
+      "eval_steps_per_second": 0.662,
+      "step": 7000
+    },
+    {
+      "epoch": 4.171562867215041,
+      "grad_norm": 4.3792805671691895,
+      "learning_rate": 0.00023742655699177434,
+      "loss": 0.2511,
+      "step": 7100
+    },
+    {
+      "epoch": 4.230317273795535,
+      "grad_norm": 1.2670930624008179,
+      "learning_rate": 0.00023654524089306697,
+      "loss": 0.2467,
+      "step": 7200
+    },
+    {
+      "epoch": 4.289071680376028,
+      "grad_norm": 4.641327381134033,
+      "learning_rate": 0.00023566392479435955,
+      "loss": 0.252,
+      "step": 7300
+    },
+    {
+      "epoch": 4.3478260869565215,
+      "grad_norm": 2.356194257736206,
+      "learning_rate": 0.00023478260869565215,
+      "loss": 0.2471,
+      "step": 7400
+    },
+    {
+      "epoch": 4.406580493537016,
+      "grad_norm": 3.5696866512298584,
+      "learning_rate": 0.00023390129259694476,
+      "loss": 0.2444,
+      "step": 7500
+    },
+    {
+      "epoch": 4.465334900117509,
+      "grad_norm": 8.092639923095703,
+      "learning_rate": 0.00023301997649823736,
+      "loss": 0.2458,
+      "step": 7600
+    },
+    {
+      "epoch": 4.524089306698002,
+      "grad_norm": 3.4449079036712646,
+      "learning_rate": 0.00023213866039952993,
+      "loss": 0.2512,
+      "step": 7700
+    },
+    {
+      "epoch": 4.582843713278496,
+      "grad_norm": 5.513228416442871,
+      "learning_rate": 0.00023125734430082256,
+      "loss": 0.2518,
+      "step": 7800
+    },
+    {
+      "epoch": 4.6415981198589895,
+      "grad_norm": 2.5132598876953125,
+      "learning_rate": 0.00023037602820211514,
+      "loss": 0.2519,
+      "step": 7900
+    },
+    {
+      "epoch": 4.700352526439483,
+      "grad_norm": 2.164031982421875,
+      "learning_rate": 0.00022949471210340774,
+      "loss": 0.2455,
+      "step": 8000
+    },
+    {
+      "epoch": 4.700352526439483,
+      "eval_accuracy": 0.7628505880164521,
+      "eval_f1_macro": 0.4485850425849249,
+      "eval_loss": 0.2332322597503662,
+      "eval_precision": 0.484949663618545,
+      "eval_recall": 0.4351442261907779,
+      "eval_runtime": 569.2371,
+      "eval_samples_per_second": 84.996,
+      "eval_steps_per_second": 0.664,
+      "step": 8000
+    },
+    {
+      "epoch": 4.759106933019977,
+      "grad_norm": 8.147943496704102,
+      "learning_rate": 0.00022861339600470035,
+      "loss": 0.2538,
+      "step": 8100
+    },
+    {
+      "epoch": 4.81786133960047,
+      "grad_norm": 3.1226038932800293,
+      "learning_rate": 0.00022773207990599292,
+      "loss": 0.2481,
+      "step": 8200
+    },
+    {
+      "epoch": 4.876615746180963,
+      "grad_norm": 1.5910353660583496,
+      "learning_rate": 0.00022685076380728553,
+      "loss": 0.2432,
+      "step": 8300
+    },
+    {
+      "epoch": 4.9353701527614575,
+      "grad_norm": 2.3687844276428223,
+      "learning_rate": 0.0002259694477085781,
+      "loss": 0.2464,
+      "step": 8400
+    },
+    {
+      "epoch": 4.994124559341951,
+      "grad_norm": 1.5202206373214722,
+      "learning_rate": 0.00022508813160987073,
+      "loss": 0.2491,
+      "step": 8500
+    },
+    {
+      "epoch": 5.052878965922444,
+      "grad_norm": 8.09229850769043,
+      "learning_rate": 0.0002242068155111633,
+      "loss": 0.245,
+      "step": 8600
+    },
+    {
+      "epoch": 5.111633372502938,
+      "grad_norm": 4.977721691131592,
+      "learning_rate": 0.0002233254994124559,
+      "loss": 0.2414,
+      "step": 8700
+    },
+    {
+      "epoch": 5.170387779083431,
+      "grad_norm": 2.690870523452759,
+      "learning_rate": 0.0002224441833137485,
+      "loss": 0.2437,
+      "step": 8800
+    },
+    {
+      "epoch": 5.229142185663925,
+      "grad_norm": 4.5524373054504395,
+      "learning_rate": 0.00022156286721504112,
+      "loss": 0.2476,
+      "step": 8900
+    },
+    {
+      "epoch": 5.287896592244419,
+      "grad_norm": 3.5273966789245605,
+      "learning_rate": 0.0002206815511163337,
+      "loss": 0.2424,
+      "step": 9000
+    },
+    {
+      "epoch": 5.287896592244419,
+      "eval_accuracy": 0.7704152284893454,
+      "eval_f1_macro": 0.40212235960322845,
+      "eval_loss": 0.22698020935058594,
+      "eval_precision": 0.5605989028870036,
+      "eval_recall": 0.3866441363925465,
+      "eval_runtime": 588.8368,
+      "eval_samples_per_second": 82.167,
+      "eval_steps_per_second": 0.642,
+      "step": 9000
+    },
+    {
+      "epoch": 5.346650998824912,
+      "grad_norm": 4.150397777557373,
+      "learning_rate": 0.00021980023501762632,
+      "loss": 0.2476,
+      "step": 9100
+    },
+    {
+      "epoch": 5.405405405405405,
+      "grad_norm": 1.5151199102401733,
+      "learning_rate": 0.0002189189189189189,
+      "loss": 0.2458,
+      "step": 9200
+    },
+    {
+      "epoch": 5.464159811985899,
+      "grad_norm": 2.2463040351867676,
+      "learning_rate": 0.0002180376028202115,
+      "loss": 0.2388,
+      "step": 9300
+    },
+    {
+      "epoch": 5.522914218566393,
+      "grad_norm": 2.767045259475708,
+      "learning_rate": 0.0002171562867215041,
+      "loss": 0.2433,
+      "step": 9400
+    },
+    {
+      "epoch": 5.581668625146886,
+      "grad_norm": 5.879153728485107,
+      "learning_rate": 0.0002162749706227967,
+      "loss": 0.2452,
+      "step": 9500
+    },
+    {
+      "epoch": 5.64042303172738,
+      "grad_norm": 4.529464244842529,
+      "learning_rate": 0.00021539365452408928,
+      "loss": 0.2437,
+      "step": 9600
+    },
+    {
+      "epoch": 5.699177438307873,
+      "grad_norm": 2.579648017883301,
+      "learning_rate": 0.00021451233842538186,
+      "loss": 0.2429,
+      "step": 9700
+    },
+    {
+      "epoch": 5.7579318448883665,
+      "grad_norm": 1.4765149354934692,
+      "learning_rate": 0.0002136310223266745,
+      "loss": 0.2347,
+      "step": 9800
+    },
+    {
+      "epoch": 5.816686251468861,
+      "grad_norm": 6.136841297149658,
+      "learning_rate": 0.00021274970622796706,
+      "loss": 0.2407,
+      "step": 9900
+    },
+    {
+      "epoch": 5.875440658049354,
+      "grad_norm": 5.470715045928955,
+      "learning_rate": 0.0002118683901292597,
+      "loss": 0.2476,
+      "step": 10000
+    },
+    {
+      "epoch": 5.875440658049354,
+      "eval_accuracy": 0.7542525267139284,
+      "eval_f1_macro": 0.4479732740452163,
+      "eval_loss": 0.2347133606672287,
+      "eval_precision": 0.495252873527299,
+      "eval_recall": 0.4323174862666415,
+      "eval_runtime": 555.2826,
+      "eval_samples_per_second": 87.132,
+      "eval_steps_per_second": 0.681,
+      "step": 10000
+    },
+    {
+      "epoch": 5.934195064629847,
+      "grad_norm": 2.95473575592041,
+      "learning_rate": 0.00021098707403055227,
+      "loss": 0.2443,
+      "step": 10100
+    },
+    {
+      "epoch": 5.992949471210341,
+      "grad_norm": 1.8928413391113281,
+      "learning_rate": 0.00021010575793184487,
+      "loss": 0.237,
+      "step": 10200
+    },
+    {
+      "epoch": 6.0517038777908345,
+      "grad_norm": 5.004413604736328,
+      "learning_rate": 0.00020922444183313745,
+      "loss": 0.2419,
+      "step": 10300
+    },
+    {
+      "epoch": 6.110458284371328,
+      "grad_norm": 2.2294819355010986,
+      "learning_rate": 0.00020834312573443008,
+      "loss": 0.2393,
+      "step": 10400
+    },
+    {
+      "epoch": 6.169212690951821,
+      "grad_norm": 3.4622702598571777,
+      "learning_rate": 0.00020746180963572265,
+      "loss": 0.2467,
+      "step": 10500
+    },
+    {
+      "epoch": 6.227967097532315,
+      "grad_norm": 11.164566993713379,
+      "learning_rate": 0.00020658049353701526,
+      "loss": 0.2405,
+      "step": 10600
+    },
+    {
+      "epoch": 6.286721504112808,
+      "grad_norm": 8.36145305633545,
+      "learning_rate": 0.00020569917743830786,
+      "loss": 0.2366,
+      "step": 10700
+    },
+    {
+      "epoch": 6.345475910693302,
+      "grad_norm": 2.6593246459960938,
+      "learning_rate": 0.00020481786133960046,
+      "loss": 0.244,
+      "step": 10800
+    },
+    {
+      "epoch": 6.404230317273796,
+      "grad_norm": 1.6066139936447144,
+      "learning_rate": 0.00020393654524089304,
+      "loss": 0.2412,
+      "step": 10900
+    },
+    {
+      "epoch": 6.462984723854289,
+      "grad_norm": 2.962965250015259,
+      "learning_rate": 0.00020305522914218567,
+      "loss": 0.2385,
+      "step": 11000
+    },
+    {
+      "epoch": 6.462984723854289,
+      "eval_accuracy": 0.7729161068970506,
+      "eval_f1_macro": 0.4305720986934003,
+      "eval_loss": 0.22273238003253937,
+      "eval_precision": 0.5147561915779448,
+      "eval_recall": 0.40734695218979783,
+      "eval_runtime": 531.4648,
+      "eval_samples_per_second": 91.037,
+      "eval_steps_per_second": 0.711,
+      "step": 11000
+    },
+    {
+      "epoch": 6.521739130434782,
+      "grad_norm": 3.933185338973999,
+      "learning_rate": 0.00020217391304347824,
+      "loss": 0.2389,
+      "step": 11100
+    },
+    {
+      "epoch": 6.580493537015276,
+      "grad_norm": 6.2505316734313965,
+      "learning_rate": 0.00020129259694477085,
+      "loss": 0.2345,
+      "step": 11200
+    },
+    {
+      "epoch": 6.63924794359577,
+      "grad_norm": 2.8899261951446533,
+      "learning_rate": 0.00020041128084606345,
+      "loss": 0.2404,
+      "step": 11300
+    },
+    {
+      "epoch": 6.698002350176263,
+      "grad_norm": 4.886023998260498,
+      "learning_rate": 0.00019952996474735602,
+      "loss": 0.2405,
+      "step": 11400
+    },
+    {
+      "epoch": 6.756756756756757,
+      "grad_norm": 3.6124258041381836,
+      "learning_rate": 0.00019864864864864863,
+      "loss": 0.2388,
+      "step": 11500
+    },
+    {
+      "epoch": 6.81551116333725,
+      "grad_norm": 2.6905336380004883,
+      "learning_rate": 0.0001977673325499412,
+      "loss": 0.2419,
+      "step": 11600
+    },
+    {
+      "epoch": 6.8742655699177435,
+      "grad_norm": 1.7078518867492676,
+      "learning_rate": 0.00019688601645123383,
+      "loss": 0.2351,
+      "step": 11700
+    },
+    {
+      "epoch": 6.933019976498238,
+      "grad_norm": 4.933712482452393,
+      "learning_rate": 0.0001960047003525264,
+      "loss": 0.2357,
+      "step": 11800
+    },
+    {
+      "epoch": 6.991774383078731,
+      "grad_norm": 4.086423873901367,
+      "learning_rate": 0.00019512338425381904,
+      "loss": 0.241,
+      "step": 11900
+    },
+    {
+      "epoch": 7.050528789659224,
+      "grad_norm": 3.3847010135650635,
+      "learning_rate": 0.00019424206815511161,
+      "loss": 0.2343,
+      "step": 12000
+    },
+    {
+      "epoch": 7.050528789659224,
+      "eval_accuracy": 0.7735568278114213,
+      "eval_f1_macro": 0.4611187564793586,
+      "eval_loss": 0.2231457531452179,
+      "eval_precision": 0.48558443620097314,
+      "eval_recall": 0.4472545142062536,
+      "eval_runtime": 576.3861,
+      "eval_samples_per_second": 83.942,
+      "eval_steps_per_second": 0.656,
+      "step": 12000
+    },
+    {
+      "epoch": 7.109283196239718,
+      "grad_norm": 2.0805375576019287,
+      "learning_rate": 0.00019336075205640422,
+      "loss": 0.238,
+      "step": 12100
+    },
+    {
+      "epoch": 7.1680376028202115,
+      "grad_norm": 3.5111265182495117,
+      "learning_rate": 0.0001924794359576968,
+      "loss": 0.2365,
+      "step": 12200
+    },
+    {
+      "epoch": 7.226792009400705,
+      "grad_norm": 2.2142751216888428,
+      "learning_rate": 0.00019159811985898942,
+      "loss": 0.2387,
+      "step": 12300
+    },
+    {
+      "epoch": 7.285546415981199,
+      "grad_norm": 5.468302249908447,
+      "learning_rate": 0.000190716803760282,
+      "loss": 0.2366,
+      "step": 12400
+    },
+    {
+      "epoch": 7.344300822561692,
+      "grad_norm": 4.9833598136901855,
+      "learning_rate": 0.0001898354876615746,
+      "loss": 0.2345,
+      "step": 12500
+    },
+    {
+      "epoch": 7.403055229142185,
+      "grad_norm": 2.4710216522216797,
+      "learning_rate": 0.0001889541715628672,
+      "loss": 0.2335,
+      "step": 12600
+    },
+    {
+      "epoch": 7.4618096357226795,
+      "grad_norm": 1.4311057329177856,
+      "learning_rate": 0.0001880728554641598,
+      "loss": 0.242,
+      "step": 12700
+    },
+    {
+      "epoch": 7.520564042303173,
+      "grad_norm": 3.9087047576904297,
+      "learning_rate": 0.00018719153936545238,
+      "loss": 0.2369,
+      "step": 12800
+    },
+    {
+      "epoch": 7.579318448883666,
+      "grad_norm": 2.0385680198669434,
+      "learning_rate": 0.000186310223266745,
+      "loss": 0.2295,
+      "step": 12900
+    },
+    {
+      "epoch": 7.63807285546416,
+      "grad_norm": 3.9989728927612305,
+      "learning_rate": 0.0001854289071680376,
+      "loss": 0.2353,
+      "step": 13000
+    },
+    {
+      "epoch": 7.63807285546416,
+      "eval_accuracy": 0.7637393299299341,
+      "eval_f1_macro": 0.42822074709027813,
+      "eval_loss": 0.2291877716779709,
+      "eval_precision": 0.5240563218793618,
+      "eval_recall": 0.4106384213710441,
+      "eval_runtime": 580.7226,
+      "eval_samples_per_second": 83.315,
+      "eval_steps_per_second": 0.651,
+      "step": 13000
+    },
+    {
+      "epoch": 7.696827262044653,
+      "grad_norm": 1.630708932876587,
+      "learning_rate": 0.00018454759106933017,
+      "loss": 0.2373,
+      "step": 13100
+    },
+    {
+      "epoch": 7.755581668625147,
+      "grad_norm": 3.2567617893218994,
+      "learning_rate": 0.0001836662749706228,
+      "loss": 0.2326,
+      "step": 13200
+    },
+    {
+      "epoch": 7.814336075205641,
+      "grad_norm": 2.3300867080688477,
+      "learning_rate": 0.00018278495887191537,
+      "loss": 0.2369,
+      "step": 13300
+    },
+    {
+      "epoch": 7.873090481786134,
+      "grad_norm": 2.068678379058838,
+      "learning_rate": 0.00018190364277320797,
+      "loss": 0.2337,
+      "step": 13400
+    },
+    {
+      "epoch": 7.931844888366627,
+      "grad_norm": 2.0448477268218994,
+      "learning_rate": 0.00018102232667450055,
+      "loss": 0.2335,
+      "step": 13500
+    },
+    {
+      "epoch": 7.990599294947121,
+      "grad_norm": 2.7080137729644775,
+      "learning_rate": 0.00018014101057579318,
+      "loss": 0.238,
+      "step": 13600
+    },
+    {
+      "epoch": 8.049353701527615,
+      "grad_norm": 1.9964938163757324,
+      "learning_rate": 0.00017925969447708576,
+      "loss": 0.2359,
+      "step": 13700
+    },
+    {
+      "epoch": 8.108108108108109,
+      "grad_norm": 2.795433759689331,
+      "learning_rate": 0.00017837837837837839,
+      "loss": 0.2374,
+      "step": 13800
+    },
+    {
+      "epoch": 8.166862514688601,
+      "grad_norm": 2.685382843017578,
+      "learning_rate": 0.00017749706227967096,
+      "loss": 0.2364,
+      "step": 13900
+    },
+    {
+      "epoch": 8.225616921269095,
+      "grad_norm": 2.214505195617676,
+      "learning_rate": 0.00017661574618096356,
+      "loss": 0.2328,
+      "step": 14000
+    },
+    {
+      "epoch": 8.225616921269095,
+      "eval_accuracy": 0.7689477709112705,
+      "eval_f1_macro": 0.4571007537767639,
+      "eval_loss": 0.22145947813987732,
+      "eval_precision": 0.4969408215399489,
+      "eval_recall": 0.44354312726305817,
+      "eval_runtime": 570.5829,
+      "eval_samples_per_second": 84.796,
+      "eval_steps_per_second": 0.662,
+      "step": 14000
+    },
+    {
+      "epoch": 8.28437132784959,
+      "grad_norm": 3.0027620792388916,
+      "learning_rate": 0.00017573443008225614,
+      "loss": 0.2328,
+      "step": 14100
+    },
+    {
+      "epoch": 8.343125734430082,
+      "grad_norm": 1.9569076299667358,
+      "learning_rate": 0.00017485311398354877,
+      "loss": 0.2315,
+      "step": 14200
+    },
+    {
+      "epoch": 8.401880141010576,
+      "grad_norm": 2.1613545417785645,
+      "learning_rate": 0.00017397179788484135,
+      "loss": 0.2314,
+      "step": 14300
+    },
+    {
+      "epoch": 8.46063454759107,
+      "grad_norm": 4.501012802124023,
+      "learning_rate": 0.00017309048178613395,
+      "loss": 0.2307,
+      "step": 14400
+    },
+    {
+      "epoch": 8.519388954171562,
+      "grad_norm": 4.018213272094727,
+      "learning_rate": 0.00017220916568742655,
+      "loss": 0.2337,
+      "step": 14500
+    },
+    {
+      "epoch": 8.578143360752057,
+      "grad_norm": 3.4571876525878906,
+      "learning_rate": 0.00017132784958871913,
+      "loss": 0.2309,
+      "step": 14600
+    },
+    {
+      "epoch": 8.63689776733255,
+      "grad_norm": 1.6010338068008423,
+      "learning_rate": 0.00017044653349001173,
+      "loss": 0.2304,
+      "step": 14700
+    },
+    {
+      "epoch": 8.695652173913043,
+      "grad_norm": 5.177122592926025,
+      "learning_rate": 0.00016956521739130433,
+      "loss": 0.2315,
+      "step": 14800
+    },
+    {
+      "epoch": 8.754406580493537,
+      "grad_norm": 1.3421051502227783,
+      "learning_rate": 0.00016868390129259694,
+      "loss": 0.2358,
+      "step": 14900
+    },
+    {
+      "epoch": 8.813160987074031,
+      "grad_norm": 5.40761137008667,
+      "learning_rate": 0.0001678025851938895,
+      "loss": 0.2297,
+      "step": 15000
+    },
+    {
+      "epoch": 8.813160987074031,
+      "eval_accuracy": 0.7784759109604613,
+      "eval_f1_macro": 0.42357466614962985,
+      "eval_loss": 0.22359371185302734,
+      "eval_precision": 0.5285253073603687,
+      "eval_recall": 0.3968963203390616,
+      "eval_runtime": 558.7605,
+      "eval_samples_per_second": 86.59,
+      "eval_steps_per_second": 0.676,
+      "step": 15000
+    },
+    {
+      "epoch": 8.871915393654524,
+      "grad_norm": 2.718010902404785,
+      "learning_rate": 0.00016692126909518214,
+      "loss": 0.2282,
+      "step": 15100
+    },
+    {
+      "epoch": 8.930669800235018,
+      "grad_norm": 2.1908445358276367,
+      "learning_rate": 0.00016603995299647472,
+      "loss": 0.2342,
+      "step": 15200
+    },
+    {
+      "epoch": 8.989424206815512,
+      "grad_norm": 1.3827928304672241,
+      "learning_rate": 0.00016515863689776732,
+      "loss": 0.2336,
+      "step": 15300
+    },
+    {
+      "epoch": 9.048178613396004,
+      "grad_norm": 2.2856316566467285,
+      "learning_rate": 0.0001642773207990599,
+      "loss": 0.2316,
+      "step": 15400
+    },
+    {
+      "epoch": 9.106933019976498,
+      "grad_norm": 9.475784301757812,
+      "learning_rate": 0.00016339600470035253,
+      "loss": 0.2338,
+      "step": 15500
+    },
+    {
+      "epoch": 9.165687426556993,
+      "grad_norm": 5.561756610870361,
+      "learning_rate": 0.0001625146886016451,
+      "loss": 0.2345,
+      "step": 15600
+    },
+    {
+      "epoch": 9.224441833137485,
+      "grad_norm": 3.0887973308563232,
+      "learning_rate": 0.00016163337250293773,
+      "loss": 0.2377,
+      "step": 15700
+    },
+    {
+      "epoch": 9.283196239717979,
+      "grad_norm": 2.1840600967407227,
+      "learning_rate": 0.0001607520564042303,
+      "loss": 0.2279,
+      "step": 15800
+    },
+    {
+      "epoch": 9.341950646298473,
+      "grad_norm": 1.5278443098068237,
+      "learning_rate": 0.0001598707403055229,
+      "loss": 0.2281,
+      "step": 15900
+    },
+    {
+      "epoch": 9.400705052878966,
+      "grad_norm": 1.8652377128601074,
+      "learning_rate": 0.00015898942420681549,
+      "loss": 0.2261,
+      "step": 16000
+    },
+    {
+      "epoch": 9.400705052878966,
+      "eval_accuracy": 0.778496579377054,
+      "eval_f1_macro": 0.4118529460883977,
+      "eval_loss": 0.2205253690481186,
+      "eval_precision": 0.5265129533046812,
+      "eval_recall": 0.38989546728234464,
+      "eval_runtime": 564.1949,
+      "eval_samples_per_second": 85.756,
+      "eval_steps_per_second": 0.67,
+      "step": 16000
+    },
+    {
+      "epoch": 9.45945945945946,
+      "grad_norm": 3.021077871322632,
+      "learning_rate": 0.0001581081081081081,
+      "loss": 0.2291,
+      "step": 16100
+    },
+    {
+      "epoch": 9.518213866039954,
+      "grad_norm": 1.2995972633361816,
+      "learning_rate": 0.0001572267920094007,
+      "loss": 0.2279,
+      "step": 16200
+    },
+    {
+      "epoch": 9.576968272620446,
+      "grad_norm": 4.3413801193237305,
+      "learning_rate": 0.00015634547591069327,
+      "loss": 0.2254,
+      "step": 16300
+    },
+    {
+      "epoch": 9.63572267920094,
+      "grad_norm": 1.8537020683288574,
+      "learning_rate": 0.0001554641598119859,
+      "loss": 0.2284,
+      "step": 16400
+    },
+    {
+      "epoch": 9.694477085781434,
+      "grad_norm": 2.3501524925231934,
+      "learning_rate": 0.00015458284371327847,
+      "loss": 0.2315,
+      "step": 16500
+    },
+    {
+      "epoch": 9.753231492361927,
+      "grad_norm": 4.062187671661377,
+      "learning_rate": 0.00015370152761457108,
+      "loss": 0.2271,
+      "step": 16600
+    },
+    {
+      "epoch": 9.811985898942421,
+      "grad_norm": 2.9765398502349854,
+      "learning_rate": 0.00015282021151586368,
+      "loss": 0.2341,
+      "step": 16700
+    },
+    {
+      "epoch": 9.870740305522915,
+      "grad_norm": 3.3737270832061768,
+      "learning_rate": 0.00015193889541715628,
+      "loss": 0.2302,
+      "step": 16800
+    },
+    {
+      "epoch": 9.929494712103407,
+      "grad_norm": 3.7637851238250732,
+      "learning_rate": 0.00015105757931844886,
+      "loss": 0.2306,
+      "step": 16900
+    },
+    {
+      "epoch": 9.988249118683902,
+      "grad_norm": 4.947080135345459,
+      "learning_rate": 0.0001501762632197415,
+      "loss": 0.2226,
+      "step": 17000
+    },
+    {
+      "epoch": 9.988249118683902,
+      "eval_accuracy": 0.7804600789533513,
+      "eval_f1_macro": 0.45841854146967403,
+      "eval_loss": 0.21761466562747955,
+      "eval_precision": 0.4921930948461919,
+      "eval_recall": 0.43836252511254314,
+      "eval_runtime": 558.7853,
+      "eval_samples_per_second": 86.586,
+      "eval_steps_per_second": 0.676,
+      "step": 17000
+    },
+    {
+      "epoch": 10.047003525264396,
+      "grad_norm": 4.100146293640137,
+      "learning_rate": 0.00014929494712103406,
+      "loss": 0.2301,
+      "step": 17100
+    },
+    {
+      "epoch": 10.105757931844888,
+      "grad_norm": 2.0130274295806885,
+      "learning_rate": 0.00014841363102232667,
+      "loss": 0.2288,
+      "step": 17200
+    },
+    {
+      "epoch": 10.164512338425382,
+      "grad_norm": 2.4523582458496094,
+      "learning_rate": 0.00014753231492361924,
+      "loss": 0.2257,
+      "step": 17300
+    },
+    {
+      "epoch": 10.223266745005876,
+      "grad_norm": 2.4732425212860107,
+      "learning_rate": 0.00014665099882491185,
+      "loss": 0.2253,
+      "step": 17400
+    },
+    {
+      "epoch": 10.282021151586369,
+      "grad_norm": 2.8159022331237793,
+      "learning_rate": 0.00014576968272620445,
+      "loss": 0.2297,
+      "step": 17500
+    },
+    {
+      "epoch": 10.340775558166863,
+      "grad_norm": 2.784027338027954,
+      "learning_rate": 0.00014488836662749705,
+      "loss": 0.2282,
+      "step": 17600
+    },
+    {
+      "epoch": 10.399529964747355,
+      "grad_norm": 2.8521196842193604,
+      "learning_rate": 0.00014400705052878965,
+      "loss": 0.2294,
+      "step": 17700
+    },
+    {
+      "epoch": 10.45828437132785,
+      "grad_norm": 3.453033685684204,
+      "learning_rate": 0.00014312573443008226,
+      "loss": 0.2252,
+      "step": 17800
+    },
+    {
+      "epoch": 10.517038777908343,
+      "grad_norm": 1.889672875404358,
+      "learning_rate": 0.00014224441833137483,
+      "loss": 0.2266,
+      "step": 17900
+    },
+    {
+      "epoch": 10.575793184488838,
+      "grad_norm": 2.153575897216797,
+      "learning_rate": 0.00014136310223266744,
+      "loss": 0.2248,
+      "step": 18000
+    },
+    {
+      "epoch": 10.575793184488838,
+      "eval_accuracy": 0.778909947708906,
+      "eval_f1_macro": 0.44892131069383673,
+      "eval_loss": 0.2148253470659256,
+      "eval_precision": 0.4964295361361209,
+      "eval_recall": 0.42785352086722733,
+      "eval_runtime": 556.2941,
+      "eval_samples_per_second": 86.974,
+      "eval_steps_per_second": 0.679,
+      "step": 18000
+    },
+    {
+      "epoch": 10.63454759106933,
+      "grad_norm": 2.2354812622070312,
+      "learning_rate": 0.00014048178613396004,
+      "loss": 0.2303,
+      "step": 18100
+    },
+    {
+      "epoch": 10.693301997649824,
+      "grad_norm": 6.267688751220703,
+      "learning_rate": 0.00013960047003525264,
+      "loss": 0.2219,
+      "step": 18200
+    },
+    {
+      "epoch": 10.752056404230316,
+      "grad_norm": 4.277271270751953,
+      "learning_rate": 0.00013871915393654524,
+      "loss": 0.2256,
+      "step": 18300
+    },
+    {
+      "epoch": 10.81081081081081,
+      "grad_norm": 4.068231582641602,
+      "learning_rate": 0.00013783783783783782,
+      "loss": 0.2278,
+      "step": 18400
+    },
+    {
+      "epoch": 10.869565217391305,
+      "grad_norm": 1.9653985500335693,
+      "learning_rate": 0.00013695652173913042,
+      "loss": 0.2326,
+      "step": 18500
+    },
+    {
+      "epoch": 10.928319623971799,
+      "grad_norm": 1.7713191509246826,
+      "learning_rate": 0.00013607520564042303,
+      "loss": 0.2312,
+      "step": 18600
+    },
+    {
+      "epoch": 10.987074030552291,
+      "grad_norm": 3.856257438659668,
+      "learning_rate": 0.0001351938895417156,
+      "loss": 0.2313,
+      "step": 18700
+    },
+    {
+      "epoch": 11.045828437132785,
+      "grad_norm": 3.794623851776123,
+      "learning_rate": 0.0001343125734430082,
+      "loss": 0.2237,
+      "step": 18800
+    },
+    {
+      "epoch": 11.104582843713278,
+      "grad_norm": 2.6339073181152344,
+      "learning_rate": 0.0001334312573443008,
+      "loss": 0.2237,
+      "step": 18900
+    },
+    {
+      "epoch": 11.163337250293772,
+      "grad_norm": 2.57064151763916,
+      "learning_rate": 0.0001325499412455934,
+      "loss": 0.2238,
+      "step": 19000
+    },
+    {
+      "epoch": 11.163337250293772,
+      "eval_accuracy": 0.783353657276316,
+      "eval_f1_macro": 0.456551925690663,
+      "eval_loss": 0.21650880575180054,
+      "eval_precision": 0.5127229337198491,
+      "eval_recall": 0.42967693938948215,
+      "eval_runtime": 585.3328,
+      "eval_samples_per_second": 82.659,
+      "eval_steps_per_second": 0.646,
+      "step": 19000
+    },
+    {
+      "epoch": 11.222091656874266,
+      "grad_norm": 2.160663366317749,
+      "learning_rate": 0.000131668625146886,
+      "loss": 0.2275,
+      "step": 19100
+    },
+    {
+      "epoch": 11.280846063454758,
+      "grad_norm": 5.847405910491943,
+      "learning_rate": 0.0001307873090481786,
+      "loss": 0.2271,
+      "step": 19200
+    },
+    {
+      "epoch": 11.339600470035252,
+      "grad_norm": 2.103134870529175,
+      "learning_rate": 0.0001299059929494712,
+      "loss": 0.2295,
+      "step": 19300
+    },
+    {
+      "epoch": 11.398354876615747,
+      "grad_norm": 2.2549660205841064,
+      "learning_rate": 0.0001290246768507638,
+      "loss": 0.2229,
+      "step": 19400
+    },
+    {
+      "epoch": 11.457109283196239,
+      "grad_norm": 3.1517040729522705,
+      "learning_rate": 0.0001281433607520564,
+      "loss": 0.2221,
+      "step": 19500
+    },
+    {
+      "epoch": 11.515863689776733,
+      "grad_norm": 2.703953266143799,
+      "learning_rate": 0.000127262044653349,
+      "loss": 0.2216,
+      "step": 19600
+    },
+    {
+      "epoch": 11.574618096357227,
+      "grad_norm": 1.5301584005355835,
+      "learning_rate": 0.0001263807285546416,
+      "loss": 0.2301,
+      "step": 19700
+    },
+    {
+      "epoch": 11.63337250293772,
+      "grad_norm": 3.967664957046509,
+      "learning_rate": 0.00012549941245593418,
+      "loss": 0.2257,
+      "step": 19800
+    },
+    {
+      "epoch": 11.692126909518214,
+      "grad_norm": 3.278876543045044,
+      "learning_rate": 0.00012461809635722678,
+      "loss": 0.2277,
+      "step": 19900
+    },
+    {
+      "epoch": 11.750881316098708,
+      "grad_norm": 3.274348497390747,
+      "learning_rate": 0.00012373678025851938,
+      "loss": 0.2262,
+      "step": 20000
+    },
+    {
+      "epoch": 11.750881316098708,
+      "eval_accuracy": 0.7706632494884567,
+      "eval_f1_macro": 0.46808422191837634,
+      "eval_loss": 0.21933460235595703,
+      "eval_precision": 0.4863537664371959,
+      "eval_recall": 0.4608368369449391,
+      "eval_runtime": 542.8512,
+      "eval_samples_per_second": 89.128,
+      "eval_steps_per_second": 0.696,
+      "step": 20000
+    },
+    {
+      "epoch": 11.8096357226792,
+      "grad_norm": 2.5668694972991943,
+      "learning_rate": 0.00012285546415981196,
+      "loss": 0.2212,
+      "step": 20100
+    },
+    {
+      "epoch": 11.868390129259694,
+      "grad_norm": 2.187702178955078,
+      "learning_rate": 0.00012197414806110456,
+      "loss": 0.2254,
+      "step": 20200
+    },
+    {
+      "epoch": 11.927144535840188,
+      "grad_norm": 2.247164487838745,
+      "learning_rate": 0.00012109283196239717,
+      "loss": 0.2268,
+      "step": 20300
+    },
+    {
+      "epoch": 11.98589894242068,
+      "grad_norm": 4.919483184814453,
+      "learning_rate": 0.00012021151586368976,
+      "loss": 0.2251,
+      "step": 20400
+    },
+    {
+      "epoch": 12.044653349001175,
+      "grad_norm": 2.580787181854248,
+      "learning_rate": 0.00011933019976498236,
+      "loss": 0.2255,
+      "step": 20500
+    },
+    {
+      "epoch": 12.103407755581669,
+      "grad_norm": 3.7776031494140625,
+      "learning_rate": 0.00011844888366627496,
+      "loss": 0.2217,
+      "step": 20600
+    },
+    {
+      "epoch": 12.162162162162161,
+      "grad_norm": 2.159958839416504,
+      "learning_rate": 0.00011756756756756755,
+      "loss": 0.2213,
+      "step": 20700
+    },
+    {
+      "epoch": 12.220916568742656,
+      "grad_norm": 1.7245205640792847,
+      "learning_rate": 0.00011668625146886015,
+      "loss": 0.2217,
+      "step": 20800
+    },
+    {
+      "epoch": 12.27967097532315,
+      "grad_norm": 1.598755955696106,
+      "learning_rate": 0.00011580493537015276,
+      "loss": 0.2227,
+      "step": 20900
+    },
+    {
+      "epoch": 12.338425381903642,
+      "grad_norm": 1.9064700603485107,
+      "learning_rate": 0.00011492361927144535,
+      "loss": 0.2239,
+      "step": 21000
+    },
+    {
+      "epoch": 12.338425381903642,
+      "eval_accuracy": 0.7744042328917181,
+      "eval_f1_macro": 0.46072293060919217,
+      "eval_loss": 0.21473053097724915,
+      "eval_precision": 0.5166471813664621,
+      "eval_recall": 0.44065307647654417,
+      "eval_runtime": 586.4383,
+      "eval_samples_per_second": 82.503,
+      "eval_steps_per_second": 0.645,
+      "step": 21000
+    },
+    {
+      "epoch": 12.397179788484136,
+      "grad_norm": 2.0441579818725586,
+      "learning_rate": 0.00011404230317273795,
+      "loss": 0.2208,
+      "step": 21100
+    },
+    {
+      "epoch": 12.45593419506463,
+      "grad_norm": 4.460620880126953,
+      "learning_rate": 0.00011316098707403055,
+      "loss": 0.2233,
+      "step": 21200
+    },
+    {
+      "epoch": 12.514688601645123,
+      "grad_norm": 2.7372050285339355,
+      "learning_rate": 0.00011227967097532314,
+      "loss": 0.2204,
+      "step": 21300
+    },
+    {
+      "epoch": 12.573443008225617,
+      "grad_norm": 3.1166772842407227,
+      "learning_rate": 0.00011139835487661574,
+      "loss": 0.2283,
+      "step": 21400
+    },
+    {
+      "epoch": 12.632197414806111,
+      "grad_norm": 3.481877565383911,
+      "learning_rate": 0.00011051703877790835,
+      "loss": 0.2206,
+      "step": 21500
+    },
+    {
+      "epoch": 12.690951821386603,
+      "grad_norm": 2.6548030376434326,
+      "learning_rate": 0.00010963572267920094,
+      "loss": 0.2241,
+      "step": 21600
+    },
+    {
+      "epoch": 12.749706227967097,
+      "grad_norm": 2.3535709381103516,
+      "learning_rate": 0.00010875440658049353,
+      "loss": 0.2213,
+      "step": 21700
+    },
+    {
+      "epoch": 12.808460634547592,
+      "grad_norm": 2.523663282394409,
+      "learning_rate": 0.00010787309048178611,
+      "loss": 0.2222,
+      "step": 21800
+    },
+    {
+      "epoch": 12.867215041128084,
+      "grad_norm": 1.9537861347198486,
+      "learning_rate": 0.00010699177438307872,
+      "loss": 0.2221,
+      "step": 21900
+    },
+    {
+      "epoch": 12.925969447708578,
+      "grad_norm": 1.9098992347717285,
+      "learning_rate": 0.00010611045828437131,
+      "loss": 0.2196,
+      "step": 22000
+    },
+    {
+      "epoch": 12.925969447708578,
+      "eval_accuracy": 0.782361573279871,
+      "eval_f1_macro": 0.45321392135920985,
+      "eval_loss": 0.21087060868740082,
+      "eval_precision": 0.5160027197721393,
+      "eval_recall": 0.4277450812600511,
+      "eval_runtime": 561.4808,
+      "eval_samples_per_second": 86.17,
+      "eval_steps_per_second": 0.673,
+      "step": 22000
+    },
+    {
+      "epoch": 12.984723854289072,
+      "grad_norm": 2.665001153945923,
+      "learning_rate": 0.00010522914218566391,
+      "loss": 0.2224,
+      "step": 22100
+    },
+    {
+      "epoch": 13.043478260869565,
+      "grad_norm": 3.2731380462646484,
+      "learning_rate": 0.00010434782608695651,
+      "loss": 0.2161,
+      "step": 22200
+    },
+    {
+      "epoch": 13.102232667450059,
+      "grad_norm": 1.7394378185272217,
+      "learning_rate": 0.0001034665099882491,
+      "loss": 0.2213,
+      "step": 22300
+    },
+    {
+      "epoch": 13.160987074030553,
+      "grad_norm": 4.027496337890625,
+      "learning_rate": 0.0001025851938895417,
+      "loss": 0.2232,
+      "step": 22400
+    },
+    {
+      "epoch": 13.219741480611045,
+      "grad_norm": 4.968031883239746,
+      "learning_rate": 0.00010170387779083431,
+      "loss": 0.2228,
+      "step": 22500
+    },
+    {
+      "epoch": 13.27849588719154,
+      "grad_norm": 2.2942428588867188,
+      "learning_rate": 0.0001008225616921269,
+      "loss": 0.2243,
+      "step": 22600
+    },
+    {
+      "epoch": 13.337250293772033,
+      "grad_norm": 1.5325312614440918,
+      "learning_rate": 9.99412455934195e-05,
+      "loss": 0.2191,
+      "step": 22700
+    },
+    {
+      "epoch": 13.396004700352526,
+      "grad_norm": 4.171008586883545,
+      "learning_rate": 9.90599294947121e-05,
+      "loss": 0.225,
+      "step": 22800
+    },
+    {
+      "epoch": 13.45475910693302,
+      "grad_norm": 2.144474506378174,
+      "learning_rate": 9.817861339600469e-05,
+      "loss": 0.2191,
+      "step": 22900
+    },
+    {
+      "epoch": 13.513513513513514,
+      "grad_norm": 1.8419458866119385,
+      "learning_rate": 9.72972972972973e-05,
+      "loss": 0.2244,
+      "step": 23000
+    },
+    {
+      "epoch": 13.513513513513514,
+      "eval_accuracy": 0.7836016782754274,
+      "eval_f1_macro": 0.46679865317019514,
+      "eval_loss": 0.21025818586349487,
+      "eval_precision": 0.5055179262576418,
+      "eval_recall": 0.4468885382196594,
+      "eval_runtime": 567.4084,
+      "eval_samples_per_second": 85.27,
+      "eval_steps_per_second": 0.666,
+      "step": 23000
+    },
+    {
+      "epoch": 13.572267920094006,
+      "grad_norm": 2.069737672805786,
+      "learning_rate": 9.64159811985899e-05,
+      "loss": 0.2171,
+      "step": 23100
+    },
+    {
+      "epoch": 13.6310223266745,
+      "grad_norm": 2.8181750774383545,
+      "learning_rate": 9.553466509988249e-05,
+      "loss": 0.2199,
+      "step": 23200
+    },
+    {
+      "epoch": 13.689776733254995,
+      "grad_norm": 1.9151453971862793,
+      "learning_rate": 9.465334900117508e-05,
+      "loss": 0.2215,
+      "step": 23300
+    },
+    {
+      "epoch": 13.748531139835487,
+      "grad_norm": 2.501735210418701,
+      "learning_rate": 9.377203290246767e-05,
+      "loss": 0.2224,
+      "step": 23400
+    },
+    {
+      "epoch": 13.807285546415981,
+      "grad_norm": 4.269018173217773,
+      "learning_rate": 9.289071680376027e-05,
+      "loss": 0.2226,
+      "step": 23500
+    },
+    {
+      "epoch": 13.866039952996475,
+      "grad_norm": 3.0230236053466797,
+      "learning_rate": 9.200940070505287e-05,
+      "loss": 0.2254,
+      "step": 23600
+    },
+    {
+      "epoch": 13.924794359576968,
+      "grad_norm": 1.6129169464111328,
+      "learning_rate": 9.112808460634546e-05,
+      "loss": 0.223,
+      "step": 23700
+    },
+    {
+      "epoch": 13.983548766157462,
+      "grad_norm": 3.050380229949951,
+      "learning_rate": 9.024676850763806e-05,
+      "loss": 0.2157,
+      "step": 23800
+    },
+    {
+      "epoch": 14.042303172737956,
+      "grad_norm": 1.8896129131317139,
+      "learning_rate": 8.936545240893067e-05,
+      "loss": 0.2203,
+      "step": 23900
+    },
+    {
+      "epoch": 14.101057579318448,
+      "grad_norm": 2.357605218887329,
+      "learning_rate": 8.848413631022326e-05,
+      "loss": 0.2181,
+      "step": 24000
+    },
+    {
+      "epoch": 14.101057579318448,
+      "eval_accuracy": 0.7799227001219436,
+      "eval_f1_macro": 0.4646864264769805,
+      "eval_loss": 0.2089788019657135,
+      "eval_precision": 0.5189278565795705,
+      "eval_recall": 0.4427888401365953,
+      "eval_runtime": 561.1944,
+      "eval_samples_per_second": 86.214,
+      "eval_steps_per_second": 0.674,
+      "step": 24000
+    },
+    {
+      "epoch": 14.159811985898942,
+      "grad_norm": 2.8449645042419434,
+      "learning_rate": 8.760282021151586e-05,
+      "loss": 0.2158,
+      "step": 24100
+    },
+    {
+      "epoch": 14.218566392479437,
+      "grad_norm": 3.220463752746582,
+      "learning_rate": 8.672150411280845e-05,
+      "loss": 0.2234,
+      "step": 24200
+    },
+    {
+      "epoch": 14.277320799059929,
+      "grad_norm": 2.0377910137176514,
+      "learning_rate": 8.584018801410105e-05,
+      "loss": 0.222,
+      "step": 24300
+    },
+    {
+      "epoch": 14.336075205640423,
+      "grad_norm": 2.213088274002075,
+      "learning_rate": 8.495887191539365e-05,
+      "loss": 0.217,
+      "step": 24400
+    },
+    {
+      "epoch": 14.394829612220917,
+      "grad_norm": 3.5318024158477783,
+      "learning_rate": 8.407755581668624e-05,
+      "loss": 0.218,
+      "step": 24500
+    },
+    {
+      "epoch": 14.45358401880141,
+      "grad_norm": 2.010096549987793,
+      "learning_rate": 8.319623971797885e-05,
+      "loss": 0.2189,
+      "step": 24600
+    },
+    {
+      "epoch": 14.512338425381904,
+      "grad_norm": 1.9498238563537598,
+      "learning_rate": 8.231492361927145e-05,
+      "loss": 0.2238,
+      "step": 24700
+    },
+    {
+      "epoch": 14.571092831962398,
+      "grad_norm": 2.8972408771514893,
+      "learning_rate": 8.143360752056404e-05,
+      "loss": 0.2133,
+      "step": 24800
+    },
+    {
+      "epoch": 14.62984723854289,
+      "grad_norm": 1.39292311668396,
+      "learning_rate": 8.055229142185663e-05,
+      "loss": 0.2229,
+      "step": 24900
+    },
+    {
+      "epoch": 14.688601645123384,
+      "grad_norm": 3.813009738922119,
+      "learning_rate": 7.967097532314922e-05,
+      "loss": 0.2165,
+      "step": 25000
+    },
+    {
+      "epoch": 14.688601645123384,
+      "eval_accuracy": 0.7839530413575017,
+      "eval_f1_macro": 0.4677806630551704,
+      "eval_loss": 0.21110670268535614,
+      "eval_precision": 0.5050811199516158,
+      "eval_recall": 0.44609252100680624,
+      "eval_runtime": 551.7967,
+      "eval_samples_per_second": 87.683,
+      "eval_steps_per_second": 0.685,
+      "step": 25000
+    },
+    {
+      "epoch": 14.747356051703878,
+      "grad_norm": 2.7742602825164795,
+      "learning_rate": 7.878965922444182e-05,
+      "loss": 0.2139,
+      "step": 25100
+    },
+    {
+      "epoch": 14.80611045828437,
+      "grad_norm": 2.4670934677124023,
+      "learning_rate": 7.790834312573442e-05,
+      "loss": 0.2183,
+      "step": 25200
+    },
+    {
+      "epoch": 14.864864864864865,
+      "grad_norm": 3.2507874965667725,
+      "learning_rate": 7.702702702702701e-05,
+      "loss": 0.2229,
+      "step": 25300
+    },
+    {
+      "epoch": 14.923619271445359,
+      "grad_norm": 1.7584885358810425,
+      "learning_rate": 7.614571092831962e-05,
+      "loss": 0.2171,
+      "step": 25400
+    },
+    {
+      "epoch": 14.982373678025851,
+      "grad_norm": 2.5273969173431396,
+      "learning_rate": 7.526439482961222e-05,
+      "loss": 0.2184,
+      "step": 25500
+    },
+    {
+      "epoch": 15.041128084606346,
+      "grad_norm": 2.622952699661255,
+      "learning_rate": 7.438307873090481e-05,
+      "loss": 0.2162,
+      "step": 25600
+    },
+    {
+      "epoch": 15.09988249118684,
+      "grad_norm": 2.1974570751190186,
+      "learning_rate": 7.350176263219741e-05,
+      "loss": 0.2104,
+      "step": 25700
+    },
+    {
+      "epoch": 15.158636897767332,
+      "grad_norm": 2.2584497928619385,
+      "learning_rate": 7.262044653349001e-05,
+      "loss": 0.2152,
+      "step": 25800
+    },
+    {
+      "epoch": 15.217391304347826,
+      "grad_norm": 3.1817431449890137,
+      "learning_rate": 7.17391304347826e-05,
+      "loss": 0.2144,
+      "step": 25900
+    },
+    {
+      "epoch": 15.27614571092832,
+      "grad_norm": 3.306057929992676,
+      "learning_rate": 7.08578143360752e-05,
+      "loss": 0.2197,
+      "step": 26000
+    },
+    {
+      "epoch": 15.27614571092832,
+      "eval_accuracy": 0.7750036169729037,
+      "eval_f1_macro": 0.4676251296697811,
+      "eval_loss": 0.21305988729000092,
+      "eval_precision": 0.49501914130610697,
+      "eval_recall": 0.45513146662613874,
+      "eval_runtime": 565.178,
+      "eval_samples_per_second": 85.607,
+      "eval_steps_per_second": 0.669,
+      "step": 26000
+    },
+    {
+      "epoch": 15.334900117508813,
+      "grad_norm": 2.0208778381347656,
+      "learning_rate": 6.99764982373678e-05,
+      "loss": 0.2214,
+      "step": 26100
+    },
+    {
+      "epoch": 15.393654524089307,
+      "grad_norm": 3.2097744941711426,
+      "learning_rate": 6.909518213866038e-05,
+      "loss": 0.2195,
+      "step": 26200
+    },
+    {
+      "epoch": 15.452408930669801,
+      "grad_norm": 2.718372344970703,
+      "learning_rate": 6.821386603995299e-05,
+      "loss": 0.2233,
+      "step": 26300
+    },
+    {
+      "epoch": 15.511163337250293,
+      "grad_norm": 3.371232032775879,
+      "learning_rate": 6.733254994124559e-05,
+      "loss": 0.2148,
+      "step": 26400
+    },
+    {
+      "epoch": 15.569917743830787,
+      "grad_norm": 1.748062014579773,
+      "learning_rate": 6.645123384253818e-05,
+      "loss": 0.2151,
+      "step": 26500
+    },
+    {
+      "epoch": 15.628672150411282,
+      "grad_norm": 2.6323885917663574,
+      "learning_rate": 6.556991774383078e-05,
+      "loss": 0.2193,
+      "step": 26600
+    },
+    {
+      "epoch": 15.687426556991774,
+      "grad_norm": 3.380427598953247,
+      "learning_rate": 6.468860164512338e-05,
+      "loss": 0.2151,
+      "step": 26700
+    },
+    {
+      "epoch": 15.746180963572268,
+      "grad_norm": 2.617914915084839,
+      "learning_rate": 6.380728554641597e-05,
+      "loss": 0.2175,
+      "step": 26800
+    },
+    {
+      "epoch": 15.804935370152762,
+      "grad_norm": 2.5959670543670654,
+      "learning_rate": 6.292596944770856e-05,
+      "loss": 0.2158,
+      "step": 26900
+    },
+    {
+      "epoch": 15.863689776733255,
+      "grad_norm": 1.8247867822647095,
+      "learning_rate": 6.204465334900117e-05,
+      "loss": 0.2173,
+      "step": 27000
+    },
+    {
+      "epoch": 15.863689776733255,
+      "eval_accuracy": 0.7821548891139449,
+      "eval_f1_macro": 0.46965749568257437,
+      "eval_loss": 0.2087218463420868,
+      "eval_precision": 0.5036667780561973,
+      "eval_recall": 0.45166012346117074,
+      "eval_runtime": 551.32,
+      "eval_samples_per_second": 87.758,
+      "eval_steps_per_second": 0.686,
+      "step": 27000
+    },
+    {
+      "epoch": 15.922444183313749,
+      "grad_norm": 2.068295478820801,
+      "learning_rate": 6.116333725029377e-05,
+      "loss": 0.2161,
+      "step": 27100
+    },
+    {
+      "epoch": 15.981198589894243,
+      "grad_norm": 2.173018455505371,
+      "learning_rate": 6.0282021151586365e-05,
+      "loss": 0.2192,
+      "step": 27200
+    },
+    {
+      "epoch": 16.039952996474735,
+      "grad_norm": 3.564419746398926,
+      "learning_rate": 5.940070505287896e-05,
+      "loss": 0.213,
+      "step": 27300
+    },
+    {
+      "epoch": 16.09870740305523,
+      "grad_norm": 2.131643772125244,
+      "learning_rate": 5.851938895417156e-05,
+      "loss": 0.219,
+      "step": 27400
+    },
+    {
+      "epoch": 16.157461809635723,
+      "grad_norm": 1.6084250211715698,
+      "learning_rate": 5.7638072855464154e-05,
+      "loss": 0.2168,
+      "step": 27500
+    },
+    {
+      "epoch": 16.216216216216218,
+      "grad_norm": 1.8609333038330078,
+      "learning_rate": 5.6756756756756757e-05,
+      "loss": 0.2203,
+      "step": 27600
+    },
+    {
+      "epoch": 16.274970622796708,
+      "grad_norm": 1.929494857788086,
+      "learning_rate": 5.5875440658049346e-05,
+      "loss": 0.2126,
+      "step": 27700
+    },
+    {
+      "epoch": 16.333725029377202,
+      "grad_norm": 1.7891273498535156,
+      "learning_rate": 5.499412455934194e-05,
+      "loss": 0.2125,
+      "step": 27800
+    },
+    {
+      "epoch": 16.392479435957696,
+      "grad_norm": 1.935006022453308,
+      "learning_rate": 5.411280846063454e-05,
+      "loss": 0.2136,
+      "step": 27900
+    },
+    {
+      "epoch": 16.45123384253819,
+      "grad_norm": 2.7039895057678223,
+      "learning_rate": 5.323149236192714e-05,
+      "loss": 0.2204,
+      "step": 28000
+    },
+    {
+      "epoch": 16.45123384253819,
+      "eval_accuracy": 0.781844862865056,
+      "eval_f1_macro": 0.47010557287584404,
+      "eval_loss": 0.20962630212306976,
+      "eval_precision": 0.5006758936230861,
+      "eval_recall": 0.45393923309607365,
+      "eval_runtime": 539.5236,
+      "eval_samples_per_second": 89.677,
+      "eval_steps_per_second": 0.701,
+      "step": 28000
+    },
+    {
+      "epoch": 16.509988249118685,
+      "grad_norm": 2.465174913406372,
+      "learning_rate": 5.235017626321974e-05,
+      "loss": 0.215,
+      "step": 28100
+    },
+    {
+      "epoch": 16.56874265569918,
+      "grad_norm": 3.626897096633911,
+      "learning_rate": 5.146886016451233e-05,
+      "loss": 0.2159,
+      "step": 28200
+    },
+    {
+      "epoch": 16.62749706227967,
+      "grad_norm": 1.9083856344223022,
+      "learning_rate": 5.0587544065804936e-05,
+      "loss": 0.2146,
+      "step": 28300
+    },
+    {
+      "epoch": 16.686251468860164,
+      "grad_norm": 1.8644742965698242,
+      "learning_rate": 4.970622796709753e-05,
+      "loss": 0.2155,
+      "step": 28400
+    },
+    {
+      "epoch": 16.745005875440658,
+      "grad_norm": 2.8223023414611816,
+      "learning_rate": 4.882491186839013e-05,
+      "loss": 0.2127,
+      "step": 28500
+    },
+    {
+      "epoch": 16.803760282021152,
+      "grad_norm": 2.9986822605133057,
+      "learning_rate": 4.794359576968272e-05,
+      "loss": 0.2168,
+      "step": 28600
+    },
+    {
+      "epoch": 16.862514688601646,
+      "grad_norm": 2.5670571327209473,
+      "learning_rate": 4.7062279670975314e-05,
+      "loss": 0.2105,
+      "step": 28700
+    },
+    {
+      "epoch": 16.92126909518214,
+      "grad_norm": 4.0372467041015625,
+      "learning_rate": 4.6180963572267917e-05,
+      "loss": 0.2116,
+      "step": 28800
+    },
+    {
+      "epoch": 16.98002350176263,
+      "grad_norm": 2.199449300765991,
+      "learning_rate": 4.529964747356051e-05,
+      "loss": 0.2188,
+      "step": 28900
+    },
+    {
+      "epoch": 17.038777908343125,
+      "grad_norm": 2.2641525268554688,
+      "learning_rate": 4.441833137485311e-05,
+      "loss": 0.2157,
+      "step": 29000
+    },
+    {
+      "epoch": 17.038777908343125,
+      "eval_accuracy": 0.7866192670979476,
+      "eval_f1_macro": 0.46819371251173775,
+      "eval_loss": 0.2086167186498642,
+      "eval_precision": 0.5101989602452385,
+      "eval_recall": 0.44473033751495855,
+      "eval_runtime": 537.4744,
+      "eval_samples_per_second": 90.019,
+      "eval_steps_per_second": 0.703,
+      "step": 29000
+    },
+    {
+      "epoch": 17.09753231492362,
+      "grad_norm": 1.7599774599075317,
+      "learning_rate": 4.353701527614571e-05,
+      "loss": 0.2147,
+      "step": 29100
+    },
+    {
+      "epoch": 17.156286721504113,
+      "grad_norm": 3.7391819953918457,
+      "learning_rate": 4.265569917743831e-05,
+      "loss": 0.2131,
+      "step": 29200
+    },
+    {
+      "epoch": 17.215041128084607,
+      "grad_norm": 1.628392219543457,
+      "learning_rate": 4.1774383078730904e-05,
+      "loss": 0.211,
+      "step": 29300
+    },
+    {
+      "epoch": 17.2737955346651,
+      "grad_norm": 2.0813705921173096,
+      "learning_rate": 4.089306698002349e-05,
+      "loss": 0.213,
+      "step": 29400
+    },
+    {
+      "epoch": 17.332549941245592,
+      "grad_norm": 1.5833709239959717,
+      "learning_rate": 4.0011750881316096e-05,
+      "loss": 0.2122,
+      "step": 29500
+    },
+    {
+      "epoch": 17.391304347826086,
+      "grad_norm": 2.216641664505005,
+      "learning_rate": 3.913043478260869e-05,
+      "loss": 0.2123,
+      "step": 29600
+    },
+    {
+      "epoch": 17.45005875440658,
+      "grad_norm": 1.9753063917160034,
+      "learning_rate": 3.824911868390129e-05,
+      "loss": 0.2121,
+      "step": 29700
+    },
+    {
+      "epoch": 17.508813160987074,
+      "grad_norm": 2.2607269287109375,
+      "learning_rate": 3.7367802585193884e-05,
+      "loss": 0.2179,
+      "step": 29800
+    },
+    {
+      "epoch": 17.56756756756757,
+      "grad_norm": 4.074460506439209,
+      "learning_rate": 3.648648648648649e-05,
+      "loss": 0.2104,
+      "step": 29900
+    },
+    {
+      "epoch": 17.626321974148063,
+      "grad_norm": 1.758702039718628,
+      "learning_rate": 3.560517038777908e-05,
+      "loss": 0.2135,
+      "step": 30000
+    },
+    {
+      "epoch": 17.626321974148063,
+      "eval_accuracy": 0.781369489283426,
+      "eval_f1_macro": 0.4699363142405637,
+      "eval_loss": 0.20687498152256012,
+      "eval_precision": 0.5076781202687786,
+      "eval_recall": 0.4518498245169957,
+      "eval_runtime": 584.9222,
+      "eval_samples_per_second": 82.717,
+      "eval_steps_per_second": 0.646,
+      "step": 30000
+    },
+    {
+      "epoch": 17.685076380728553,
+      "grad_norm": 2.3470921516418457,
+      "learning_rate": 3.472385428907168e-05,
+      "loss": 0.2152,
+      "step": 30100
+    },
+    {
+      "epoch": 17.743830787309047,
+      "grad_norm": 2.4279568195343018,
+      "learning_rate": 3.3842538190364276e-05,
+      "loss": 0.2144,
+      "step": 30200
+    },
+    {
+      "epoch": 17.80258519388954,
+      "grad_norm": 1.7226651906967163,
+      "learning_rate": 3.296122209165687e-05,
+      "loss": 0.214,
+      "step": 30300
+    },
+    {
+      "epoch": 17.861339600470036,
+      "grad_norm": 3.090634346008301,
+      "learning_rate": 3.207990599294947e-05,
+      "loss": 0.2175,
+      "step": 30400
+    },
+    {
+      "epoch": 17.92009400705053,
+      "grad_norm": 1.787026286125183,
+      "learning_rate": 3.1198589894242064e-05,
+      "loss": 0.2112,
+      "step": 30500
+    },
+    {
+      "epoch": 17.978848413631024,
+      "grad_norm": 1.844420313835144,
+      "learning_rate": 3.0317273795534663e-05,
+      "loss": 0.2132,
+      "step": 30600
+    },
+    {
+      "epoch": 18.037602820211514,
+      "grad_norm": 1.7395440340042114,
+      "learning_rate": 2.9435957696827263e-05,
+      "loss": 0.2091,
+      "step": 30700
+    },
+    {
+      "epoch": 18.09635722679201,
+      "grad_norm": 1.9990227222442627,
+      "learning_rate": 2.8554641598119856e-05,
+      "loss": 0.2123,
+      "step": 30800
+    },
+    {
+      "epoch": 18.155111633372503,
+      "grad_norm": 2.4807918071746826,
+      "learning_rate": 2.7673325499412452e-05,
+      "loss": 0.2082,
+      "step": 30900
+    },
+    {
+      "epoch": 18.213866039952997,
+      "grad_norm": 2.496959924697876,
+      "learning_rate": 2.679200940070505e-05,
+      "loss": 0.2088,
+      "step": 31000
+    },
+    {
+      "epoch": 18.213866039952997,
+      "eval_accuracy": 0.783353657276316,
+      "eval_f1_macro": 0.4700197150315801,
+      "eval_loss": 0.20601825416088104,
+      "eval_precision": 0.5112111682738877,
+      "eval_recall": 0.45050131045693015,
+      "eval_runtime": 542.6455,
+      "eval_samples_per_second": 89.161,
+      "eval_steps_per_second": 0.697,
+      "step": 31000
+    },
+    {
+      "epoch": 18.27262044653349,
+      "grad_norm": 2.345229387283325,
+      "learning_rate": 2.591069330199765e-05,
+      "loss": 0.2152,
+      "step": 31100
+    },
+    {
+      "epoch": 18.331374853113985,
+      "grad_norm": 2.492675542831421,
+      "learning_rate": 2.5029377203290243e-05,
+      "loss": 0.2115,
+      "step": 31200
+    },
+    {
+      "epoch": 18.390129259694476,
+      "grad_norm": 2.251948833465576,
+      "learning_rate": 2.4148061104582843e-05,
+      "loss": 0.2111,
+      "step": 31300
+    },
+    {
+      "epoch": 18.44888366627497,
+      "grad_norm": 2.327437400817871,
+      "learning_rate": 2.326674500587544e-05,
+      "loss": 0.2085,
+      "step": 31400
+    },
+    {
+      "epoch": 18.507638072855464,
+      "grad_norm": 2.292947292327881,
+      "learning_rate": 2.238542890716804e-05,
+      "loss": 0.2141,
+      "step": 31500
+    },
+    {
+      "epoch": 18.566392479435958,
+      "grad_norm": 2.319504499435425,
+      "learning_rate": 2.150411280846063e-05,
+      "loss": 0.2133,
+      "step": 31600
+    },
+    {
+      "epoch": 18.625146886016452,
+      "grad_norm": 2.6240415573120117,
+      "learning_rate": 2.062279670975323e-05,
+      "loss": 0.2161,
+      "step": 31700
+    },
+    {
+      "epoch": 18.683901292596946,
+      "grad_norm": 1.637320637702942,
+      "learning_rate": 1.9741480611045827e-05,
+      "loss": 0.2084,
+      "step": 31800
+    },
+    {
+      "epoch": 18.742655699177437,
+      "grad_norm": 1.8866758346557617,
+      "learning_rate": 1.8860164512338426e-05,
+      "loss": 0.2099,
+      "step": 31900
+    },
+    {
+      "epoch": 18.80141010575793,
+      "grad_norm": 1.5909967422485352,
+      "learning_rate": 1.7978848413631022e-05,
+      "loss": 0.215,
+      "step": 32000
+    },
+    {
+      "epoch": 18.80141010575793,
+      "eval_accuracy": 0.7825475890292045,
+      "eval_f1_macro": 0.4729882480682901,
+      "eval_loss": 0.2060166597366333,
+      "eval_precision": 0.5092421494744399,
+      "eval_recall": 0.45578106777502264,
+      "eval_runtime": 530.5298,
+      "eval_samples_per_second": 91.198,
+      "eval_steps_per_second": 0.712,
+      "step": 32000
+    },
+    {
+      "epoch": 18.860164512338425,
+      "grad_norm": 2.3732552528381348,
+      "learning_rate": 1.709753231492362e-05,
+      "loss": 0.2126,
+      "step": 32100
+    },
+    {
+      "epoch": 18.91891891891892,
+      "grad_norm": 2.5773391723632812,
+      "learning_rate": 1.6216216216216215e-05,
+      "loss": 0.2095,
+      "step": 32200
+    },
+    {
+      "epoch": 18.977673325499413,
+      "grad_norm": 2.3607606887817383,
+      "learning_rate": 1.533490011750881e-05,
+      "loss": 0.2099,
+      "step": 32300
+    },
+    {
+      "epoch": 19.036427732079908,
+      "grad_norm": 3.814934730529785,
+      "learning_rate": 1.445358401880141e-05,
+      "loss": 0.21,
+      "step": 32400
+    },
+    {
+      "epoch": 19.095182138660398,
+      "grad_norm": 1.9512494802474976,
+      "learning_rate": 1.3572267920094006e-05,
+      "loss": 0.2098,
+      "step": 32500
+    },
+    {
+      "epoch": 19.153936545240892,
+      "grad_norm": 2.072913408279419,
+      "learning_rate": 1.2690951821386604e-05,
+      "loss": 0.2156,
+      "step": 32600
+    },
+    {
+      "epoch": 19.212690951821386,
+      "grad_norm": 3.2823166847229004,
+      "learning_rate": 1.18096357226792e-05,
+      "loss": 0.2058,
+      "step": 32700
+    },
+    {
+      "epoch": 19.27144535840188,
+      "grad_norm": 2.4737155437469482,
+      "learning_rate": 1.0928319623971798e-05,
+      "loss": 0.2084,
+      "step": 32800
+    },
+    {
+      "epoch": 19.330199764982375,
+      "grad_norm": 2.195495843887329,
+      "learning_rate": 1.0047003525264394e-05,
+      "loss": 0.2099,
+      "step": 32900
+    },
+    {
+      "epoch": 19.38895417156287,
+      "grad_norm": 2.7346057891845703,
+      "learning_rate": 9.16568742655699e-06,
+      "loss": 0.2138,
+      "step": 33000
+    },
+    {
+      "epoch": 19.38895417156287,
+      "eval_accuracy": 0.7822788996135006,
+      "eval_f1_macro": 0.47654790915623974,
+      "eval_loss": 0.2069149762392044,
+      "eval_precision": 0.5055123102220627,
+      "eval_recall": 0.4617181765781322,
+      "eval_runtime": 563.9047,
+      "eval_samples_per_second": 85.8,
+      "eval_steps_per_second": 0.67,
+      "step": 33000
+    },
+    {
+      "epoch": 19.44770857814336,
+      "grad_norm": 1.7967709302902222,
+      "learning_rate": 8.284371327849588e-06,
+      "loss": 0.2113,
+      "step": 33100
+    },
+    {
+      "epoch": 19.506462984723854,
+      "grad_norm": 2.043199300765991,
+      "learning_rate": 7.403055229142185e-06,
+      "loss": 0.2134,
+      "step": 33200
+    },
+    {
+      "epoch": 19.565217391304348,
+      "grad_norm": 2.2731423377990723,
+      "learning_rate": 6.521739130434782e-06,
+      "loss": 0.2053,
+      "step": 33300
+    },
+    {
+      "epoch": 19.623971797884842,
+      "grad_norm": 1.7844184637069702,
+      "learning_rate": 5.640423031727379e-06,
+      "loss": 0.2102,
+      "step": 33400
+    },
+    {
+      "epoch": 19.682726204465336,
+      "grad_norm": 2.1626009941101074,
+      "learning_rate": 4.759106933019976e-06,
+      "loss": 0.2165,
+      "step": 33500
+    },
+    {
+      "epoch": 19.74148061104583,
+      "grad_norm": 2.5223422050476074,
+      "learning_rate": 3.877790834312573e-06,
+      "loss": 0.2061,
+      "step": 33600
+    },
+    {
+      "epoch": 19.80023501762632,
+      "grad_norm": 2.5433812141418457,
+      "learning_rate": 2.99647473560517e-06,
+      "loss": 0.2083,
+      "step": 33700
+    },
+    {
+      "epoch": 19.858989424206815,
+      "grad_norm": 2.087890148162842,
+      "learning_rate": 2.1151586368977672e-06,
+      "loss": 0.2117,
+      "step": 33800
+    },
+    {
+      "epoch": 19.91774383078731,
+      "grad_norm": 2.9558768272399902,
+      "learning_rate": 1.2338425381903642e-06,
+      "loss": 0.2139,
+      "step": 33900
+    },
+    {
+      "epoch": 19.976498237367803,
+      "grad_norm": 2.22611927986145,
+      "learning_rate": 3.525264394829612e-07,
+      "loss": 0.2125,
+      "step": 34000
+    },
+    {
+      "epoch": 19.976498237367803,
+      "eval_accuracy": 0.7837876940247608,
+      "eval_f1_macro": 0.4752409882222652,
+      "eval_loss": 0.20548085868358612,
+      "eval_precision": 0.506773667085311,
+      "eval_recall": 0.45863653128297405,
+      "eval_runtime": 556.0383,
+      "eval_samples_per_second": 87.014,
+      "eval_steps_per_second": 0.68,
+      "step": 34000
+    }
+  ],
+  "logging_steps": 100,
+  "max_steps": 34040,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 20,
+  "save_steps": 1000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.2885422443855616e+18,
+  "train_batch_size": 256,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e4ae76ec58b487fbfb990b251c8d99d9f248f8fa34093d0f6f5edb76bc44251
+size 5304