Upload 8 files

README.md

@@ -1,3 +1,111 @@
----
-license: apache-2.0
----
+---
+license: apache-2.0
+base_model: Derify/ModChemBERT-IR-BASE
+library_name: transformers
+tags:
+- modernbert
+- ModChemBERT
+- cheminformatics
+- chemical-language-model
+pipeline_tag: fill-mask
+---
+
+# ModChemBERT: ModernBERT as a Chemical Language Model
+ModChemBERT-IR-BASE is a ModernBERT-based chemical language model (CLM) pretrained on SMILES strings using masked language modeling (MLM). This model serves as a base model for training embedding, retrieval, and reranking models for molecular information retrieval tasks.
+
+## Usage
+Install the `transformers` library starting from v4.56.1:
+
+```bash
+pip install -U transformers>=4.56.1
+```
+
+### Load Model
+```python
+from transformers import AutoModelForMaskedLM, AutoTokenizer
+
+model_id = "Derify/ModChemBERT-IR-BASE"
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+model = AutoModelForMaskedLM.from_pretrained(
+    model_id,
+    trust_remote_code=True,
+    dtype="bfloat16",
+    device_map="auto",
+)
+```
+
+### Fill-Mask Pipeline
+```python
+from transformers import pipeline
+
+fill = pipeline("fill-mask", model=model, tokenizer=tokenizer)
+print(fill("c1ccccc1[MASK]"))
+```
+
+## Architecture
+- Backbone: ModernBERT [1]
+- Hidden size: 1024
+- Intermediate size: 1536
+- Encoder Layers: 22
+- Attention heads: 16
+- Max sequence length: 512 tokens
+- Tokenizer: BPE tokenizer using [MolFormer's vocab](https://github.com/emapco/ModChemBERT/blob/main/modchembert/tokenizers/molformer/vocab.json) (2362 tokens)
+
+## Dataset
+- Pretraining: [PubChem 110M dataset (canonical SMILES strings)](https://ibm.ent.box.com/v/MoLFormer-data)
+
+## Pooling (Classifier / Regressor Head)
+Kallergis et al. [2] demonstrated that the CLM embedding method prior to the prediction head was the strongest contributor to downstream performance among evaluated hyperparameters.
+
+Behrendt et al. [3] noted that the last few layers contain task-specific information and that pooling methods leveraging information from multiple layers can enhance model performance. Their results further demonstrated that the `max_seq_mha` pooling method was particularly effective in low-data regimes.
+
+This base model includes configurable pooling strategies for downstream fine-tuning. When fine-tuned for embedding, retrieval, or reranking tasks (e.g., with Sentence Transformers), various pooling methods can be explored:
+- `cls`: Last layer [CLS]
+- `mean`: Mean over last hidden layer
+- `max_cls`: Max over last k layers of [CLS]
+- `cls_mha`: MHA with [CLS] as query
+- `max_seq_mha`: MHA with max pooled sequence as KV and max pooled [CLS] as query
+- `sum_mean`: Sum over all layers then mean tokens
+- `sum_sum`: Sum over all layers then sum tokens
+- `mean_mean`: Mean over all layers then mean tokens
+- `mean_sum`: Mean over all layers then sum tokens
+- `max_seq_mean`: Max over last k layers then mean tokens
+
+Note: ModChemBERT's `max_seq_mha` differs from MaxPoolBERT [3]. MaxPoolBERT uses PyTorch `nn.MultiheadAttention`, whereas ModChemBERT's `ModChemBertPoolingAttention` adapts ModernBERT's `ModernBertAttention`. 
+On ChemBERTa-3 benchmarks this variant produced stronger validation metrics and avoided the training instabilities (sporadic zero / NaN losses and gradient norms) seen with `nn.MultiheadAttention`. Training instability with ModernBERT has been reported in the past ([discussion 1](https://huggingface.co/answerdotai/ModernBERT-base/discussions/59) and [discussion 2](https://huggingface.co/answerdotai/ModernBERT-base/discussions/63)).
+
+## Intended Use
+* Primary: Base model for training embedding, retrieval, and reranking models for chemical information retrieval tasks using frameworks such as Sentence Transformers.
+* Appropriate for: Fine-tuning for semantic search of chemical compounds, molecular similarity tasks, chemical information retrieval systems, and as a foundation for building chemical embedding models.
+* Not intended for: Direct molecular property prediction without fine-tuning, generating novel molecules, or production use without domain-specific validation.
+
+## Limitations
+- This is a base model pretrained only on masked language modeling; it requires fine-tuning for specific information retrieval tasks.
+- Performance on out-of-domain chemical spaces may vary: very long SMILES (>512 tokens), inorganic/organometallic compounds, polymers, or charged/enumerated tautomers may not be well represented in the training corpus.
+- The model reflects the chemical space distribution of PubChem and may not generalize equally well to all chemical domains.
+
+## Ethical Considerations & Responsible Use
+- This base model is intended for research and development purposes in chemical information retrieval.
+- When fine-tuned for downstream applications, users should validate performance on their specific domain and use case.
+- Do not deploy in clinical, regulatory, or safety-critical settings without rigorous domain-specific validation and appropriate oversight.
+
+## Hardware
+Training was performed on two NVIDIA RTX 3090 GPUs using `accelerate` for distributed (DDP) training.
+
+## Citation
+If you use ModChemBERT-IR-BASE in your research, please cite the checkpoint and the following:
+```
+@software{cortes-2025-modchembert,
+  author = {Emmanuel Cortes},
+  title = {ModChemBERT: ModernBERT as a Chemical Language Model},
+  year = {2025},
+  publisher = {GitHub},
+  howpublished = {GitHub repository},
+  url = {https://github.com/emapco/ModChemBERT}
+}
+```
+
+## References
+1. Warner, Benjamin, et al. "Smarter, better, faster, longer: A modern bidirectional encoder for fast, memory efficient, and long context finetuning and inference." arXiv preprint arXiv:2412.13663 (2024).
+2. Kallergis, G., Asgari, E., Empting, M. et al. Domain adaptable language modeling of chemical compounds identifies potent pathoblockers for Pseudomonas aeruginosa. Commun Chem 8, 114 (2025). https://doi.org/10.1038/s42004-025-01484-4
+3. Behrendt, Maike, Stefan Sylvius Wagner, and Stefan Harmeling. "MaxPoolBERT: Enhancing BERT Classification via Layer-and Token-Wise Aggregation." arXiv preprint arXiv:2505.15696 (2025).

config.json

@@ -0,0 +1,53 @@
+{
+  "architectures": [
+    "ModChemBertForMaskedLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.1,
+  "auto_map": {
+    "AutoConfig": "configuration_modchembert.ModChemBertConfig",
+    "AutoModel": "modeling_modchembert.ModChemBertModel",
+    "AutoModelForMaskedLM": "modeling_modchembert.ModChemBertForMaskedLM",
+    "AutoModelForSequenceClassification": "modeling_modchembert.ModChemBertForSequenceClassification"
+  },
+  "bos_token_id": 0,
+  "classifier_activation": "gelu",
+  "classifier_bias": false,
+  "classifier_dropout": 0.0,
+  "classifier_pooling": "max_seq_mha",
+  "classifier_pooling_attention_dropout": 0.1,
+  "classifier_pooling_last_k": 5,
+  "classifier_pooling_num_attention_heads": 4,
+  "cls_token_id": 0,
+  "decoder_bias": true,
+  "deterministic_flash_attn": false,
+  "dtype": "bfloat16",
+  "embedding_dropout": 0.1,
+  "eos_token_id": 1,
+  "global_attn_every_n_layers": 3,
+  "global_rope_theta": 160000.0,
+  "hidden_activation": "gelu",
+  "hidden_size": 1024,
+  "initializer_cutoff_factor": 2.0,
+  "initializer_range": 0.02,
+  "intermediate_size": 1536,
+  "layer_norm_eps": 1e-05,
+  "local_attention": 8,
+  "local_rope_theta": 10000.0,
+  "max_position_embeddings": 512,
+  "mlp_bias": false,
+  "mlp_dropout": 0.1,
+  "model_type": "modchembert",
+  "norm_bias": false,
+  "norm_eps": 1e-05,
+  "num_attention_heads": 16,
+  "num_hidden_layers": 22,
+  "pad_token_id": 2,
+  "position_embedding_type": "absolute",
+  "repad_logits_with_grad": false,
+  "sep_token_id": 1,
+  "sparse_pred_ignore_index": -100,
+  "sparse_prediction": false,
+  "transformers_version": "4.56.1",
+  "vocab_size": 2362
+}

configuration_modchembert.py

@@ -0,0 +1,90 @@
+# Copyright 2025 Emmanuel Cortes, All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Literal
+
+from transformers.models.modernbert.configuration_modernbert import ModernBertConfig
+
+
+class ModChemBertConfig(ModernBertConfig):
+    """
+    Configuration class for ModChemBert models.
+
+    This configuration class extends ModernBertConfig with additional parameters specific to
+    chemical molecule modeling and custom pooling strategies for classification/regression tasks.
+    It accepts all arguments and keyword arguments from ModernBertConfig.
+
+    Args:
+        classifier_pooling (str, optional): Pooling strategy for sequence classification.
+            Available options:
+            - "cls": Use CLS token representation
+            - "mean": Attention-weighted average pooling
+            - "sum_mean": Sum all hidden states across layers, then mean pool over sequence (ChemLM approach)
+            - "sum_sum": Sum all hidden states across layers, then sum pool over sequence
+            - "mean_mean": Mean all hidden states across layers, then mean pool over sequence
+            - "mean_sum": Mean all hidden states across layers, then sum pool over sequence
+            - "max_cls": Element-wise max pooling over last k hidden states, then take CLS token
+            - "cls_mha": Multi-head attention with CLS token as query and full sequence as keys/values
+            - "max_seq_mha": Max pooling over last k states + multi-head attention with CLS as query
+            - "max_seq_mean": Max pooling over last k hidden states, then mean pooling over sequence
+            Defaults to "sum_mean".
+        classifier_pooling_num_attention_heads (int, optional): Number of attention heads for multi-head attention
+            pooling strategies (cls_mha, max_seq_mha). Defaults to 4.
+        classifier_pooling_attention_dropout (float, optional): Dropout probability for multi-head attention
+            pooling strategies (cls_mha, max_seq_mha). Defaults to 0.0.
+        classifier_pooling_last_k (int, optional): Number of last hidden layers to use for max pooling
+            strategies (max_cls, max_seq_mha, max_seq_mean). Defaults to 8.
+        *args: Variable length argument list passed to ModernBertConfig.
+        **kwargs: Arbitrary keyword arguments passed to ModernBertConfig.
+
+    Note:
+        This class inherits all configuration parameters from ModernBertConfig including
+        hidden_size, num_hidden_layers, num_attention_heads, intermediate_size, etc.
+    """
+
+    model_type = "modchembert"
+
+    def __init__(
+        self,
+        *args,
+        classifier_pooling: Literal[
+            "cls",
+            "mean",
+            "sum_mean",
+            "sum_sum",
+            "mean_mean",
+            "mean_sum",
+            "max_cls",
+            "cls_mha",
+            "max_seq_mha",
+            "max_seq_mean",
+        ] = "max_seq_mha",
+        classifier_pooling_num_attention_heads: int = 4,
+        classifier_pooling_attention_dropout: float = 0.0,
+        classifier_pooling_last_k: int = 8,
+        **kwargs,
+    ):
+        # Pass classifier_pooling="cls" to circumvent ValueError in ModernBertConfig init
+        super().__init__(*args, classifier_pooling="cls", **kwargs)
+        # Override with custom value
+        self.classifier_pooling = classifier_pooling
+        self.classifier_pooling_num_attention_heads = classifier_pooling_num_attention_heads
+        self.classifier_pooling_attention_dropout = classifier_pooling_attention_dropout
+        self.classifier_pooling_last_k = classifier_pooling_last_k
+        self.auto_map = {
+            "AutoConfig": "configuration_modchembert.ModChemBertConfig",
+            "AutoModel": "modeling_modchembert.ModChemBertModel",
+            "AutoModelForMaskedLM": "modeling_modchembert.ModChemBertForMaskedLM",
+            "AutoModelForSequenceClassification": "modeling_modchembert.ModChemBertForSequenceClassification",
+        }

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d213c4cf107cb25234351be1c000c35e5353cd9faced047d57f2bb3d9680fc4
+size 399215212

modeling_modchembert.py

@@ -0,0 +1,734 @@
+# Copyright 2025 Emmanuel Cortes, All Rights Reserved.
+#
+# Copyright 2024 Answer.AI, LightOn, and contributors, and the HuggingFace Inc. team. All rights reserved.
+#
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# This file is adapted from the transformers library.
+# Modifications include:
+# - Additional classifier_pooling options for ModChemBertForSequenceClassification
+#   - sum_mean, sum_sum, mean_sum, mean_mean: from ChemLM (utilizes all hidden states)
+#   - max_cls, cls_mha, max_seq_mha: from MaxPoolBERT (utilizes last k hidden states)
+#   - max_seq_mean: a merge between sum_mean and max_cls (utilizes last k hidden states)
+# - Addition of ModChemBertPoolingAttention for cls_mha and max_seq_mha pooling options
+
+import copy
+import math
+import typing
+from contextlib import nullcontext
+
+import torch
+import torch.nn as nn
+from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
+from transformers.modeling_attn_mask_utils import _prepare_4d_attention_mask
+from transformers.modeling_outputs import BaseModelOutput, MaskedLMOutput, SequenceClassifierOutput
+from transformers.models.modernbert.modeling_modernbert import (
+    MODERNBERT_ATTENTION_FUNCTION,
+    ModernBertEmbeddings,
+    ModernBertEncoderLayer,
+    ModernBertModel,
+    ModernBertPredictionHead,
+    ModernBertPreTrainedModel,
+    ModernBertRotaryEmbedding,
+    _pad_modernbert_output,
+    _unpad_modernbert_input,
+)
+from transformers.utils import logging
+
+from .configuration_modchembert import ModChemBertConfig
+
+logger = logging.get_logger(__name__)
+
+
+class InitWeightsMixin:
+    def _init_weights(self, module: nn.Module):
+        super()._init_weights(module)  # type: ignore
+
+        cutoff_factor = self.config.initializer_cutoff_factor  # type: ignore
+        if cutoff_factor is None:
+            cutoff_factor = 3
+
+        def init_weight(module: nn.Module, std: float):
+            if isinstance(module, nn.Linear):
+                nn.init.trunc_normal_(
+                    module.weight,
+                    mean=0.0,
+                    std=std,
+                    a=-cutoff_factor * std,
+                    b=cutoff_factor * std,
+                )
+                if module.bias is not None:
+                    nn.init.zeros_(module.bias)
+
+        stds = {
+            "in": self.config.initializer_range,  # type: ignore
+            "out": self.config.initializer_range / math.sqrt(2.0 * self.config.num_hidden_layers),  # type: ignore
+            "final_out": self.config.hidden_size**-0.5,  # type: ignore
+        }
+
+        if isinstance(module, ModChemBertForMaskedLM):
+            init_weight(module.decoder, stds["out"])
+        elif isinstance(module, ModChemBertForSequenceClassification):
+            init_weight(module.classifier, stds["final_out"])
+        elif isinstance(module, ModChemBertPoolingAttention):
+            init_weight(module.Wq, stds["in"])
+            init_weight(module.Wk, stds["in"])
+            init_weight(module.Wv, stds["in"])
+            init_weight(module.Wo, stds["out"])
+
+
+class ModChemBertPoolingAttention(nn.Module):
+    """Performs multi-headed self attention on a batch of sequences."""
+
+    def __init__(self, config: ModChemBertConfig):
+        super().__init__()
+        self.config = copy.deepcopy(config)
+        # Override num_attention_heads to use classifier_pooling_num_attention_heads
+        self.config.num_attention_heads = config.classifier_pooling_num_attention_heads
+        # Override attention_dropout to use classifier_pooling_attention_dropout
+        self.config.attention_dropout = config.classifier_pooling_attention_dropout
+
+        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 "
+                f"({config.num_attention_heads})"
+            )
+
+        self.attention_dropout = config.attention_dropout
+        self.num_heads = config.num_attention_heads
+        self.head_dim = config.hidden_size // config.num_attention_heads
+        self.all_head_size = self.head_dim * self.num_heads
+        self.Wq = nn.Linear(config.hidden_size, self.all_head_size, bias=config.attention_bias)
+        self.Wk = nn.Linear(config.hidden_size, self.all_head_size, bias=config.attention_bias)
+        self.Wv = nn.Linear(config.hidden_size, self.all_head_size, bias=config.attention_bias)
+
+        # Use global attention
+        self.local_attention = (-1, -1)
+        rope_theta = config.global_rope_theta
+        # sdpa path from original ModernBert implementation
+        config_copy = copy.deepcopy(config)
+        config_copy.rope_theta = rope_theta
+        self.rotary_emb = ModernBertRotaryEmbedding(config=config_copy)
+
+        self.Wo = nn.Linear(config.hidden_size, config.hidden_size, bias=config.attention_bias)
+        self.out_drop = nn.Dropout(config.attention_dropout) if config.attention_dropout > 0.0 else nn.Identity()
+        self.pruned_heads = set()
+
+    def forward(
+        self,
+        q: torch.Tensor,
+        kv: torch.Tensor,
+        attention_mask: torch.Tensor | None = None,
+        **kwargs,
+    ) -> torch.Tensor:
+        bs, seq_len = kv.shape[:2]
+        q_proj: torch.Tensor = self.Wq(q)
+        k_proj: torch.Tensor = self.Wk(kv)
+        v_proj: torch.Tensor = self.Wv(kv)
+        qkv = torch.stack(
+            (
+                q_proj.reshape(bs, seq_len, self.num_heads, self.head_dim),
+                k_proj.reshape(bs, seq_len, self.num_heads, self.head_dim),
+                v_proj.reshape(bs, seq_len, self.num_heads, self.head_dim),
+            ),
+            dim=2,
+        )  # (bs, seq_len, 3, num_heads, head_dim)
+
+        device = kv.device
+        if attention_mask is None:
+            attention_mask = torch.ones((bs, seq_len), device=device, dtype=torch.bool)
+        position_ids = torch.arange(seq_len, device=device).unsqueeze(0).long()
+
+        attn_outputs = MODERNBERT_ATTENTION_FUNCTION["sdpa"](
+            self,
+            qkv=qkv,
+            attention_mask=_prepare_4d_attention_mask(attention_mask, kv.dtype),
+            sliding_window_mask=None,  # not needed when using global attention
+            position_ids=position_ids,
+            local_attention=self.local_attention,
+            bs=bs,
+            dim=self.all_head_size,
+            **kwargs,
+        )
+        hidden_states = attn_outputs[0]
+        hidden_states = self.out_drop(self.Wo(hidden_states))
+
+        return hidden_states
+
+
+class ModChemBertModel(ModernBertPreTrainedModel):
+    config_class = ModChemBertConfig
+
+    def __init__(self, config: ModChemBertConfig):
+        super().__init__(config)
+        self.config = config
+        self.embeddings = ModernBertEmbeddings(config)
+        self.layers = nn.ModuleList(
+            [ModernBertEncoderLayer(config, layer_id) for layer_id in range(config.num_hidden_layers)]
+        )
+        self.final_norm = nn.LayerNorm(config.hidden_size, eps=config.norm_eps, bias=config.norm_bias)
+        self.gradient_checkpointing = False
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embeddings.tok_embeddings
+
+    def set_input_embeddings(self, value):
+        self.embeddings.tok_embeddings = value  # type: ignore
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor | None = None,
+        attention_mask: torch.Tensor | None = None,
+        sliding_window_mask: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        inputs_embeds: torch.Tensor | None = None,
+        indices: torch.Tensor | None = None,
+        cu_seqlens: torch.Tensor | None = None,
+        max_seqlen: int | None = None,
+        batch_size: int | None = None,
+        seq_len: int | None = None,
+        output_attentions: bool | None = None,
+        output_hidden_states: bool | None = None,
+        return_dict: bool | None = None,
+    ) -> tuple[torch.Tensor, ...] | BaseModelOutput:
+        r"""
+        sliding_window_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Mask to avoid performing attention on padding or far-away tokens. In ModernBert, only every few layers
+            perform global attention, while the rest perform local attention. This mask is used to avoid attending to
+            far-away tokens in the local attention layers when not using Flash Attention.
+        indices (`torch.Tensor` of shape `(total_unpadded_tokens,)`, *optional*):
+            Indices of the non-padding tokens in the input sequence. Used for unpadding the output.
+        cu_seqlens (`torch.Tensor` of shape `(batch + 1,)`, *optional*):
+            Cumulative sequence lengths of the input sequences. Used to index the unpadded tensors.
+        max_seqlen (`int`, *optional*):
+            Maximum sequence length in the batch excluding padding tokens. Used to unpad input_ids and pad output tensors.
+        batch_size (`int`, *optional*):
+            Batch size of the input sequences. Used to pad the output tensors.
+        seq_len (`int`, *optional*):
+            Sequence length of the input sequences including padding tokens. Used to pad the output tensors.
+        """  # noqa: E501
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
+
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attentions = () if output_attentions else None
+
+        self._maybe_set_compile()
+
+        if input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
+
+        if batch_size is None and seq_len is None:
+            if inputs_embeds is not None:
+                batch_size, seq_len = inputs_embeds.shape[:2]
+            else:
+                batch_size, seq_len = input_ids.shape[:2]  # type: ignore
+        device = input_ids.device if input_ids is not None else inputs_embeds.device  # type: ignore
+
+        if attention_mask is None:
+            attention_mask = torch.ones((batch_size, seq_len), device=device, dtype=torch.bool)  # type: ignore
+
+        repad = False
+        if self.config._attn_implementation == "flash_attention_2":
+            if indices is None and cu_seqlens is None and max_seqlen is None:
+                repad = True
+                if inputs_embeds is None:
+                    with torch.no_grad():
+                        input_ids, indices, cu_seqlens, max_seqlen, *_ = _unpad_modernbert_input(
+                            inputs=input_ids,  # type: ignore
+                            attention_mask=attention_mask,  # type: ignore
+                        )
+                else:
+                    inputs_embeds, indices, cu_seqlens, max_seqlen, *_ = _unpad_modernbert_input(
+                        inputs=inputs_embeds,
+                        attention_mask=attention_mask,  # type: ignore
+                    )
+        else:
+            if position_ids is None:
+                position_ids = torch.arange(seq_len, device=device).unsqueeze(0)  # type: ignore
+
+            attention_mask, sliding_window_mask = self._update_attention_mask(
+                attention_mask,  # type: ignore
+                output_attentions=output_attentions,  # type: ignore
+            )
+
+        hidden_states = self.embeddings(input_ids=input_ids, inputs_embeds=inputs_embeds)
+
+        for encoder_layer in self.layers:
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)  # type: ignore
+
+            layer_outputs = encoder_layer(
+                hidden_states,
+                attention_mask=attention_mask,
+                sliding_window_mask=sliding_window_mask,
+                position_ids=position_ids,
+                cu_seqlens=cu_seqlens,
+                max_seqlen=max_seqlen,
+                output_attentions=output_attentions,
+            )
+            hidden_states = layer_outputs[0]
+            if output_attentions and len(layer_outputs) > 1:
+                all_self_attentions = all_self_attentions + (layer_outputs[1],)  # type: ignore
+
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)  # type: ignore
+
+        hidden_states = self.final_norm(hidden_states)
+
+        if repad:
+            hidden_states = _pad_modernbert_output(
+                inputs=hidden_states,
+                indices=indices,  # type: ignore
+                batch=batch_size,  # type: ignore
+                seqlen=seq_len,  # type: ignore
+            )
+            if all_hidden_states is not None:
+                all_hidden_states = tuple(
+                    _pad_modernbert_output(inputs=hs, indices=indices, batch=batch_size, seqlen=seq_len)  # type: ignore
+                    for hs in all_hidden_states
+                )
+
+        if not return_dict:
+            return tuple(v for v in [hidden_states, all_hidden_states, all_self_attentions] if v is not None)
+        return BaseModelOutput(
+            last_hidden_state=hidden_states,  # type: ignore
+            hidden_states=all_hidden_states,  # type: ignore
+            attentions=all_self_attentions,
+        )
+
+    def _update_attention_mask(self, attention_mask: torch.Tensor, output_attentions: bool) -> torch.Tensor:
+        if output_attentions:
+            if self.config._attn_implementation == "sdpa":
+                logger.warning_once(  # type: ignore
+                    "Outputting attentions is only supported with the 'eager' attention implementation, "
+                    'not with "sdpa". Falling back to `attn_implementation="eager"`.'
+                )
+                self.config._attn_implementation = "eager"
+            elif self.config._attn_implementation != "eager":
+                logger.warning_once(  # type: ignore
+                    "Outputting attentions is only supported with the eager attention implementation, "
+                    f'not with {self.config._attn_implementation}. Consider setting `attn_implementation="eager"`.'
+                    " Setting `output_attentions=False`."
+                )
+
+        global_attention_mask = _prepare_4d_attention_mask(attention_mask, self.dtype)
+
+        # Create position indices
+        rows = torch.arange(global_attention_mask.shape[2]).unsqueeze(0)
+        # Calculate distance between positions
+        distance = torch.abs(rows - rows.T)
+
+        # Create sliding window mask (1 for positions within window, 0 outside)
+        window_mask = (distance <= self.config.local_attention // 2).unsqueeze(0).unsqueeze(0).to(attention_mask.device)
+        # Combine with existing mask
+        sliding_window_mask = global_attention_mask.masked_fill(window_mask.logical_not(), torch.finfo(self.dtype).min)
+
+        return global_attention_mask, sliding_window_mask  # type: ignore
+
+
+class ModChemBertForMaskedLM(InitWeightsMixin, ModernBertPreTrainedModel):
+    config_class = ModChemBertConfig
+    _tied_weights_keys = ["decoder.weight"]
+
+    def __init__(self, config: ModChemBertConfig):
+        super().__init__(config)
+        self.config = config
+        self.model = ModChemBertModel(config)
+        self.head = ModernBertPredictionHead(config)
+        self.decoder = nn.Linear(config.hidden_size, config.vocab_size, bias=config.decoder_bias)
+
+        self.sparse_prediction = self.config.sparse_prediction
+        self.sparse_pred_ignore_index = self.config.sparse_pred_ignore_index
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_output_embeddings(self):
+        return self.decoder
+
+    def set_output_embeddings(self, new_embeddings: nn.Linear):
+        self.decoder = new_embeddings
+
+    @torch.compile(dynamic=True)
+    def compiled_head(self, output: torch.Tensor) -> torch.Tensor:
+        return self.decoder(self.head(output))
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor | None = None,
+        attention_mask: torch.Tensor | None = None,
+        sliding_window_mask: torch.Tensor | None = None,
+        position_ids: torch.Tensor | None = None,
+        inputs_embeds: torch.Tensor | None = None,
+        labels: torch.Tensor | None = None,
+        indices: torch.Tensor | None = None,
+        cu_seqlens: torch.Tensor | None = None,
+        max_seqlen: int | None = None,
+        batch_size: int | None = None,
+        seq_len: int | None = None,
+        output_attentions: bool | None = None,
+        output_hidden_states: bool | None = None,
+        return_dict: bool | None = None,
+        **kwargs,
+    ) -> tuple[torch.Tensor] | tuple[torch.Tensor, typing.Any] | MaskedLMOutput:
+        r"""
+        sliding_window_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Mask to avoid performing attention on padding or far-away tokens. In ModernBert, only every few layers
+            perform global attention, while the rest perform local attention. This mask is used to avoid attending to
+            far-away tokens in the local attention layers when not using Flash Attention.
+        indices (`torch.Tensor` of shape `(total_unpadded_tokens,)`, *optional*):
+            Indices of the non-padding tokens in the input sequence. Used for unpadding the output.
+        cu_seqlens (`torch.Tensor` of shape `(batch + 1,)`, *optional*):
+            Cumulative sequence lengths of the input sequences. Used to index the unpadded tensors.
+        max_seqlen (`int`, *optional*):
+            Maximum sequence length in the batch excluding padding tokens. Used to unpad input_ids & pad output tensors.
+        batch_size (`int`, *optional*):
+            Batch size of the input sequences. Used to pad the output tensors.
+        seq_len (`int`, *optional*):
+            Sequence length of the input sequences including padding tokens. Used to pad the output tensors.
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        self._maybe_set_compile()
+
+        if self.config._attn_implementation == "flash_attention_2":  # noqa: SIM102
+            if indices is None and cu_seqlens is None and max_seqlen is None:
+                if batch_size is None and seq_len is None:
+                    if inputs_embeds is not None:
+                        batch_size, seq_len = inputs_embeds.shape[:2]
+                    else:
+                        batch_size, seq_len = input_ids.shape[:2]  # type: ignore
+                device = input_ids.device if input_ids is not None else inputs_embeds.device  # type: ignore
+
+                if attention_mask is None:
+                    attention_mask = torch.ones((batch_size, seq_len), device=device, dtype=torch.bool)  # type: ignore
+
+                if inputs_embeds is None:
+                    with torch.no_grad():
+                        input_ids, indices, cu_seqlens, max_seqlen, position_ids, labels = _unpad_modernbert_input(
+                            inputs=input_ids,  # type: ignore
+                            attention_mask=attention_mask,  # type: ignore
+                            position_ids=position_ids,
+                            labels=labels,
+                        )
+                else:
+                    inputs_embeds, indices, cu_seqlens, max_seqlen, position_ids, labels = _unpad_modernbert_input(
+                        inputs=inputs_embeds,
+                        attention_mask=attention_mask,  # type: ignore
+                        position_ids=position_ids,
+                        labels=labels,
+                    )
+
+        outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            sliding_window_mask=sliding_window_mask,
+            position_ids=position_ids,
+            inputs_embeds=inputs_embeds,
+            indices=indices,
+            cu_seqlens=cu_seqlens,
+            max_seqlen=max_seqlen,
+            batch_size=batch_size,
+            seq_len=seq_len,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        last_hidden_state = outputs[0]
+
+        if self.sparse_prediction and labels is not None:
+            # flatten labels and output first
+            labels = labels.view(-1)
+            last_hidden_state = last_hidden_state.view(labels.shape[0], -1)
+
+            # then filter out the non-masked tokens
+            mask_tokens = labels != self.sparse_pred_ignore_index
+            last_hidden_state = last_hidden_state[mask_tokens]
+            labels = labels[mask_tokens]
+
+        logits = (
+            self.compiled_head(last_hidden_state)
+            if self.config.reference_compile
+            else self.decoder(self.head(last_hidden_state))
+        )
+
+        loss = None
+        if labels is not None:
+            loss = self.loss_function(logits, labels, vocab_size=self.config.vocab_size, **kwargs)
+
+        if self.config._attn_implementation == "flash_attention_2":
+            with nullcontext() if self.config.repad_logits_with_grad or labels is None else torch.no_grad():
+                logits = _pad_modernbert_output(inputs=logits, indices=indices, batch=batch_size, seqlen=seq_len)  # type: ignore
+
+        if not return_dict:
+            output = (logits,)
+            return ((loss,) + output) if loss is not None else output
+
+        return MaskedLMOutput(
+            loss=loss,
+            logits=typing.cast(torch.FloatTensor, logits),
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+class ModChemBertForSequenceClassification(InitWeightsMixin, ModernBertPreTrainedModel):
+    config_class = ModChemBertConfig
+
+    def __init__(self, config: ModChemBertConfig):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+        self.config = config
+
+        self.model = ModernBertModel(config)
+        if self.config.classifier_pooling in {"cls_mha", "max_seq_mha"}:
+            self.pooling_attn = ModChemBertPoolingAttention(config=self.config)
+        else:
+            self.pooling_attn = None
+        self.head = ModernBertPredictionHead(config)
+        self.drop = torch.nn.Dropout(config.classifier_dropout)
+        self.classifier = nn.Linear(config.hidden_size, config.num_labels)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor | None = None,
+        attention_mask: torch.Tensor | None = None,
+        sliding_window_mask: torch.Tensor | None = None,
+        position_ids: torch.Tensor | None = None,
+        inputs_embeds: torch.Tensor | None = None,
+        labels: torch.Tensor | None = None,
+        indices: torch.Tensor | None = None,
+        cu_seqlens: torch.Tensor | None = None,
+        max_seqlen: int | None = None,
+        batch_size: int | None = None,
+        seq_len: int | None = None,
+        output_attentions: bool | None = None,
+        output_hidden_states: bool | None = None,
+        return_dict: bool | None = None,
+        **kwargs,
+    ) -> tuple[torch.Tensor] | tuple[torch.Tensor, typing.Any] | SequenceClassifierOutput:
+        r"""
+        sliding_window_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Mask to avoid performing attention on padding or far-away tokens. In ModernBert, only every few layers
+            perform global attention, while the rest perform local attention. This mask is used to avoid attending to
+            far-away tokens in the local attention layers when not using Flash Attention.
+        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
+            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
+            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
+        indices (`torch.Tensor` of shape `(total_unpadded_tokens,)`, *optional*):
+            Indices of the non-padding tokens in the input sequence. Used for unpadding the output.
+        cu_seqlens (`torch.Tensor` of shape `(batch + 1,)`, *optional*):
+            Cumulative sequence lengths of the input sequences. Used to index the unpadded tensors.
+        max_seqlen (`int`, *optional*):
+            Maximum sequence length in the batch excluding padding tokens. Used to unpad input_ids & pad output tensors.
+        batch_size (`int`, *optional*):
+            Batch size of the input sequences. Used to pad the output tensors.
+        seq_len (`int`, *optional*):
+            Sequence length of the input sequences including padding tokens. Used to pad the output tensors.
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        self._maybe_set_compile()
+
+        if input_ids is not None:
+            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
+
+        if batch_size is None and seq_len is None:
+            if inputs_embeds is not None:
+                batch_size, seq_len = inputs_embeds.shape[:2]
+            else:
+                batch_size, seq_len = input_ids.shape[:2]  # type: ignore
+        device = input_ids.device if input_ids is not None else inputs_embeds.device  # type: ignore
+
+        if attention_mask is None:
+            attention_mask = torch.ones((batch_size, seq_len), device=device, dtype=torch.bool)  # type: ignore
+
+        # Ensure output_hidden_states is True in case pooling mode requires all hidden states
+        output_hidden_states = True
+
+        outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            sliding_window_mask=sliding_window_mask,
+            position_ids=position_ids,
+            inputs_embeds=inputs_embeds,
+            indices=indices,
+            cu_seqlens=cu_seqlens,
+            max_seqlen=max_seqlen,
+            batch_size=batch_size,
+            seq_len=seq_len,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        last_hidden_state = outputs[0]
+        hidden_states = outputs[1]
+
+        last_hidden_state = _pool_modchembert_output(
+            self,
+            last_hidden_state,
+            hidden_states,
+            typing.cast(torch.Tensor, attention_mask),
+        )
+        pooled_output = self.head(last_hidden_state)
+        pooled_output = self.drop(pooled_output)
+        logits = self.classifier(pooled_output)
+
+        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 = 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 = CrossEntropyLoss()
+                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
+            elif self.config.problem_type == "multi_label_classification":
+                loss_fct = BCEWithLogitsLoss()
+                loss = loss_fct(logits, labels)
+
+        if not return_dict:
+            output = (logits,)
+            return ((loss,) + output) if loss is not None else output
+
+        return SequenceClassifierOutput(
+            loss=loss,
+            logits=logits,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+def _pool_modchembert_output(
+    module: ModChemBertForSequenceClassification,
+    last_hidden_state: torch.Tensor,
+    hidden_states: list[torch.Tensor],
+    attention_mask: torch.Tensor,
+):
+    """
+    Apply pooling strategy to hidden states for sequence-level classification/regression tasks.
+
+    This function implements various pooling strategies to aggregate sequence representations
+    into a single vector for downstream classification or regression tasks. The pooling method
+    is determined by the `classifier_pooling` configuration parameter.
+
+    Available pooling strategies:
+    - cls: Use the CLS token ([CLS]) representation from the last hidden state
+    - mean: Average pooling over all tokens in the sequence (attention-weighted)
+    - max_cls: Element-wise max pooling over the last k hidden states, then take CLS token
+    - cls_mha: Multi-head attention with CLS token as query and full sequence as keys/values
+    - max_seq_mha: Max pooling over last k states + multi-head attention with CLS as query
+    - max_seq_mean: Max pooling over last k hidden states, then mean pooling over sequence
+    - sum_mean: Sum all hidden states across layers, then mean pool over sequence
+    - sum_sum: Sum all hidden states across layers, then sum pool over sequence
+    - mean_sum: Mean all hidden states across layers, then sum pool over sequence
+    - mean_mean: Mean all hidden states across layers, then mean pool over sequence
+
+    Args:
+        module: The model instance containing configuration and pooling attention if needed
+        last_hidden_state: Final layer hidden states of shape (batch_size, seq_len, hidden_size)
+        hidden_states: List of hidden states from all layers, each of shape (batch_size, seq_len, hidden_size)
+        attention_mask: Attention mask of shape (batch_size, seq_len) indicating valid tokens
+
+    Returns:
+        torch.Tensor: Pooled representation of shape (batch_size, hidden_size)
+
+    Note:
+        Some pooling strategies (cls_mha, max_seq_mha) require the module to have a pooling_attn
+        attribute containing a ModChemBertPoolingAttention instance.
+    """
+    config = typing.cast(ModChemBertConfig, module.config)
+    if config.classifier_pooling == "cls":
+        last_hidden_state = last_hidden_state[:, 0]
+    elif config.classifier_pooling == "mean":
+        last_hidden_state = (last_hidden_state * attention_mask.unsqueeze(-1)).sum(dim=1) / attention_mask.sum(
+            dim=1, keepdim=True
+        )
+    elif config.classifier_pooling == "max_cls":
+        k_hidden_states = hidden_states[-config.classifier_pooling_last_k :]
+        theta = torch.stack(k_hidden_states, dim=1)  # (batch, k, seq_len, hidden)
+        pooled_seq = torch.max(theta, dim=1).values  # Element-wise max over k -> (batch, seq_len, hidden)
+        last_hidden_state = pooled_seq[:, 0, :]  # (batch, hidden)
+    elif config.classifier_pooling == "cls_mha":
+        # Similar to max_seq_mha but without the max pooling step
+        # Query is CLS token (position 0); Keys/Values are full sequence
+        q = last_hidden_state[:, 0, :].unsqueeze(1)  # (batch, 1, hidden)
+        q = q.expand(-1, last_hidden_state.shape[1], -1)  # (batch, seq_len, hidden)
+        attn_out: torch.Tensor = module.pooling_attn(  # type: ignore
+            q=q, kv=last_hidden_state, attention_mask=attention_mask
+        )  # (batch, seq_len, hidden)
+        last_hidden_state = torch.mean(attn_out, dim=1)
+    elif config.classifier_pooling == "max_seq_mha":
+        k_hidden_states = hidden_states[-config.classifier_pooling_last_k :]
+        theta = torch.stack(k_hidden_states, dim=1)  # (batch, k, seq_len, hidden)
+        pooled_seq = torch.max(theta, dim=1).values  # Element-wise max over k -> (batch, seq_len, hidden)
+        # Query is pooled CLS token (position 0); Keys/Values are pooled sequence
+        q = pooled_seq[:, 0, :].unsqueeze(1)  # (batch, 1, hidden)
+        q = q.expand(-1, pooled_seq.shape[1], -1)  # (batch, seq_len, hidden)
+        attn_out: torch.Tensor = module.pooling_attn(  # type: ignore
+            q=q, kv=pooled_seq, attention_mask=attention_mask
+        )  # (batch, seq_len, hidden)
+        last_hidden_state = torch.mean(attn_out, dim=1)
+    elif config.classifier_pooling == "max_seq_mean":
+        k_hidden_states = hidden_states[-config.classifier_pooling_last_k :]
+        theta = torch.stack(k_hidden_states, dim=1)  # (batch, k, seq_len, hidden)
+        pooled_seq = torch.max(theta, dim=1).values  # Element-wise max over k -> (batch, seq_len, hidden)
+        last_hidden_state = torch.mean(pooled_seq, dim=1)  # Mean over sequence length
+    elif config.classifier_pooling == "sum_mean":
+        # ChemLM uses the mean of all hidden states
+        # which outperforms using just the last layer mean or the cls embedding
+        # https://doi.org/10.1038/s42004-025-01484-4
+        # https://static-content.springer.com/esm/art%3A10.1038%2Fs42004-025-01484-4/MediaObjects/42004_2025_1484_MOESM2_ESM.pdf
+        all_hidden_states = torch.stack(hidden_states)
+        w = torch.sum(all_hidden_states, dim=0)
+        last_hidden_state = torch.mean(w, dim=1)
+    elif config.classifier_pooling == "sum_sum":
+        all_hidden_states = torch.stack(hidden_states)
+        w = torch.sum(all_hidden_states, dim=0)
+        last_hidden_state = torch.sum(w, dim=1)
+    elif config.classifier_pooling == "mean_sum":
+        all_hidden_states = torch.stack(hidden_states)
+        w = torch.mean(all_hidden_states, dim=0)
+        last_hidden_state = torch.sum(w, dim=1)
+    elif config.classifier_pooling == "mean_mean":
+        all_hidden_states = torch.stack(hidden_states)
+        w = torch.mean(all_hidden_states, dim=0)
+        last_hidden_state = torch.mean(w, dim=1)
+    return last_hidden_state
+
+
+__all__ = [
+    "ModChemBertForMaskedLM",
+    "ModChemBertForSequenceClassification",
+]

special_tokens_map.json

@@ -0,0 +1,37 @@
+{
+  "cls_token": {
+    "content": "[CLS]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "mask_token": {
+    "content": "[MASK]",
+    "lstrip": true,
+    "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.json

@@ -0,0 +1,2542 @@
+{
+  "version": "1.0",
+  "truncation": null,
+  "padding": null,
+  "added_tokens": [
+    {
+      "id": 0,
+      "content": "[CLS]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 1,
+      "content": "[SEP]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 2,
+      "content": "[PAD]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 3,
+      "content": "[MASK]",
+      "single_word": false,
+      "lstrip": true,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 2361,
+      "content": "[UNK]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    }
+  ],
+  "normalizer": null,
+  "pre_tokenizer": {
+    "type": "ByteLevel",
+    "add_prefix_space": false,
+    "trim_offsets": true,
+    "use_regex": true
+  },
+  "post_processor": {
+    "type": "TemplateProcessing",
+    "single": [
+      {
+        "SpecialToken": {
+          "id": "[CLS]",
+          "type_id": 0
+        }
+      },
+      {
+        "Sequence": {
+          "id": "A",
+          "type_id": 0
+        }
+      },
+      {
+        "SpecialToken": {
+          "id": "[SEP]",
+          "type_id": 0
+        }
+      }
+    ],
+    "pair": [
+      {
+        "SpecialToken": {
+          "id": "[CLS]",
+          "type_id": 0
+        }
+      },
+      {
+        "Sequence": {
+          "id": "A",
+          "type_id": 0
+        }
+      },
+      {
+        "SpecialToken": {
+          "id": "[SEP]",
+          "type_id": 0
+        }
+      },
+      {
+        "Sequence": {
+          "id": "B",
+          "type_id": 0
+        }
+      },
+      {
+        "SpecialToken": {
+          "id": "[SEP]",
+          "type_id": 0
+        }
+      }
+    ],
+    "special_tokens": {
+      "[CLS]": {
+        "id": "[CLS]",
+        "ids": [
+          0
+        ],
+        "tokens": [
+          "[CLS]"
+        ]
+      },
+      "[MASK]": {
+        "id": "[MASK]",
+        "ids": [
+          3
+        ],
+        "tokens": [
+          "[MASK]"
+        ]
+      },
+      "[PAD]": {
+        "id": "[PAD]",
+        "ids": [
+          2
+        ],
+        "tokens": [
+          "[PAD]"
+        ]
+      },
+      "[SEP]": {
+        "id": "[SEP]",
+        "ids": [
+          1
+        ],
+        "tokens": [
+          "[SEP]"
+        ]
+      },
+      "[UNK]": {
+        "id": "[UNK]",
+        "ids": [
+          2361
+        ],
+        "tokens": [
+          "[UNK]"
+        ]
+      }
+    }
+  },
+  "decoder": {
+    "type": "ByteLevel",
+    "add_prefix_space": false,
+    "trim_offsets": true,
+    "use_regex": true
+  },
+  "model": {
+    "type": "BPE",
+    "dropout": null,
+    "unk_token": "[UNK]",
+    "continuing_subword_prefix": null,
+    "end_of_word_suffix": null,
+    "fuse_unk": false,
+    "byte_fallback": false,
+    "ignore_merges": false,
+    "vocab": {
+      "[CLS]": 0,
+      "[SEP]": 1,
+      "[PAD]": 2,
+      "[MASK]": 3,
+      "C": 4,
+      "c": 5,
+      "(": 6,
+      ")": 7,
+      "1": 8,
+      "O": 9,
+      "N": 10,
+      "2": 11,
+      "=": 12,
+      "n": 13,
+      "3": 14,
+      "[C@H]": 15,
+      "[C@@H]": 16,
+      "F": 17,
+      "S": 18,
+      "4": 19,
+      "Cl": 20,
+      "-": 21,
+      "o": 22,
+      "s": 23,
+      "[nH]": 24,
+      "#": 25,
+      "/": 26,
+      "Br": 27,
+      "[C@]": 28,
+      "[C@@]": 29,
+      "[N+]": 30,
+      "[O-]": 31,
+      "5": 32,
+      "\\": 33,
+      ".": 34,
+      "I": 35,
+      "6": 36,
+      "[S@]": 37,
+      "[S@@]": 38,
+      "P": 39,
+      "[N-]": 40,
+      "[Si]": 41,
+      "7": 42,
+      "[n+]": 43,
+      "[2H]": 44,
+      "8": 45,
+      "[NH+]": 46,
+      "B": 47,
+      "9": 48,
+      "[C-]": 49,
+      "[Na+]": 50,
+      "[Cl-]": 51,
+      "[c-]": 52,
+      "[CH]": 53,
+      "%10": 54,
+      "[NH2+]": 55,
+      "[P+]": 56,
+      "[B]": 57,
+      "[I-]": 58,
+      "%11": 59,
+      "[CH2-]": 60,
+      "[O+]": 61,
+      "[NH3+]": 62,
+      "[C]": 63,
+      "[Br-]": 64,
+      "[IH2]": 65,
+      "[S-]": 66,
+      "[cH-]": 67,
+      "%12": 68,
+      "[nH+]": 69,
+      "[B-]": 70,
+      "[K+]": 71,
+      "[Sn]": 72,
+      "[Se]": 73,
+      "[CH-]": 74,
+      "[HH]": 75,
+      "[Y]": 76,
+      "[n-]": 77,
+      "[CH3-]": 78,
+      "[SiH]": 79,
+      "[S+]": 80,
+      "%13": 81,
+      "[SiH2]": 82,
+      "[Li+]": 83,
+      "[NH-]": 84,
+      "%14": 85,
+      "[Na]": 86,
+      "[CH2]": 87,
+      "[O-2]": 88,
+      "[U+2]": 89,
+      "[W]": 90,
+      "[Al]": 91,
+      "[P@]": 92,
+      "[Fe+2]": 93,
+      "[PH+]": 94,
+      "%15": 95,
+      "[Cl+3]": 96,
+      "[Zn+2]": 97,
+      "[Ir]": 98,
+      "[Mg+2]": 99,
+      "[Pt+2]": 100,
+      "[OH2+]": 101,
+      "[As]": 102,
+      "[Fe]": 103,
+      "[OH+]": 104,
+      "[Zr+2]": 105,
+      "[3H]": 106,
+      "[Ge]": 107,
+      "[SiH3]": 108,
+      "[OH-]": 109,
+      "[NH4+]": 110,
+      "[Cu+2]": 111,
+      "[P@@]": 112,
+      "p": 113,
+      "[Pt]": 114,
+      "%16": 115,
+      "[Ca+2]": 116,
+      "[Zr]": 117,
+      "[F-]": 118,
+      "[C+]": 119,
+      "[Ti]": 120,
+      "[P-]": 121,
+      "[V]": 122,
+      "[se]": 123,
+      "[U]": 124,
+      "[O]": 125,
+      "[Ni+2]": 126,
+      "[Zn]": 127,
+      "[Co]": 128,
+      "[Ni]": 129,
+      "[Pd+2]": 130,
+      "[Cu]": 131,
+      "%17": 132,
+      "[Cu+]": 133,
+      "[Te]": 134,
+      "[H+]": 135,
+      "[CH+]": 136,
+      "[Li]": 137,
+      "[Pd]": 138,
+      "[Mo]": 139,
+      "[Ru+2]": 140,
+      "[o+]": 141,
+      "[Re]": 142,
+      "[SH+]": 143,
+      "%18": 144,
+      "[Ac]": 145,
+      "[Cr]": 146,
+      "[NH2-]": 147,
+      "[K]": 148,
+      "[13CH2]": 149,
+      "[c]": 150,
+      "[Zr+4]": 151,
+      "[Tl]": 152,
+      "[13C]": 153,
+      "[Mn]": 154,
+      "[N@+]": 155,
+      "[Hg]": 156,
+      "[Rh]": 157,
+      "[Ti+4]": 158,
+      "[Sb]": 159,
+      "[Co+2]": 160,
+      "[Ag+]": 161,
+      "[Ru]": 162,
+      "%19": 163,
+      "[N@@+]": 164,
+      "[Ti+2]": 165,
+      "[Al+3]": 166,
+      "[Pb]": 167,
+      "[I+]": 168,
+      "[18F]": 169,
+      "[s+]": 170,
+      "[Rb+]": 171,
+      "[Ba+2]": 172,
+      "[H-]": 173,
+      "[Fe+3]": 174,
+      "[Ir+3]": 175,
+      "[13cH]": 176,
+      "%20": 177,
+      "[AlH2]": 178,
+      "[Au+]": 179,
+      "[13c]": 180,
+      "[SH2+]": 181,
+      "[Sn+2]": 182,
+      "[Mn+2]": 183,
+      "[Si-]": 184,
+      "[Ag]": 185,
+      "[N]": 186,
+      "[Bi]": 187,
+      "%21": 188,
+      "[In]": 189,
+      "[CH2+]": 190,
+      "[Y+3]": 191,
+      "[Ga]": 192,
+      "%22": 193,
+      "[Co+3]": 194,
+      "[Au]": 195,
+      "[13CH3]": 196,
+      "[Mg]": 197,
+      "[Cs+]": 198,
+      "[W+2]": 199,
+      "[Hf]": 200,
+      "[Zn+]": 201,
+      "[Se-]": 202,
+      "[S-2]": 203,
+      "[Ca]": 204,
+      "[pH]": 205,
+      "[ClH+]": 206,
+      "[Ti+3]": 207,
+      "%23": 208,
+      "[Ru+]": 209,
+      "[SH-]": 210,
+      "[13CH]": 211,
+      "[IH+]": 212,
+      "[Hf+4]": 213,
+      "[Rf]": 214,
+      "[OH3+]": 215,
+      "%24": 216,
+      "[Pt+4]": 217,
+      "[Zr+3]": 218,
+      "[PH3+]": 219,
+      "[Sr+2]": 220,
+      "[Cd+2]": 221,
+      "[Cd]": 222,
+      "%25": 223,
+      "[Os]": 224,
+      "[BH-]": 225,
+      "[Sn+4]": 226,
+      "[Cr+3]": 227,
+      "[Ru+3]": 228,
+      "[PH2+]": 229,
+      "[Rh+2]": 230,
+      "[V+2]": 231,
+      "%26": 232,
+      "[Gd+3]": 233,
+      "[Pb+2]": 234,
+      "[PH]": 235,
+      "[Hg+]": 236,
+      "[Mo+2]": 237,
+      "[AlH]": 238,
+      "[Sn+]": 239,
+      "%27": 240,
+      "[Pd+]": 241,
+      "b": 242,
+      "[Rh+3]": 243,
+      "[Hg+2]": 244,
+      "[15NH]": 245,
+      "[14C]": 246,
+      "%28": 247,
+      "[Mn+3]": 248,
+      "[Si+]": 249,
+      "[SeH]": 250,
+      "[13C@H]": 251,
+      "[NH]": 252,
+      "[Ga+3]": 253,
+      "[SiH-]": 254,
+      "[13C@@H]": 255,
+      "[Ce]": 256,
+      "[Au+3]": 257,
+      "[Bi+3]": 258,
+      "[15N]": 259,
+      "%29": 260,
+      "[BH3-]": 261,
+      "[14cH]": 262,
+      "[Ti+]": 263,
+      "[Gd]": 264,
+      "[cH+]": 265,
+      "[Cr+2]": 266,
+      "[Sb-]": 267,
+      "%30": 268,
+      "[Be+2]": 269,
+      "[Al+]": 270,
+      "[te]": 271,
+      "[11CH3]": 272,
+      "[Sm]": 273,
+      "[Pr]": 274,
+      "[La]": 275,
+      "%31": 276,
+      "[Al-]": 277,
+      "[Ta]": 278,
+      "[125I]": 279,
+      "[BH2-]": 280,
+      "[Nb]": 281,
+      "[Si@]": 282,
+      "%32": 283,
+      "[14c]": 284,
+      "[Sb+3]": 285,
+      "[Ba]": 286,
+      "%33": 287,
+      "[Os+2]": 288,
+      "[Si@@]": 289,
+      "[La+3]": 290,
+      "[15n]": 291,
+      "[15NH2]": 292,
+      "[Nd+3]": 293,
+      "%34": 294,
+      "[14CH2]": 295,
+      "[18O]": 296,
+      "[Nd]": 297,
+      "[GeH]": 298,
+      "[Ni+3]": 299,
+      "[Eu]": 300,
+      "[Dy+3]": 301,
+      "[Sc]": 302,
+      "%36": 303,
+      "[Se-2]": 304,
+      "[As+]": 305,
+      "%35": 306,
+      "[AsH]": 307,
+      "[Tb]": 308,
+      "[Sb+5]": 309,
+      "[Se+]": 310,
+      "[Ce+3]": 311,
+      "[c+]": 312,
+      "[In+3]": 313,
+      "[SnH]": 314,
+      "[Mo+4]": 315,
+      "%37": 316,
+      "[V+4]": 317,
+      "[Eu+3]": 318,
+      "[Hf+2]": 319,
+      "%38": 320,
+      "[Pt+]": 321,
+      "[p+]": 322,
+      "[123I]": 323,
+      "[Tl+]": 324,
+      "[Sm+3]": 325,
+      "%39": 326,
+      "[Yb+3]": 327,
+      "%40": 328,
+      "[Yb]": 329,
+      "[Os+]": 330,
+      "%41": 331,
+      "[10B]": 332,
+      "[Sc+3]": 333,
+      "[Al+2]": 334,
+      "%42": 335,
+      "[Sr]": 336,
+      "[Tb+3]": 337,
+      "[Po]": 338,
+      "[Tc]": 339,
+      "[PH-]": 340,
+      "[AlH3]": 341,
+      "[Ar]": 342,
+      "[U+4]": 343,
+      "[SnH2]": 344,
+      "[Cl+2]": 345,
+      "[si]": 346,
+      "[Fe+]": 347,
+      "[14CH3]": 348,
+      "[U+3]": 349,
+      "[Cl+]": 350,
+      "%43": 351,
+      "[GeH2]": 352,
+      "%44": 353,
+      "[Er+3]": 354,
+      "[Mo+3]": 355,
+      "[I+2]": 356,
+      "[Fe+4]": 357,
+      "[99Tc]": 358,
+      "%45": 359,
+      "[11C]": 360,
+      "%46": 361,
+      "[SnH3]": 362,
+      "[S]": 363,
+      "[Te+]": 364,
+      "[Er]": 365,
+      "[Lu+3]": 366,
+      "[11B]": 367,
+      "%47": 368,
+      "%48": 369,
+      "[P]": 370,
+      "[Tm]": 371,
+      "[Th]": 372,
+      "[Dy]": 373,
+      "[Pr+3]": 374,
+      "[Ta+5]": 375,
+      "[Nb+5]": 376,
+      "[Rb]": 377,
+      "[GeH3]": 378,
+      "[Br+2]": 379,
+      "%49": 380,
+      "[131I]": 381,
+      "[Fm]": 382,
+      "[Cs]": 383,
+      "[BH4-]": 384,
+      "[Lu]": 385,
+      "[15nH]": 386,
+      "%50": 387,
+      "[Ru+6]": 388,
+      "[b-]": 389,
+      "[Ho]": 390,
+      "[Th+4]": 391,
+      "[Ru+4]": 392,
+      "%52": 393,
+      "[14CH]": 394,
+      "%51": 395,
+      "[Cr+6]": 396,
+      "[18OH]": 397,
+      "[Ho+3]": 398,
+      "[Ce+4]": 399,
+      "[Bi+2]": 400,
+      "[Co+]": 401,
+      "%53": 402,
+      "[Yb+2]": 403,
+      "[Fe+6]": 404,
+      "[Be]": 405,
+      "%54": 406,
+      "[SH3+]": 407,
+      "[Np]": 408,
+      "[As-]": 409,
+      "%55": 410,
+      "[14C@@H]": 411,
+      "[Ir+2]": 412,
+      "[GaH3]": 413,
+      "[p-]": 414,
+      "[GeH4]": 415,
+      "[Sn+3]": 416,
+      "[Os+4]": 417,
+      "%56": 418,
+      "[14C@H]": 419,
+      "[sH+]": 420,
+      "[19F]": 421,
+      "[Eu+2]": 422,
+      "[TlH]": 423,
+      "%57": 424,
+      "[Cr+4]": 425,
+      "%58": 426,
+      "[B@@-]": 427,
+      "[SiH+]": 428,
+      "[At]": 429,
+      "[Am]": 430,
+      "[Fe+5]": 431,
+      "[AsH2]": 432,
+      "[Si+4]": 433,
+      "[B@-]": 434,
+      "[Pu]": 435,
+      "[SbH]": 436,
+      "[P-2]": 437,
+      "[Tm+3]": 438,
+      "*": 439,
+      "%59": 440,
+      "[se+]": 441,
+      "%60": 442,
+      "[oH+]": 443,
+      "[1H]": 444,
+      "[15N+]": 445,
+      "[124I]": 446,
+      "[S@@+]": 447,
+      "[P-3]": 448,
+      "[H]": 449,
+      "[IH2+]": 450,
+      "[TeH]": 451,
+      "[Xe]": 452,
+      "[PH4+]": 453,
+      "[Cr+]": 454,
+      "[Cm]": 455,
+      "[I+3]": 456,
+      "%61": 457,
+      "[Nb+2]": 458,
+      "[Ru+5]": 459,
+      "%62": 460,
+      "[Ta+2]": 461,
+      "[Tc+4]": 462,
+      "[CH3+]": 463,
+      "[Pm]": 464,
+      "[Si@H]": 465,
+      "[No]": 466,
+      "%63": 467,
+      "[Cr+5]": 468,
+      "[Th+2]": 469,
+      "[Zn-2]": 470,
+      "[13C@]": 471,
+      "[Lr]": 472,
+      "%64": 473,
+      "[99Tc+3]": 474,
+      "%65": 475,
+      "[13C@@]": 476,
+      "%66": 477,
+      "[Fe-]": 478,
+      "[17O]": 479,
+      "[siH]": 480,
+      "[Sb+]": 481,
+      "[OH]": 482,
+      "[IH]": 483,
+      "[11CH2]": 484,
+      "[Cf]": 485,
+      "[SiH2+]": 486,
+      "[Gd+2]": 487,
+      "[In+]": 488,
+      "[Si@@H]": 489,
+      "[Mn+]": 490,
+      "[99Tc+4]": 491,
+      "[Ga-]": 492,
+      "%67": 493,
+      "[S@+]": 494,
+      "[Ge+4]": 495,
+      "[Tl+3]": 496,
+      "[16OH]": 497,
+      "%68": 498,
+      "[2H-]": 499,
+      "[Ra]": 500,
+      "[si-]": 501,
+      "[NiH2]": 502,
+      "[P@@H]": 503,
+      "[Rh+]": 504,
+      "[12C]": 505,
+      "[35S]": 506,
+      "[32P]": 507,
+      "[SiH2-]": 508,
+      "[AlH2+]": 509,
+      "[16O]": 510,
+      "%69": 511,
+      "[BiH]": 512,
+      "[BiH2]": 513,
+      "[Zn-]": 514,
+      "[BH]": 515,
+      "[Tc+3]": 516,
+      "[Ir+]": 517,
+      "[Ni+]": 518,
+      "%70": 519,
+      "[InH2]": 520,
+      "[InH]": 521,
+      "[Nb+3]": 522,
+      "[PbH]": 523,
+      "[Bi+]": 524,
+      "%71": 525,
+      "[As+3]": 526,
+      "%72": 527,
+      "[18O-]": 528,
+      "[68Ga+3]": 529,
+      "%73": 530,
+      "[Pa]": 531,
+      "[76Br]": 532,
+      "[Tc+5]": 533,
+      "[pH+]": 534,
+      "[64Cu+2]": 535,
+      "[Ru+8]": 536,
+      "%74": 537,
+      "[PH2-]": 538,
+      "[Si+2]": 539,
+      "[17OH]": 540,
+      "[RuH]": 541,
+      "[111In+3]": 542,
+      "[AlH+]": 543,
+      "%75": 544,
+      "%76": 545,
+      "[W+]": 546,
+      "[SbH2]": 547,
+      "[PoH]": 548,
+      "[Ru-]": 549,
+      "[XeH]": 550,
+      "[Tc+2]": 551,
+      "[13C-]": 552,
+      "[Br+]": 553,
+      "[Pt-2]": 554,
+      "[Es]": 555,
+      "[Cu-]": 556,
+      "[Mg+]": 557,
+      "[3HH]": 558,
+      "[P@H]": 559,
+      "[ClH2+]": 560,
+      "%77": 561,
+      "[SH]": 562,
+      "[Au-]": 563,
+      "[2HH]": 564,
+      "%78": 565,
+      "[Sn-]": 566,
+      "[11CH]": 567,
+      "[PdH2]": 568,
+      "0": 569,
+      "[Os+6]": 570,
+      "%79": 571,
+      "[Mo+]": 572,
+      "%80": 573,
+      "[al]": 574,
+      "[PbH2]": 575,
+      "[64Cu]": 576,
+      "[Cl]": 577,
+      "[12CH3]": 578,
+      "%81": 579,
+      "[Tc+7]": 580,
+      "[11c]": 581,
+      "%82": 582,
+      "[Li-]": 583,
+      "[99Tc+5]": 584,
+      "[He]": 585,
+      "[12c]": 586,
+      "[Kr]": 587,
+      "[RuH+2]": 588,
+      "[35Cl]": 589,
+      "[Pd-2]": 590,
+      "[GaH2]": 591,
+      "[4H]": 592,
+      "[Sg]": 593,
+      "[Cu-2]": 594,
+      "[Br+3]": 595,
+      "%83": 596,
+      "[37Cl]": 597,
+      "[211At]": 598,
+      "[IrH+2]": 599,
+      "[Mt]": 600,
+      "[Ir-2]": 601,
+      "[In-]": 602,
+      "[12cH]": 603,
+      "[12CH2]": 604,
+      "[RuH2]": 605,
+      "[99Tc+7]": 606,
+      "%84": 607,
+      "[15n+]": 608,
+      "[ClH2+2]": 609,
+      "[16N]": 610,
+      "[111In]": 611,
+      "[Tc+]": 612,
+      "[Ru-2]": 613,
+      "[12CH]": 614,
+      "[si+]": 615,
+      "[Tc+6]": 616,
+      "%85": 617,
+      "%86": 618,
+      "[90Y]": 619,
+      "[Pd-]": 620,
+      "[188Re]": 621,
+      "[RuH+]": 622,
+      "[NiH]": 623,
+      "[SiH3-]": 624,
+      "[14n]": 625,
+      "[CH3]": 626,
+      "[14N]": 627,
+      "[10BH2]": 628,
+      "%88": 629,
+      "%89": 630,
+      "%90": 631,
+      "[34S]": 632,
+      "[77Br]": 633,
+      "[GaH]": 634,
+      "[Br]": 635,
+      "[Ge@]": 636,
+      "[B@@H-]": 637,
+      "[CuH]": 638,
+      "[SiH4]": 639,
+      "[3H-]": 640,
+      "%87": 641,
+      "%91": 642,
+      "%92": 643,
+      "[67Cu]": 644,
+      "[I]": 645,
+      "[177Lu]": 646,
+      "[ReH]": 647,
+      "[67Ga+3]": 648,
+      "[Db]": 649,
+      "[177Lu+3]": 650,
+      "[AlH2-]": 651,
+      "[Si+3]": 652,
+      "[Ti-2]": 653,
+      "[RuH+3]": 654,
+      "[al+]": 655,
+      "[68Ga]": 656,
+      "[2H+]": 657,
+      "[B@H-]": 658,
+      "[WH2]": 659,
+      "[OsH]": 660,
+      "[Ir-3]": 661,
+      "[AlH-]": 662,
+      "[Bk]": 663,
+      "[75Se]": 664,
+      "[14C@]": 665,
+      "[Pt-]": 666,
+      "[N@@H+]": 667,
+      "[Nb-]": 668,
+      "[13NH2]": 669,
+      "%93": 670,
+      "[186Re]": 671,
+      "[Tb+4]": 672,
+      "[PtH]": 673,
+      "[IrH2]": 674,
+      "[Hg-2]": 675,
+      "[AlH3-]": 676,
+      "[PdH+]": 677,
+      "[Md]": 678,
+      "[RhH+2]": 679,
+      "[11cH]": 680,
+      "[Co-2]": 681,
+      "[15N-]": 682,
+      "[ZrH2]": 683,
+      "%94": 684,
+      "[Hg-]": 685,
+      "[127I]": 686,
+      "[AsH2+]": 687,
+      "[MoH2]": 688,
+      "[Te+4]": 689,
+      "[14C@@]": 690,
+      "[As+5]": 691,
+      "[SnH+3]": 692,
+      "[Ge@@]": 693,
+      "[6Li+]": 694,
+      "[WH]": 695,
+      "[Ne]": 696,
+      "[14NH2]": 697,
+      "[14NH]": 698,
+      "[12C@@H]": 699,
+      "[Os+7]": 700,
+      "[RhH]": 701,
+      "[Al-3]": 702,
+      "[SnH+]": 703,
+      "[15NH3+]": 704,
+      "[Zr+]": 705,
+      "[197Hg+]": 706,
+      "%95": 707,
+      "%96": 708,
+      "[90Y+3]": 709,
+      "[Os-2]": 710,
+      "[98Tc+5]": 711,
+      "[15NH3]": 712,
+      "[bH-]": 713,
+      "[33P]": 714,
+      "[Zr-2]": 715,
+      "[15O]": 716,
+      "[Rh-]": 717,
+      "[PbH3]": 718,
+      "[PH2]": 719,
+      "[Ni-]": 720,
+      "[CuH+]": 721,
+      "%97": 722,
+      "%98": 723,
+      "%99": 724,
+      "[Os+5]": 725,
+      "[PtH+]": 726,
+      "[ReH4]": 727,
+      "[16NH]": 728,
+      "[82Br]": 729,
+      "[W-]": 730,
+      "[18F-]": 731,
+      "[15NH4+]": 732,
+      "[Se+4]": 733,
+      "[SeH-]": 734,
+      "[67Cu+2]": 735,
+      "[12C@H]": 736,
+      "[AsH3]": 737,
+      "[HgH]": 738,
+      "[10B-]": 739,
+      "[99Tc+6]": 740,
+      "[117Sn+4]": 741,
+      "[Te@]": 742,
+      "[P@+]": 743,
+      "[35SH]": 744,
+      "[SeH+]": 745,
+      "[Ni-2]": 746,
+      "[Al-2]": 747,
+      "[TeH2]": 748,
+      "[Bh]": 749,
+      "[99Tc+2]": 750,
+      "[Os+8]": 751,
+      "[PH-2]": 752,
+      "[7Li+]": 753,
+      "[14nH]": 754,
+      "[AlH+2]": 755,
+      "[18FH]": 756,
+      "[SnH4]": 757,
+      "[18O-2]": 758,
+      "[IrH]": 759,
+      "[13N]": 760,
+      "[Te@@]": 761,
+      "[Rh-3]": 762,
+      "[15NH+]": 763,
+      "[AsH3+]": 764,
+      "[SeH2]": 765,
+      "[AsH+]": 766,
+      "[CoH2]": 767,
+      "[16NH2]": 768,
+      "[AsH-]": 769,
+      "[203Hg+]": 770,
+      "[P@@+]": 771,
+      "[166Ho+3]": 772,
+      "[60Co+3]": 773,
+      "[13CH2-]": 774,
+      "[SeH2+]": 775,
+      "[75Br]": 776,
+      "[TlH2]": 777,
+      "[80Br]": 778,
+      "[siH+]": 779,
+      "[Ca+]": 780,
+      "[153Sm+3]": 781,
+      "[PdH]": 782,
+      "[225Ac]": 783,
+      "[13CH3-]": 784,
+      "[AlH4-]": 785,
+      "[FeH]": 786,
+      "[13CH-]": 787,
+      "[14C-]": 788,
+      "[11C-]": 789,
+      "[153Sm]": 790,
+      "[Re-]": 791,
+      "[te+]": 792,
+      "[13CH4]": 793,
+      "[ClH+2]": 794,
+      "[8CH2]": 795,
+      "[99Mo]": 796,
+      "[ClH3+3]": 797,
+      "[SbH3]": 798,
+      "[25Mg+2]": 799,
+      "[16N+]": 800,
+      "[SnH2+]": 801,
+      "[11C@H]": 802,
+      "[122I]": 803,
+      "[Re-2]": 804,
+      "[RuH2+2]": 805,
+      "[ZrH]": 806,
+      "[Bi-]": 807,
+      "[Pr+]": 808,
+      "[Rn]": 809,
+      "[Fr]": 810,
+      "[36Cl]": 811,
+      "[18o]": 812,
+      "[YH]": 813,
+      "[79Br]": 814,
+      "[121I]": 815,
+      "[113In+3]": 816,
+      "[TaH]": 817,
+      "[RhH2]": 818,
+      "[Ta-]": 819,
+      "[67Ga]": 820,
+      "[ZnH+]": 821,
+      "[SnH2-]": 822,
+      "[OsH2]": 823,
+      "[16F]": 824,
+      "[FeH2]": 825,
+      "[14O]": 826,
+      "[PbH2+2]": 827,
+      "[BH2]": 828,
+      "[6H]": 829,
+      "[125Te]": 830,
+      "[197Hg]": 831,
+      "[TaH2]": 832,
+      "[TaH3]": 833,
+      "[76As]": 834,
+      "[Nb-2]": 835,
+      "[14N+]": 836,
+      "[125I-]": 837,
+      "[33S]": 838,
+      "[IH2+2]": 839,
+      "[NH2]": 840,
+      "[PtH2]": 841,
+      "[MnH]": 842,
+      "[19C]": 843,
+      "[17F]": 844,
+      "[1H-]": 845,
+      "[SnH4+2]": 846,
+      "[Mn-2]": 847,
+      "[15NH2+]": 848,
+      "[TiH2]": 849,
+      "[ReH7]": 850,
+      "[Cd-2]": 851,
+      "[Fe-3]": 852,
+      "[SH2]": 853,
+      "[17O-]": 854,
+      "[siH-]": 855,
+      "[CoH+]": 856,
+      "[VH]": 857,
+      "[10BH]": 858,
+      "[Ru-3]": 859,
+      "[13O]": 860,
+      "[5H]": 861,
+      "[15n-]": 862,
+      "[153Gd]": 863,
+      "[12C@]": 864,
+      "[11CH3-]": 865,
+      "[IrH3]": 866,
+      "[RuH3]": 867,
+      "[74Se]": 868,
+      "[Se@]": 869,
+      "[Hf+]": 870,
+      "[77Se]": 871,
+      "[166Ho]": 872,
+      "[59Fe+2]": 873,
+      "[203Hg]": 874,
+      "[18OH-]": 875,
+      "[8CH]": 876,
+      "[12C@@]": 877,
+      "[11CH4]": 878,
+      "[15C]": 879,
+      "[249Cf]": 880,
+      "[PbH4]": 881,
+      "[64Zn]": 882,
+      "[99Tc+]": 883,
+      "[14c-]": 884,
+      "[149Pm]": 885,
+      "[IrH4]": 886,
+      "[Se@@]": 887,
+      "[13OH]": 888,
+      "[14CH3-]": 889,
+      "[28Si]": 890,
+      "[Rh-2]": 891,
+      "[Fe-2]": 892,
+      "[131I-]": 893,
+      "[51Cr]": 894,
+      "[62Cu+2]": 895,
+      "[81Br]": 896,
+      "[121Sb]": 897,
+      "[7Li]": 898,
+      "[89Zr+4]": 899,
+      "[SbH3+]": 900,
+      "[11C@@H]": 901,
+      "[98Tc]": 902,
+      "[59Fe+3]": 903,
+      "[BiH2+]": 904,
+      "[SbH+]": 905,
+      "[TiH]": 906,
+      "[14NH3]": 907,
+      "[15OH]": 908,
+      "[119Sn]": 909,
+      "[201Hg]": 910,
+      "[MnH+]": 911,
+      "[201Tl]": 912,
+      "[51Cr+3]": 913,
+      "[123I-]": 914,
+      "[MoH]": 915,
+      "[AlH6-3]": 916,
+      "[MnH2]": 917,
+      "[WH3]": 918,
+      "[213Bi+3]": 919,
+      "[SnH2+2]": 920,
+      "[123IH]": 921,
+      "[13CH+]": 922,
+      "[Zr-]": 923,
+      "[74As]": 924,
+      "[13C+]": 925,
+      "[32P+]": 926,
+      "[KrH]": 927,
+      "[SiH+2]": 928,
+      "[ClH3+2]": 929,
+      "[13NH]": 930,
+      "[9CH2]": 931,
+      "[ZrH2+2]": 932,
+      "[87Sr+2]": 933,
+      "[35s]": 934,
+      "[239Pu]": 935,
+      "[198Au]": 936,
+      "[241Am]": 937,
+      "[203Hg+2]": 938,
+      "[V+]": 939,
+      "[YH2]": 940,
+      "[195Pt]": 941,
+      "[203Pb]": 942,
+      "[RuH4]": 943,
+      "[ThH2]": 944,
+      "[AuH]": 945,
+      "[66Ga+3]": 946,
+      "[11B-]": 947,
+      "[F]": 948,
+      "[24Na+]": 949,
+      "[85Sr+2]": 950,
+      "[201Tl+]": 951,
+      "[14CH4]": 952,
+      "[32S]": 953,
+      "[TeH2+]": 954,
+      "[ClH2+3]": 955,
+      "[AgH]": 956,
+      "[Ge@H]": 957,
+      "[44Ca+2]": 958,
+      "[Os-]": 959,
+      "[31P]": 960,
+      "[15nH+]": 961,
+      "[SbH4]": 962,
+      "[TiH+]": 963,
+      "[Ba+]": 964,
+      "[57Co+2]": 965,
+      "[Ta+]": 966,
+      "[125IH]": 967,
+      "[77As]": 968,
+      "[129I]": 969,
+      "[Fe-4]": 970,
+      "[Ta-2]": 971,
+      "[19O]": 972,
+      "[12O]": 973,
+      "[BiH3]": 974,
+      "[237Np]": 975,
+      "[252Cf]": 976,
+      "[86Y]": 977,
+      "[Cr-2]": 978,
+      "[89Y]": 979,
+      "[195Pt+2]": 980,
+      "[si+2]": 981,
+      "[58Fe+2]": 982,
+      "[Hs]": 983,
+      "[S@@H]": 984,
+      "[8CH4]": 985,
+      "[164Dy+3]": 986,
+      "[47Ca+2]": 987,
+      "[57Co]": 988,
+      "[NbH2]": 989,
+      "[ReH2]": 990,
+      "[ZnH2]": 991,
+      "[CrH2]": 992,
+      "[17NH]": 993,
+      "[ZrH3]": 994,
+      "[RhH3]": 995,
+      "[12C-]": 996,
+      "[18O+]": 997,
+      "[Bi-2]": 998,
+      "[ClH4+3]": 999,
+      "[Ni-3]": 1000,
+      "[Ag-]": 1001,
+      "[111In-]": 1002,
+      "[Mo-2]": 1003,
+      "[55Fe+3]": 1004,
+      "[204Hg+]": 1005,
+      "[35Cl-]": 1006,
+      "[211Pb]": 1007,
+      "[75Ge]": 1008,
+      "[8B]": 1009,
+      "[TeH3]": 1010,
+      "[SnH3+]": 1011,
+      "[Zr-3]": 1012,
+      "[28F]": 1013,
+      "[249Bk]": 1014,
+      "[169Yb]": 1015,
+      "[34SH]": 1016,
+      "[6Li]": 1017,
+      "[94Tc]": 1018,
+      "[197Au]": 1019,
+      "[195Pt+4]": 1020,
+      "[169Yb+3]": 1021,
+      "[32Cl]": 1022,
+      "[82Se]": 1023,
+      "[159Gd+3]": 1024,
+      "[213Bi]": 1025,
+      "[CoH+2]": 1026,
+      "[36S]": 1027,
+      "[35P]": 1028,
+      "[Ru-4]": 1029,
+      "[Cr-3]": 1030,
+      "[60Co]": 1031,
+      "[1H+]": 1032,
+      "[18CH2]": 1033,
+      "[Cd-]": 1034,
+      "[152Sm+3]": 1035,
+      "[106Ru]": 1036,
+      "[238Pu]": 1037,
+      "[220Rn]": 1038,
+      "[45Ca+2]": 1039,
+      "[89Sr+2]": 1040,
+      "[239Np]": 1041,
+      "[90Sr+2]": 1042,
+      "[137Cs+]": 1043,
+      "[165Dy]": 1044,
+      "[68GaH3]": 1045,
+      "[65Zn+2]": 1046,
+      "[89Zr]": 1047,
+      "[BiH2+2]": 1048,
+      "[62Cu]": 1049,
+      "[165Dy+3]": 1050,
+      "[238U]": 1051,
+      "[105Rh+3]": 1052,
+      "[70Zn]": 1053,
+      "[12B]": 1054,
+      "[12OH]": 1055,
+      "[18CH]": 1056,
+      "[17CH]": 1057,
+      "[42K]": 1058,
+      "[76Br-]": 1059,
+      "[71As]": 1060,
+      "[NbH3]": 1061,
+      "[ReH3]": 1062,
+      "[OsH-]": 1063,
+      "[WH4]": 1064,
+      "[MoH3]": 1065,
+      "[OsH4]": 1066,
+      "[RuH6]": 1067,
+      "[PtH3]": 1068,
+      "[CuH2]": 1069,
+      "[CoH3]": 1070,
+      "[TiH4]": 1071,
+      "[64Zn+2]": 1072,
+      "[Si-2]": 1073,
+      "[79BrH]": 1074,
+      "[14CH2-]": 1075,
+      "[PtH2+2]": 1076,
+      "[Os-3]": 1077,
+      "[29Si]": 1078,
+      "[Ti-]": 1079,
+      "[Se+6]": 1080,
+      "[22Na+]": 1081,
+      "[42K+]": 1082,
+      "[131Cs+]": 1083,
+      "[86Rb+]": 1084,
+      "[134Cs+]": 1085,
+      "[209Po]": 1086,
+      "[208Po]": 1087,
+      "[81Rb+]": 1088,
+      "[203Tl+]": 1089,
+      "[Zr-4]": 1090,
+      "[148Sm]": 1091,
+      "[147Sm]": 1092,
+      "[37Cl-]": 1093,
+      "[12CH4]": 1094,
+      "[Ge@@H]": 1095,
+      "[63Cu]": 1096,
+      "[13CH2+]": 1097,
+      "[AsH2-]": 1098,
+      "[CeH]": 1099,
+      "[SnH-]": 1100,
+      "[UH]": 1101,
+      "[9c]": 1102,
+      "[21CH3]": 1103,
+      "[TeH+]": 1104,
+      "[57Co+3]": 1105,
+      "[8BH2]": 1106,
+      "[12BH2]": 1107,
+      "[19BH2]": 1108,
+      "[9BH2]": 1109,
+      "[YbH2]": 1110,
+      "[CrH+2]": 1111,
+      "[208Bi]": 1112,
+      "[152Gd]": 1113,
+      "[61Cu]": 1114,
+      "[115In]": 1115,
+      "[60Co+2]": 1116,
+      "[13NH2-]": 1117,
+      "[120I]": 1118,
+      "[18OH2]": 1119,
+      "[75SeH]": 1120,
+      "[SbH2+]": 1121,
+      "[144Ce]": 1122,
+      "[16n]": 1123,
+      "[113In]": 1124,
+      "[22nH]": 1125,
+      "[129I-]": 1126,
+      "[InH3]": 1127,
+      "[32PH3]": 1128,
+      "[234U]": 1129,
+      "[235U]": 1130,
+      "[59Fe]": 1131,
+      "[82Rb+]": 1132,
+      "[65Zn]": 1133,
+      "[244Cm]": 1134,
+      "[147Pm]": 1135,
+      "[91Y]": 1136,
+      "[237Pu]": 1137,
+      "[231Pa]": 1138,
+      "[253Cf]": 1139,
+      "[127Te]": 1140,
+      "[187Re]": 1141,
+      "[236Np]": 1142,
+      "[235Np]": 1143,
+      "[72Zn]": 1144,
+      "[253Es]": 1145,
+      "[159Dy]": 1146,
+      "[62Zn]": 1147,
+      "[101Tc]": 1148,
+      "[149Tb]": 1149,
+      "[124I-]": 1150,
+      "[SeH3+]": 1151,
+      "[210Pb]": 1152,
+      "[40K]": 1153,
+      "[210Po]": 1154,
+      "[214Pb]": 1155,
+      "[218Po]": 1156,
+      "[214Po]": 1157,
+      "[7Be]": 1158,
+      "[212Pb]": 1159,
+      "[205Pb]": 1160,
+      "[209Pb]": 1161,
+      "[123Te]": 1162,
+      "[202Pb]": 1163,
+      "[72As]": 1164,
+      "[201Pb]": 1165,
+      "[70As]": 1166,
+      "[73Ge]": 1167,
+      "[200Pb]": 1168,
+      "[198Pb]": 1169,
+      "[66Ga]": 1170,
+      "[73Se]": 1171,
+      "[195Pb]": 1172,
+      "[199Pb]": 1173,
+      "[144Ce+3]": 1174,
+      "[235U+2]": 1175,
+      "[90Tc]": 1176,
+      "[114In+3]": 1177,
+      "[128I]": 1178,
+      "[100Tc+]": 1179,
+      "[82Br-]": 1180,
+      "[191Pt+2]": 1181,
+      "[191Pt+4]": 1182,
+      "[193Pt+4]": 1183,
+      "[31PH3]": 1184,
+      "[125I+2]": 1185,
+      "[131I+2]": 1186,
+      "[125Te+4]": 1187,
+      "[82Sr+2]": 1188,
+      "[149Sm]": 1189,
+      "[81BrH]": 1190,
+      "[129Xe]": 1191,
+      "[193Pt+2]": 1192,
+      "[123I+2]": 1193,
+      "[Cr-]": 1194,
+      "[Co-]": 1195,
+      "[227Th+4]": 1196,
+      "[249Cf+3]": 1197,
+      "[252Cf+3]": 1198,
+      "[187Os]": 1199,
+      "[16O-]": 1200,
+      "[17O+]": 1201,
+      "[16OH-]": 1202,
+      "[98Tc+7]": 1203,
+      "[58Co+2]": 1204,
+      "[69Ga+3]": 1205,
+      "[57Fe+2]": 1206,
+      "[43K+]": 1207,
+      "[16C]": 1208,
+      "[52Fe+3]": 1209,
+      "[SeH5]": 1210,
+      "[194Pb]": 1211,
+      "[196Pb]": 1212,
+      "[197Pb]": 1213,
+      "[213Pb]": 1214,
+      "[9B]": 1215,
+      "[19B]": 1216,
+      "[11CH-]": 1217,
+      "[9CH]": 1218,
+      "[20OH]": 1219,
+      "[25OH]": 1220,
+      "[8cH]": 1221,
+      "[TiH+3]": 1222,
+      "[SnH6+3]": 1223,
+      "[N@H+]": 1224,
+      "[52Mn+2]": 1225,
+      "[64Ga]": 1226,
+      "[13B]": 1227,
+      "[216Bi]": 1228,
+      "[117Sn+2]": 1229,
+      "[232Th]": 1230,
+      "[SnH+2]": 1231,
+      "[BiH5]": 1232,
+      "[77Kr]": 1233,
+      "[103Cd]": 1234,
+      "[62Ni]": 1235,
+      "[LaH3]": 1236,
+      "[SmH3]": 1237,
+      "[EuH3]": 1238,
+      "[MoH5]": 1239,
+      "[64Ni]": 1240,
+      "[66Zn]": 1241,
+      "[68Zn]": 1242,
+      "[186W]": 1243,
+      "[FeH4]": 1244,
+      "[MoH4]": 1245,
+      "[HgH2]": 1246,
+      "[15NH2-]": 1247,
+      "[UH2]": 1248,
+      "[204Hg]": 1249,
+      "[GaH4-]": 1250,
+      "[ThH4]": 1251,
+      "[WH6]": 1252,
+      "[PtH4]": 1253,
+      "[VH2]": 1254,
+      "[UH3]": 1255,
+      "[FeH3]": 1256,
+      "[RuH5]": 1257,
+      "[BiH4]": 1258,
+      "[80Br-]": 1259,
+      "[CeH3]": 1260,
+      "[37ClH]": 1261,
+      "[157Gd+3]": 1262,
+      "[205Tl]": 1263,
+      "[203Tl]": 1264,
+      "[62Cu+]": 1265,
+      "[64Cu+]": 1266,
+      "[61Cu+]": 1267,
+      "[37SH2]": 1268,
+      "[30Si]": 1269,
+      "[28Al]": 1270,
+      "[19OH2]": 1271,
+      "[8He]": 1272,
+      "[6He]": 1273,
+      "[153Pm]": 1274,
+      "[209Bi]": 1275,
+      "[66Zn+2]": 1276,
+      "[10CH4]": 1277,
+      "[191Ir]": 1278,
+      "[66Cu]": 1279,
+      "[16O+]": 1280,
+      "[25O]": 1281,
+      "[10c]": 1282,
+      "[Co-3]": 1283,
+      "[Sn@@]": 1284,
+      "[17OH-]": 1285,
+      "[206Po]": 1286,
+      "[204Po]": 1287,
+      "[202Po]": 1288,
+      "[201Po]": 1289,
+      "[200Po]": 1290,
+      "[199Po]": 1291,
+      "[198Po]": 1292,
+      "[197Po]": 1293,
+      "[196Po]": 1294,
+      "[195Po]": 1295,
+      "[194Po]": 1296,
+      "[193Po]": 1297,
+      "[192Po]": 1298,
+      "[191Po]": 1299,
+      "[190Po]": 1300,
+      "[217Po]": 1301,
+      "[BiH4-]": 1302,
+      "[TeH4]": 1303,
+      "[222Ra]": 1304,
+      "[62Ga]": 1305,
+      "[39Ar]": 1306,
+      "[144Sm]": 1307,
+      "[58Fe]": 1308,
+      "[153Eu]": 1309,
+      "[85Rb]": 1310,
+      "[171Yb]": 1311,
+      "[172Yb]": 1312,
+      "[114Cd]": 1313,
+      "[51Fe]": 1314,
+      "[142Ce]": 1315,
+      "[207Tl]": 1316,
+      "[92Mo]": 1317,
+      "[115Sn]": 1318,
+      "[140Ce]": 1319,
+      "[202Hg]": 1320,
+      "[180W]": 1321,
+      "[182W]": 1322,
+      "[183W]": 1323,
+      "[184W]": 1324,
+      "[96Mo]": 1325,
+      "[47Ti]": 1326,
+      "[111Cd]": 1327,
+      "[143Nd]": 1328,
+      "[145Nd]": 1329,
+      "[126Te]": 1330,
+      "[128Te]": 1331,
+      "[130Te]": 1332,
+      "[185Re]": 1333,
+      "[97Mo]": 1334,
+      "[98Mo]": 1335,
+      "[183Re]": 1336,
+      "[52V]": 1337,
+      "[80Se]": 1338,
+      "[87Kr]": 1339,
+      "[137Xe]": 1340,
+      "[196Au]": 1341,
+      "[146Ce]": 1342,
+      "[88Kr]": 1343,
+      "[51Ti]": 1344,
+      "[138Xe]": 1345,
+      "[112Cd]": 1346,
+      "[116Sn]": 1347,
+      "[120Sn]": 1348,
+      "[28SiH3]": 1349,
+      "[35S-]": 1350,
+      "[15NH-]": 1351,
+      "[13CH3+]": 1352,
+      "[34S+]": 1353,
+      "[34s]": 1354,
+      "[SiH4-]": 1355,
+      "[100Tc+5]": 1356,
+      "[NiH2+2]": 1357,
+      "[239Th]": 1358,
+      "[186Lu]": 1359,
+      "[AuH3]": 1360,
+      "[I@@-]": 1361,
+      "[XeH2]": 1362,
+      "[B+]": 1363,
+      "[16CH2]": 1364,
+      "[8C]": 1365,
+      "[TaH5]": 1366,
+      "[FeH4-]": 1367,
+      "[19C@H]": 1368,
+      "[10NH]": 1369,
+      "[FeH6-3]": 1370,
+      "[22CH]": 1371,
+      "[25N]": 1372,
+      "[25N+]": 1373,
+      "[25N-]": 1374,
+      "[21CH2]": 1375,
+      "[18cH]": 1376,
+      "[113I]": 1377,
+      "[ScH3]": 1378,
+      "[30PH3]": 1379,
+      "[43Ca+2]": 1380,
+      "[41Ca+2]": 1381,
+      "[106Cd]": 1382,
+      "[122Sn]": 1383,
+      "[18CH3]": 1384,
+      "[58Co+3]": 1385,
+      "[98Tc+4]": 1386,
+      "[70Ge]": 1387,
+      "[76Ge]": 1388,
+      "[108Cd]": 1389,
+      "[116Cd]": 1390,
+      "[130Xe]": 1391,
+      "[94Mo]": 1392,
+      "[124Sn]": 1393,
+      "[186Os]": 1394,
+      "[188Os]": 1395,
+      "[190Os]": 1396,
+      "[192Os]": 1397,
+      "[106Pd]": 1398,
+      "[110Pd]": 1399,
+      "[120Te]": 1400,
+      "[132Ba]": 1401,
+      "[134Ba]": 1402,
+      "[136Ba]": 1403,
+      "[136Ce]": 1404,
+      "[138Ce]": 1405,
+      "[156Dy]": 1406,
+      "[158Dy]": 1407,
+      "[160Dy]": 1408,
+      "[163Dy]": 1409,
+      "[162Er]": 1410,
+      "[164Er]": 1411,
+      "[167Er]": 1412,
+      "[176Hf]": 1413,
+      "[26Mg]": 1414,
+      "[144Nd]": 1415,
+      "[150Nd]": 1416,
+      "[41K]": 1417,
+      "[46Ti]": 1418,
+      "[48Ti]": 1419,
+      "[49Ti]": 1420,
+      "[50Ti]": 1421,
+      "[170Yb]": 1422,
+      "[173Yb]": 1423,
+      "[91Zr]": 1424,
+      "[92Zr]": 1425,
+      "[96Zr]": 1426,
+      "[34S-]": 1427,
+      "[CuH2-]": 1428,
+      "[38Cl]": 1429,
+      "[25Mg]": 1430,
+      "[51V]": 1431,
+      "[93Nb]": 1432,
+      "[95Mo]": 1433,
+      "[45Sc]": 1434,
+      "[123Sb]": 1435,
+      "[139La]": 1436,
+      "[9Be]": 1437,
+      "[99Y+3]": 1438,
+      "[99Y]": 1439,
+      "[156Ho]": 1440,
+      "[67Zn]": 1441,
+      "[144Ce+4]": 1442,
+      "[210Tl]": 1443,
+      "[42Ca]": 1444,
+      "[54Fe]": 1445,
+      "[193Ir]": 1446,
+      "[92Nb]": 1447,
+      "[141Cs]": 1448,
+      "[52Cr]": 1449,
+      "[35ClH]": 1450,
+      "[46Ca]": 1451,
+      "[139Cs]": 1452,
+      "[65Cu]": 1453,
+      "[71Ga]": 1454,
+      "[60Ni]": 1455,
+      "[16NH3]": 1456,
+      "[148Nd]": 1457,
+      "[72Ge]": 1458,
+      "[161Dy]": 1459,
+      "[49Ca]": 1460,
+      "[43Ca]": 1461,
+      "[8Be]": 1462,
+      "[48Ca]": 1463,
+      "[44Ca]": 1464,
+      "[120Xe]": 1465,
+      "[80Rb]": 1466,
+      "[215At]": 1467,
+      "[180Re]": 1468,
+      "[146Sm]": 1469,
+      "[19Ne]": 1470,
+      "[74Kr]": 1471,
+      "[134La]": 1472,
+      "[76Kr]": 1473,
+      "[219Fr]": 1474,
+      "[121Xe]": 1475,
+      "[220Fr]": 1476,
+      "[216At]": 1477,
+      "[223Ac]": 1478,
+      "[218At]": 1479,
+      "[37Ar]": 1480,
+      "[135I]": 1481,
+      "[110Cd]": 1482,
+      "[94Tc+7]": 1483,
+      "[86Y+3]": 1484,
+      "[135I-]": 1485,
+      "[15O-2]": 1486,
+      "[151Eu+3]": 1487,
+      "[161Tb+3]": 1488,
+      "[197Hg+2]": 1489,
+      "[109Cd+2]": 1490,
+      "[191Os+4]": 1491,
+      "[170Tm+3]": 1492,
+      "[205Bi+3]": 1493,
+      "[233U+4]": 1494,
+      "[126Sb+3]": 1495,
+      "[127Sb+3]": 1496,
+      "[132Cs+]": 1497,
+      "[136Eu+3]": 1498,
+      "[136Eu]": 1499,
+      "[125Sn+4]": 1500,
+      "[175Yb+3]": 1501,
+      "[100Mo]": 1502,
+      "[22Ne]": 1503,
+      "[13c-]": 1504,
+      "[13NH4+]": 1505,
+      "[17C]": 1506,
+      "[9C]": 1507,
+      "[31S]": 1508,
+      "[31SH]": 1509,
+      "[133I]": 1510,
+      "[126I]": 1511,
+      "[36SH]": 1512,
+      "[30S]": 1513,
+      "[32SH]": 1514,
+      "[19CH2]": 1515,
+      "[19c]": 1516,
+      "[18c]": 1517,
+      "[15F]": 1518,
+      "[10C]": 1519,
+      "[RuH-]": 1520,
+      "[62Zn+2]": 1521,
+      "[32ClH]": 1522,
+      "[33ClH]": 1523,
+      "[78BrH]": 1524,
+      "[12Li+]": 1525,
+      "[12Li]": 1526,
+      "[233Ra]": 1527,
+      "[68Ge+4]": 1528,
+      "[44Sc+3]": 1529,
+      "[91Y+3]": 1530,
+      "[106Ru+3]": 1531,
+      "[PoH2]": 1532,
+      "[AtH]": 1533,
+      "[55Fe]": 1534,
+      "[233U]": 1535,
+      "[210PoH2]": 1536,
+      "[230Th]": 1537,
+      "[228Th]": 1538,
+      "[222Rn]": 1539,
+      "[35SH2]": 1540,
+      "[227Th]": 1541,
+      "[192Ir]": 1542,
+      "[133Xe]": 1543,
+      "[81Kr]": 1544,
+      "[95Zr]": 1545,
+      "[240Pu]": 1546,
+      "[54Mn]": 1547,
+      "[103Ru]": 1548,
+      "[95Nb]": 1549,
+      "[109Cd]": 1550,
+      "[141Ce]": 1551,
+      "[85Kr]": 1552,
+      "[110Ag]": 1553,
+      "[58Co]": 1554,
+      "[241Pu]": 1555,
+      "[234Th]": 1556,
+      "[140La]": 1557,
+      "[63Ni]": 1558,
+      "[152Eu]": 1559,
+      "[132IH]": 1560,
+      "[226Rn]": 1561,
+      "[154Eu]": 1562,
+      "[36ClH]": 1563,
+      "[228Ac]": 1564,
+      "[155Eu]": 1565,
+      "[106Rh]": 1566,
+      "[243Am]": 1567,
+      "[227Ac]": 1568,
+      "[243Cm]": 1569,
+      "[236U]": 1570,
+      "[144Pr]": 1571,
+      "[232U]": 1572,
+      "[32SH2]": 1573,
+      "[88Y]": 1574,
+      "[82BrH]": 1575,
+      "[135IH]": 1576,
+      "[242Cm]": 1577,
+      "[115Cd]": 1578,
+      "[242Pu]": 1579,
+      "[46Sc]": 1580,
+      "[56Mn]": 1581,
+      "[234Pa]": 1582,
+      "[41Ar]": 1583,
+      "[147Nd]": 1584,
+      "[187W]": 1585,
+      "[151Sm]": 1586,
+      "[59Ni]": 1587,
+      "[233Pa]": 1588,
+      "[52Mn]": 1589,
+      "[94Nb]": 1590,
+      "[219Rn]": 1591,
+      "[236Pu]": 1592,
+      "[13NH3]": 1593,
+      "[93Zr]": 1594,
+      "[51Cr+6]": 1595,
+      "[TlH3]": 1596,
+      "[123Xe]": 1597,
+      "[160Tb]": 1598,
+      "[170Tm]": 1599,
+      "[182Ta]": 1600,
+      "[175Yb]": 1601,
+      "[93Mo]": 1602,
+      "[143Ce]": 1603,
+      "[191Os]": 1604,
+      "[126IH]": 1605,
+      "[48V]": 1606,
+      "[113Cd]": 1607,
+      "[47Sc]": 1608,
+      "[181Hf]": 1609,
+      "[185W]": 1610,
+      "[143Pr]": 1611,
+      "[191Pt]": 1612,
+      "[181W]": 1613,
+      "[33PH3]": 1614,
+      "[97Ru]": 1615,
+      "[97Tc]": 1616,
+      "[111Ag]": 1617,
+      "[169Er]": 1618,
+      "[107Pd]": 1619,
+      "[103Ru+2]": 1620,
+      "[34SH2]": 1621,
+      "[137Ce]": 1622,
+      "[242Am]": 1623,
+      "[117SnH2]": 1624,
+      "[57Ni]": 1625,
+      "[239U]": 1626,
+      "[60Cu]": 1627,
+      "[250Cf]": 1628,
+      "[193Au]": 1629,
+      "[69Zn]": 1630,
+      "[55Co]": 1631,
+      "[139Ce]": 1632,
+      "[127Xe]": 1633,
+      "[159Gd]": 1634,
+      "[56Co]": 1635,
+      "[177Hf]": 1636,
+      "[244Pu]": 1637,
+      "[38ClH]": 1638,
+      "[142Pr]": 1639,
+      "[199Hg]": 1640,
+      "[179Hf]": 1641,
+      "[178Hf]": 1642,
+      "[237U]": 1643,
+      "[156Eu]": 1644,
+      "[157Eu]": 1645,
+      "[105Ru]": 1646,
+      "[171Tm]": 1647,
+      "[199Au]": 1648,
+      "[155Sm]": 1649,
+      "[80BrH]": 1650,
+      "[108Ag]": 1651,
+      "[128IH]": 1652,
+      "[48Sc]": 1653,
+      "[45Ti]": 1654,
+      "[176Lu]": 1655,
+      "[121SnH2]": 1656,
+      "[148Pm]": 1657,
+      "[57Fe]": 1658,
+      "[10BH3]": 1659,
+      "[96Tc]": 1660,
+      "[133IH]": 1661,
+      "[143Pm]": 1662,
+      "[105Rh]": 1663,
+      "[130IH]": 1664,
+      "[134IH]": 1665,
+      "[131IH]": 1666,
+      "[71Zn]": 1667,
+      "[105Ag]": 1668,
+      "[97Zr]": 1669,
+      "[235Pu]": 1670,
+      "[231Th]": 1671,
+      "[109Pd]": 1672,
+      "[93Y]": 1673,
+      "[190Ir]": 1674,
+      "[135Xe]": 1675,
+      "[53Mn]": 1676,
+      "[134Ce]": 1677,
+      "[234Np]": 1678,
+      "[240Am]": 1679,
+      "[246Cf]": 1680,
+      "[240Cm]": 1681,
+      "[241Cm]": 1682,
+      "[226Th]": 1683,
+      "[39ClH]": 1684,
+      "[229Th]": 1685,
+      "[245Cm]": 1686,
+      "[240U]": 1687,
+      "[240Np]": 1688,
+      "[249Cm]": 1689,
+      "[243Pu]": 1690,
+      "[145Pm]": 1691,
+      "[199Pt]": 1692,
+      "[246Bk]": 1693,
+      "[193Pt]": 1694,
+      "[230U]": 1695,
+      "[250Cm]": 1696,
+      "[44Ti]": 1697,
+      "[175Hf]": 1698,
+      "[254Fm]": 1699,
+      "[255Fm]": 1700,
+      "[257Fm]": 1701,
+      "[92Y]": 1702,
+      "[188Ir]": 1703,
+      "[171Lu]": 1704,
+      "[257Md]": 1705,
+      "[247Bk]": 1706,
+      "[121IH]": 1707,
+      "[250Bk]": 1708,
+      "[179Lu]": 1709,
+      "[224Ac]": 1710,
+      "[195Hg]": 1711,
+      "[244Am]": 1712,
+      "[246Pu]": 1713,
+      "[194Au]": 1714,
+      "[252Fm]": 1715,
+      "[173Hf]": 1716,
+      "[246Cm]": 1717,
+      "[135Ce]": 1718,
+      "[49Cr]": 1719,
+      "[248Cf]": 1720,
+      "[247Cm]": 1721,
+      "[248Cm]": 1722,
+      "[174Ta]": 1723,
+      "[176Ta]": 1724,
+      "[154Tb]": 1725,
+      "[172Ta]": 1726,
+      "[177Ta]": 1727,
+      "[175Ta]": 1728,
+      "[180Ta]": 1729,
+      "[158Tb]": 1730,
+      "[115Ag]": 1731,
+      "[189Os]": 1732,
+      "[251Cf]": 1733,
+      "[145Pr]": 1734,
+      "[147Pr]": 1735,
+      "[76BrH]": 1736,
+      "[102Rh]": 1737,
+      "[238Np]": 1738,
+      "[185Os]": 1739,
+      "[246Am]": 1740,
+      "[233Np]": 1741,
+      "[166Dy]": 1742,
+      "[254Es]": 1743,
+      "[244Cf]": 1744,
+      "[193Os]": 1745,
+      "[245Am]": 1746,
+      "[245Bk]": 1747,
+      "[239Am]": 1748,
+      "[238Am]": 1749,
+      "[97Nb]": 1750,
+      "[245Pu]": 1751,
+      "[254Cf]": 1752,
+      "[188W]": 1753,
+      "[250Es]": 1754,
+      "[251Es]": 1755,
+      "[237Am]": 1756,
+      "[182Hf]": 1757,
+      "[258Md]": 1758,
+      "[232Np]": 1759,
+      "[238Cm]": 1760,
+      "[60Fe]": 1761,
+      "[109Pd+2]": 1762,
+      "[234Pu]": 1763,
+      "[141Ce+3]": 1764,
+      "[136Nd]": 1765,
+      "[136Pr]": 1766,
+      "[173Ta]": 1767,
+      "[110Ru]": 1768,
+      "[147Tb]": 1769,
+      "[253Fm]": 1770,
+      "[139Nd]": 1771,
+      "[178Re]": 1772,
+      "[177Re]": 1773,
+      "[200Au]": 1774,
+      "[182Re]": 1775,
+      "[156Tb]": 1776,
+      "[155Tb]": 1777,
+      "[157Tb]": 1778,
+      "[161Tb]": 1779,
+      "[161Ho]": 1780,
+      "[167Tm]": 1781,
+      "[173Lu]": 1782,
+      "[179Ta]": 1783,
+      "[171Er]": 1784,
+      "[44Sc]": 1785,
+      "[49Sc]": 1786,
+      "[49V]": 1787,
+      "[51Mn]": 1788,
+      "[90Nb]": 1789,
+      "[88Nb]": 1790,
+      "[88Zr]": 1791,
+      "[36SH2]": 1792,
+      "[174Yb]": 1793,
+      "[178Lu]": 1794,
+      "[179W]": 1795,
+      "[83BrH]": 1796,
+      "[107Cd]": 1797,
+      "[75BrH]": 1798,
+      "[62Co]": 1799,
+      "[48Cr]": 1800,
+      "[63Zn]": 1801,
+      "[102Ag]": 1802,
+      "[154Sm]": 1803,
+      "[168Er]": 1804,
+      "[65Ni]": 1805,
+      "[137La]": 1806,
+      "[187Ir]": 1807,
+      "[144Pm]": 1808,
+      "[146Pm]": 1809,
+      "[160Gd]": 1810,
+      "[166Yb]": 1811,
+      "[162Dy]": 1812,
+      "[47V]": 1813,
+      "[141Nd]": 1814,
+      "[141Sm]": 1815,
+      "[166Er]": 1816,
+      "[150Sm]": 1817,
+      "[146Eu]": 1818,
+      "[149Eu]": 1819,
+      "[174Lu]": 1820,
+      "[17NH3]": 1821,
+      "[102Ru]": 1822,
+      "[170Hf]": 1823,
+      "[188Pt]": 1824,
+      "[61Ni]": 1825,
+      "[56Ni]": 1826,
+      "[149Gd]": 1827,
+      "[151Gd]": 1828,
+      "[141Pm]": 1829,
+      "[147Gd]": 1830,
+      "[146Gd]": 1831,
+      "[161Er]": 1832,
+      "[103Ag]": 1833,
+      "[145Eu]": 1834,
+      "[153Tb]": 1835,
+      "[155Dy]": 1836,
+      "[184Re]": 1837,
+      "[180Os]": 1838,
+      "[182Os]": 1839,
+      "[186Pt]": 1840,
+      "[181Os]": 1841,
+      "[181Re]": 1842,
+      "[151Tb]": 1843,
+      "[178Ta]": 1844,
+      "[178W]": 1845,
+      "[189Pt]": 1846,
+      "[194Hg]": 1847,
+      "[145Sm]": 1848,
+      "[150Tb]": 1849,
+      "[132La]": 1850,
+      "[158Gd]": 1851,
+      "[104Ag]": 1852,
+      "[193Hg]": 1853,
+      "[94Ru]": 1854,
+      "[137Pr]": 1855,
+      "[155Ho]": 1856,
+      "[117Cd]": 1857,
+      "[99Ru]": 1858,
+      "[146Nd]": 1859,
+      "[218Rn]": 1860,
+      "[95Y]": 1861,
+      "[79Kr]": 1862,
+      "[120IH]": 1863,
+      "[138Pr]": 1864,
+      "[100Pd]": 1865,
+      "[166Tm]": 1866,
+      "[90Mo]": 1867,
+      "[151Nd]": 1868,
+      "[231U]": 1869,
+      "[138Nd]": 1870,
+      "[89Nb]": 1871,
+      "[98Nb]": 1872,
+      "[162Ho]": 1873,
+      "[142Sm]": 1874,
+      "[186Ta]": 1875,
+      "[104Tc]": 1876,
+      "[184Ta]": 1877,
+      "[185Ta]": 1878,
+      "[170Er]": 1879,
+      "[107Rh]": 1880,
+      "[131La]": 1881,
+      "[169Lu]": 1882,
+      "[74BrH]": 1883,
+      "[150Pm]": 1884,
+      "[172Tm]": 1885,
+      "[197Pt]": 1886,
+      "[230Pu]": 1887,
+      "[170Lu]": 1888,
+      "[86Zr]": 1889,
+      "[176W]": 1890,
+      "[177W]": 1891,
+      "[101Pd]": 1892,
+      "[105Pd]": 1893,
+      "[108Pd]": 1894,
+      "[149Nd]": 1895,
+      "[164Ho]": 1896,
+      "[159Ho]": 1897,
+      "[167Ho]": 1898,
+      "[176Yb]": 1899,
+      "[156Sm]": 1900,
+      "[77BrH]": 1901,
+      "[189Re]": 1902,
+      "[99Rh]": 1903,
+      "[100Rh]": 1904,
+      "[151Pm]": 1905,
+      "[232Pa]": 1906,
+      "[228Pa]": 1907,
+      "[230Pa]": 1908,
+      "[66Ni]": 1909,
+      "[194Os]": 1910,
+      "[135La]": 1911,
+      "[138La]": 1912,
+      "[141La]": 1913,
+      "[142La]": 1914,
+      "[195Ir]": 1915,
+      "[96Nb]": 1916,
+      "[157Ho]": 1917,
+      "[183Hf]": 1918,
+      "[162Tm]": 1919,
+      "[172Er]": 1920,
+      "[148Eu]": 1921,
+      "[150Eu]": 1922,
+      "[15CH4]": 1923,
+      "[89Kr]": 1924,
+      "[143La]": 1925,
+      "[58Ni]": 1926,
+      "[61Co]": 1927,
+      "[158Eu]": 1928,
+      "[165Er]": 1929,
+      "[167Yb]": 1930,
+      "[173Tm]": 1931,
+      "[175Tm]": 1932,
+      "[172Hf]": 1933,
+      "[172Lu]": 1934,
+      "[93Tc]": 1935,
+      "[177Yb]": 1936,
+      "[124IH]": 1937,
+      "[194Ir]": 1938,
+      "[147Eu]": 1939,
+      "[101Mo]": 1940,
+      "[180Hf]": 1941,
+      "[189Ir]": 1942,
+      "[87Y]": 1943,
+      "[43Sc]": 1944,
+      "[195Au]": 1945,
+      "[112Ag]": 1946,
+      "[84BrH]": 1947,
+      "[106Ag]": 1948,
+      "[109Ag]": 1949,
+      "[101Rh]": 1950,
+      "[162Yb]": 1951,
+      "[228Rn]": 1952,
+      "[139Pr]": 1953,
+      "[94Y]": 1954,
+      "[201Au]": 1955,
+      "[40PH3]": 1956,
+      "[110Ag+]": 1957,
+      "[104Cd]": 1958,
+      "[133Ba+2]": 1959,
+      "[226Ac]": 1960,
+      "[145Gd]": 1961,
+      "[186Ir]": 1962,
+      "[184Ir]": 1963,
+      "[224Rn]": 1964,
+      "[185Ir]": 1965,
+      "[182Ir]": 1966,
+      "[184Hf]": 1967,
+      "[200Pt]": 1968,
+      "[227Pa]": 1969,
+      "[178Yb]": 1970,
+      "[72Br-]": 1971,
+      "[72BrH]": 1972,
+      "[248Am]": 1973,
+      "[238Th]": 1974,
+      "[161Gd]": 1975,
+      "[35S-2]": 1976,
+      "[107Ag]": 1977,
+      "[FeH6-4]": 1978,
+      "[89Sr]": 1979,
+      "[SnH3-]": 1980,
+      "[SeH3]": 1981,
+      "[TeH3+]": 1982,
+      "[SbH4+]": 1983,
+      "[AsH4+]": 1984,
+      "[4He]": 1985,
+      "[AsH3-]": 1986,
+      "[1HH]": 1987,
+      "[3H+]": 1988,
+      "[82Rb]": 1989,
+      "[85Sr]": 1990,
+      "[90Sr]": 1991,
+      "[137Cs]": 1992,
+      "[133Ba]": 1993,
+      "[131Cs]": 1994,
+      "[SbH5]": 1995,
+      "[224Ra]": 1996,
+      "[22Na]": 1997,
+      "[210Bi]": 1998,
+      "[214Bi]": 1999,
+      "[228Ra]": 2000,
+      "[127Sb]": 2001,
+      "[136Cs]": 2002,
+      "[125Sb]": 2003,
+      "[134Cs]": 2004,
+      "[140Ba]": 2005,
+      "[45Ca]": 2006,
+      "[206Pb]": 2007,
+      "[207Pb]": 2008,
+      "[24Na]": 2009,
+      "[86Rb]": 2010,
+      "[212Bi]": 2011,
+      "[208Pb]": 2012,
+      "[124Sb]": 2013,
+      "[204Pb]": 2014,
+      "[44K]": 2015,
+      "[129Te]": 2016,
+      "[113Sn]": 2017,
+      "[204Tl]": 2018,
+      "[87Sr]": 2019,
+      "[208Tl]": 2020,
+      "[87Rb]": 2021,
+      "[47Ca]": 2022,
+      "[135Cs]": 2023,
+      "[216Po]": 2024,
+      "[137Ba]": 2025,
+      "[207Bi]": 2026,
+      "[212Po]": 2027,
+      "[79Se]": 2028,
+      "[223Ra]": 2029,
+      "[86Sr]": 2030,
+      "[122Sb]": 2031,
+      "[26Al]": 2032,
+      "[32Si]": 2033,
+      "[126Sn]": 2034,
+      "[225Ra]": 2035,
+      "[114In]": 2036,
+      "[72Ga]": 2037,
+      "[132Te]": 2038,
+      "[10Be]": 2039,
+      "[125Sn]": 2040,
+      "[73As]": 2041,
+      "[206Bi]": 2042,
+      "[117Sn]": 2043,
+      "[40Ca]": 2044,
+      "[41Ca]": 2045,
+      "[89Rb]": 2046,
+      "[116In]": 2047,
+      "[129Sb]": 2048,
+      "[91Sr]": 2049,
+      "[71Ge]": 2050,
+      "[139Ba]": 2051,
+      "[69Ga]": 2052,
+      "[120Sb]": 2053,
+      "[121Sn]": 2054,
+      "[123Sn]": 2055,
+      "[131Te]": 2056,
+      "[77Ge]": 2057,
+      "[135Ba]": 2058,
+      "[82Sr]": 2059,
+      "[43K]": 2060,
+      "[131Ba]": 2061,
+      "[92Sr]": 2062,
+      "[88Rb]": 2063,
+      "[129Cs]": 2064,
+      "[144Cs]": 2065,
+      "[127Cs]": 2066,
+      "[200Tl]": 2067,
+      "[202Tl]": 2068,
+      "[141Ba]": 2069,
+      "[117Sb]": 2070,
+      "[116Sb]": 2071,
+      "[78As]": 2072,
+      "[131Sb]": 2073,
+      "[126Sb]": 2074,
+      "[128Sb]": 2075,
+      "[130Sb]": 2076,
+      "[67Ge]": 2077,
+      "[68Ge]": 2078,
+      "[78Ge]": 2079,
+      "[66Ge]": 2080,
+      "[223Fr]": 2081,
+      "[132Cs]": 2082,
+      "[125Cs]": 2083,
+      "[138Cs]": 2084,
+      "[133Te]": 2085,
+      "[84Rb]": 2086,
+      "[83Rb]": 2087,
+      "[81Rb]": 2088,
+      "[142Ba]": 2089,
+      "[200Bi]": 2090,
+      "[115Sb]": 2091,
+      "[194Tl]": 2092,
+      "[70Se]": 2093,
+      "[112In]": 2094,
+      "[118Sb]": 2095,
+      "[70Ga]": 2096,
+      "[27Mg]": 2097,
+      "[202Bi]": 2098,
+      "[83Se]": 2099,
+      "[9Li]": 2100,
+      "[69As]": 2101,
+      "[79Rb]": 2102,
+      "[81Sr]": 2103,
+      "[83Sr]": 2104,
+      "[78Se]": 2105,
+      "[109In]": 2106,
+      "[29Al]": 2107,
+      "[118Sn]": 2108,
+      "[117In]": 2109,
+      "[119Sb]": 2110,
+      "[114Sn]": 2111,
+      "[138Ba]": 2112,
+      "[69Ge]": 2113,
+      "[73Ga]": 2114,
+      "[74Ge]": 2115,
+      "[206Tl]": 2116,
+      "[199Tl]": 2117,
+      "[130Cs]": 2118,
+      "[28Mg]": 2119,
+      "[116Te]": 2120,
+      "[112Sn]": 2121,
+      "[126Ba]": 2122,
+      "[211Bi]": 2123,
+      "[81Se]": 2124,
+      "[127Sn]": 2125,
+      "[143Cs]": 2126,
+      "[134Te]": 2127,
+      "[80Sr]": 2128,
+      "[45K]": 2129,
+      "[215Po]": 2130,
+      "[207Po]": 2131,
+      "[111Sn]": 2132,
+      "[211Po]": 2133,
+      "[128Ba]": 2134,
+      "[198Tl]": 2135,
+      "[227Ra]": 2136,
+      "[213Po]": 2137,
+      "[220Ra]": 2138,
+      "[128Sn]": 2139,
+      "[203Po]": 2140,
+      "[205Po]": 2141,
+      "[65Ga]": 2142,
+      "[197Tl]": 2143,
+      "[88Sr]": 2144,
+      "[110In]": 2145,
+      "[31Si]": 2146,
+      "[201Bi]": 2147,
+      "[121Te]": 2148,
+      "[205Bi]": 2149,
+      "[203Bi]": 2150,
+      "[195Tl]": 2151,
+      "[209Tl]": 2152,
+      "[110Sn]": 2153,
+      "[222Fr]": 2154,
+      "[207At]": 2155,
+      "[119In]": 2156,
+      "[As@]": 2157,
+      "[129IH]": 2158,
+      "[157Dy]": 2159,
+      "[111IH]": 2160,
+      "[230Ra]": 2161,
+      "[144Pr+3]": 2162,
+      "[SiH3+]": 2163,
+      "[3He]": 2164,
+      "[AsH5]": 2165,
+      "[72Se]": 2166,
+      "[95Tc]": 2167,
+      "[103Pd]": 2168,
+      "[121Sn+2]": 2169,
+      "[211Rn]": 2170,
+      "[38SH2]": 2171,
+      "[127IH]": 2172,
+      "[74Br-]": 2173,
+      "[133I-]": 2174,
+      "[100Tc+4]": 2175,
+      "[100Tc]": 2176,
+      "[36Cl-]": 2177,
+      "[89Y+3]": 2178,
+      "[104Rh]": 2179,
+      "[152Sm]": 2180,
+      "[226Ra]": 2181,
+      "[19FH]": 2182,
+      "[104Pd]": 2183,
+      "[148Gd]": 2184,
+      "[157Lu]": 2185,
+      "[33SH2]": 2186,
+      "[121I-]": 2187,
+      "[17FH]": 2188,
+      "[71Se]": 2189,
+      "[157Sm]": 2190,
+      "[148Tb]": 2191,
+      "[164Dy]": 2192,
+      "[15OH2]": 2193,
+      "[15O+]": 2194,
+      "[39K]": 2195,
+      "[40Ar]": 2196,
+      "[50Cr+3]": 2197,
+      "[50Cr]": 2198,
+      "[52Ti]": 2199,
+      "[103Pd+2]": 2200,
+      "[130Ba]": 2201,
+      "[142Pm]": 2202,
+      "[153Gd+3]": 2203,
+      "[151Eu]": 2204,
+      "[103Rh]": 2205,
+      "[124Xe]": 2206,
+      "[152Tb]": 2207,
+      "[17OH2]": 2208,
+      "[20Ne]": 2209,
+      "[52Fe]": 2210,
+      "[94Zr+4]": 2211,
+      "[94Zr]": 2212,
+      "[149Pr]": 2213,
+      "[16OH2]": 2214,
+      "[53Cr+6]": 2215,
+      "[53Cr]": 2216,
+      "[81Br-]": 2217,
+      "[112Pd]": 2218,
+      "[125Xe]": 2219,
+      "[155Gd]": 2220,
+      "[157Gd]": 2221,
+      "[168Yb]": 2222,
+      "[184Os]": 2223,
+      "[166Tb]": 2224,
+      "[221Fr]": 2225,
+      "[212Ra]": 2226,
+      "[75Br-]": 2227,
+      "[79Br-]": 2228,
+      "[113Ag]": 2229,
+      "[23Na]": 2230,
+      "[34Cl-]": 2231,
+      "[34ClH]": 2232,
+      "[38Cl-]": 2233,
+      "[56Fe]": 2234,
+      "[68Cu]": 2235,
+      "[77Br-]": 2236,
+      "[90Zr+4]": 2237,
+      "[90Zr]": 2238,
+      "[102Pd]": 2239,
+      "[154Eu+3]": 2240,
+      "[57Mn]": 2241,
+      "[165Tm]": 2242,
+      "[152Dy]": 2243,
+      "[217At]": 2244,
+      "[77se]": 2245,
+      "[13cH-]": 2246,
+      "[122Te]": 2247,
+      "[156Gd]": 2248,
+      "[124Te]": 2249,
+      "[53Ni]": 2250,
+      "[131Xe]": 2251,
+      "[174Hf+4]": 2252,
+      "[174Hf]": 2253,
+      "[76Se]": 2254,
+      "[168Tm]": 2255,
+      "[167Dy]": 2256,
+      "[154Gd]": 2257,
+      "[95Ru]": 2258,
+      "[210At]": 2259,
+      "[85Br]": 2260,
+      "[59Co]": 2261,
+      "[122Xe]": 2262,
+      "[27Al]": 2263,
+      "[54Cr]": 2264,
+      "[198Hg]": 2265,
+      "[85Rb+]": 2266,
+      "[214Tl]": 2267,
+      "[229Rn]": 2268,
+      "[218Pb]": 2269,
+      "[218Bi]": 2270,
+      "[167Tm+3]": 2271,
+      "[18o+]": 2272,
+      "[P@@H+]": 2273,
+      "[P@H+]": 2274,
+      "[13N+]": 2275,
+      "[212Pb+2]": 2276,
+      "[217Bi]": 2277,
+      "[249Cf+2]": 2278,
+      "[18OH3+]": 2279,
+      "[90Sr-]": 2280,
+      "[Cf+3]": 2281,
+      "[200Hg]": 2282,
+      "[86Tc]": 2283,
+      "[141Pr+3]": 2284,
+      "[141Pr]": 2285,
+      "[16nH]": 2286,
+      "[14NH4+]": 2287,
+      "[132Xe]": 2288,
+      "[83Kr]": 2289,
+      "[70Zn+2]": 2290,
+      "[137Ba+2]": 2291,
+      "[36Ar]": 2292,
+      "[38Ar]": 2293,
+      "[21Ne]": 2294,
+      "[126Xe]": 2295,
+      "[136Xe]": 2296,
+      "[128Xe]": 2297,
+      "[134Xe]": 2298,
+      "[84Kr]": 2299,
+      "[86Kr]": 2300,
+      "[78Kr]": 2301,
+      "[80Kr]": 2302,
+      "[82Kr]": 2303,
+      "[67Zn+2]": 2304,
+      "[65Cu+2]": 2305,
+      "[110Te]": 2306,
+      "[58Fe+3]": 2307,
+      "[142Nd]": 2308,
+      "[38K]": 2309,
+      "[198Au+3]": 2310,
+      "[122IH]": 2311,
+      "[38PH3]": 2312,
+      "[130I-]": 2313,
+      "[40K+]": 2314,
+      "[38K+]": 2315,
+      "[28Mg+2]": 2316,
+      "[208Tl+]": 2317,
+      "[13OH2]": 2318,
+      "[198Bi]": 2319,
+      "[192Bi]": 2320,
+      "[194Bi]": 2321,
+      "[196Bi]": 2322,
+      "[132I-]": 2323,
+      "[83Sr+2]": 2324,
+      "[169Er+3]": 2325,
+      "[122I-]": 2326,
+      "[120I-]": 2327,
+      "[92Sr+2]": 2328,
+      "[126I-]": 2329,
+      "[24Mg]": 2330,
+      "[84Sr]": 2331,
+      "[118Pd+2]": 2332,
+      "[118Pd]": 2333,
+      "[AsH4]": 2334,
+      "[127I-]": 2335,
+      "[9C-]": 2336,
+      "[11CH3+]": 2337,
+      "[17B]": 2338,
+      "[7B]": 2339,
+      "[4HH]": 2340,
+      "[18C-]": 2341,
+      "[22CH3-]": 2342,
+      "[22CH4]": 2343,
+      "[17C-]": 2344,
+      "[15CH3]": 2345,
+      "[16CH3]": 2346,
+      "[11NH3]": 2347,
+      "[21NH3]": 2348,
+      "[11N-]": 2349,
+      "[11NH]": 2350,
+      "[16CH]": 2351,
+      "[17CH2]": 2352,
+      "[99Ru+2]": 2353,
+      "[181Ta+2]": 2354,
+      "[181Ta]": 2355,
+      "[20CH]": 2356,
+      "[32PH2]": 2357,
+      "[55Fe+2]": 2358,
+      "[SH3]": 2359,
+      "[S@H]": 2360,
+      "[UNK]": 2361
+    },
+    "merges": []
+  }
+}

tokenizer_config.json

@@ -0,0 +1,57 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "[CLS]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "[SEP]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "[PAD]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "3": {
+      "content": "[MASK]",
+      "lstrip": true,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2361": {
+      "content": "[UNK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "clean_up_tokenization_spaces": true,
+  "cls_token": "[CLS]",
+  "extra_special_tokens": {},
+  "mask_token": "[MASK]",
+  "model_input_names": [
+    "input_ids",
+    "attention_mask"
+  ],
+  "model_max_length": 512,
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "tokenizer_class": "PreTrainedTokenizerFast",
+  "unk_token": "[UNK]"
+}