Skip to content

Format of the Queries and Answers for the Molecular Verifier Server

All queries to the Molecular Verifier server and its responses follow specific formats defined using Pydantic models. Below we outline the expected structure for both requests and responses.

MolecularVerifierServerQuery

Bases: BaseModel

Input query model for the Molecular Verifier server.

Represents a complete request to the molecular verifier service, containing metadata for scoring and completions to evaluate.

Attributes:

Name Type Description
metadata List[dict[str, Any]]

List of metadata dictionaries, one per query item. Each dictionary will be converted to a MolecularVerifierMetadata object for scoring.
query List[str]

List of completion strings from the language model.

The parser extracts content between these tags based on 'parsing_method'
prompts Optional[List[str]]

Optional. Original prompts used to generate the completions. Useful for tracking and debugging. If provided, should have same length as query list.

Notes: This model also supports single-item requests not in list form. For example: 

{
  "query": "Here is a molecules: <answer>CC(C)Cc1ccc(cc1)C(C)C(=O)O</answer>",
  "prompt": "Generate a molecule that binds to my target protein with high affinity and has more than 3 rotatable bonds.",
  "metadata": {
    "properties": ["CalcNumRotatableBonds", "sample_228234_model_0"],
    "objectives": ["above", "minimize"],
    "target": [3.0, 0.0]
  }
}
is also accepted and will be internally converted to: 
{      "query": ["Here is a molecules: <answer>CC(C)Cc1ccc(cc1)C(C)C(=O)O</answer>"],
  "prompt": ["Generate a molecule that binds to my target protein with high affinity and has more than 3 rotatable bonds."],
  "metadata": [
    {
      "properties": ["CalcNumRotatableBonds", "sample_228234_model_0"],
      "objectives": ["above", "minimize"],
      "target": [3.0, 0.0]
    }
  ]
}

Example: 

{
  "query": "Here is a molecules: <answer>CC(C)Cc1ccc(cc1)C(C)C(=O)O</answer>",
  "prompt": "Generate a molecule that binds to my target protein with high affinity and has more than 3 rotatable bonds.",
  "metadata": [
    {
      "properties": ["CalcNumRotatableBonds", "sample_228234_model_0"],
      "objectives": ["above", "minimize"],
      "target": [3.0, 0.0]
    }
  ]
}

Source code in mol_gen_docking/server_utils/utils.py 
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
class MolecularVerifierServerQuery(BaseModel):
    """Input query model for the Molecular Verifier server.

    Represents a complete request to the molecular verifier service,
    containing metadata for scoring and completions to evaluate.

    Attributes:
        metadata: List of metadata dictionaries, one per query item.
            Each dictionary will be converted to a MolecularVerifierMetadata
            object for scoring.

        query: List of completion strings from the language model.

            The parser extracts content between these tags based on
            'parsing_method'

        prompts: Optional. Original prompts used to generate the completions.
            Useful for tracking and debugging. If provided, should have
            same length as query list.

    Notes:
    This model also supports single-item requests not in list form.
    For example:
    ```json
    {
      "query": "Here is a molecules: <answer>CC(C)Cc1ccc(cc1)C(C)C(=O)O</answer>",
      "prompt": "Generate a molecule that binds to my target protein with high affinity and has more than 3 rotatable bonds.",
      "metadata": {
        "properties": ["CalcNumRotatableBonds", "sample_228234_model_0"],
        "objectives": ["above", "minimize"],
        "target": [3.0, 0.0]
      }
    }
    ```
    is also accepted and will be internally converted to:
    ```json
    {      "query": ["Here is a molecules: <answer>CC(C)Cc1ccc(cc1)C(C)C(=O)O</answer>"],
      "prompt": ["Generate a molecule that binds to my target protein with high affinity and has more than 3 rotatable bonds."],
      "metadata": [
        {
          "properties": ["CalcNumRotatableBonds", "sample_228234_model_0"],
          "objectives": ["above", "minimize"],
          "target": [3.0, 0.0]
        }
      ]
    }
    ```

    Example:
    ```json
    {
      "query": "Here is a molecules: <answer>CC(C)Cc1ccc(cc1)C(C)C(=O)O</answer>",
      "prompt": "Generate a molecule that binds to my target protein with high affinity and has more than 3 rotatable bonds.",
      "metadata": [
        {
          "properties": ["CalcNumRotatableBonds", "sample_228234_model_0"],
          "objectives": ["above", "minimize"],
          "target": [3.0, 0.0]
        }
      ]
    }
    ```
    """

    metadata: List[dict[str, Any]] = Field(
        ..., description="List of metadata dictionaries, one per query item."
    )
    query: List[str] = Field(
        ..., description="List of completion strings from the language model."
    )
    prompts: Optional[List[str]] = Field(
        None, description="Original prompts used to generate the completions."
    )

    @field_validator("metadata", "query", "prompts", mode="before")
    @classmethod
    def convert_to_lists(cls, v: Any) -> Any:
        """Automatically convert fields to lists if they are not already.

        Ensures that metadata, query, and prompts fields are lists.
        If a field is not a list, it will be wrapped in a list.
        None values are left as-is for optional fields.

        Args:
            v: The field value to potentially convert

        Returns:
            The value as a list, or None if the input is None
        """
        if v is None:
            return None
        if not isinstance(v, list):
            return [v]
        return v

    @model_validator(mode="after")
    def validate_equal_lengths(self) -> "MolecularVerifierServerQuery":
        """Validate that all fields have equal length.

        Ensures that metadata, query, and prompts lists all have the same length.
        This validator runs after all fields have been set and converted.

        Returns:
            The validated instance

        Raises:
            ValueError: If field lengths are not equal
        """
        # Extract lengths of non-None fields
        lengths = {}
        lengths["metadata"] = len(self.metadata)
        lengths["query"] = len(self.query)
        if self.prompts is not None:
            lengths["prompts"] = len(self.prompts)
        unique_lengths = set(lengths.values())
        if len(unique_lengths) > 1:
            raise ValueError(f"All query fields must have equal length. Got: {lengths}")
        return self

convert_to_lists(v) classmethod

Automatically convert fields to lists if they are not already.

Ensures that metadata, query, and prompts fields are lists. If a field is not a list, it will be wrapped in a list. None values are left as-is for optional fields.

Parameters:

Name Type Description Default
v Any

The field value to potentially convert

 required

Returns:

Type Description
Any

The value as a list, or None if the input is None
Source code in mol_gen_docking/server_utils/utils.py 
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
@field_validator("metadata", "query", "prompts", mode="before")
@classmethod
def convert_to_lists(cls, v: Any) -> Any:
    """Automatically convert fields to lists if they are not already.

    Ensures that metadata, query, and prompts fields are lists.
    If a field is not a list, it will be wrapped in a list.
    None values are left as-is for optional fields.

    Args:
        v: The field value to potentially convert

    Returns:
        The value as a list, or None if the input is None
    """
    if v is None:
        return None
    if not isinstance(v, list):
        return [v]
    return v

validate_equal_lengths()

Validate that all fields have equal length.

Ensures that metadata, query, and prompts lists all have the same length. This validator runs after all fields have been set and converted.

Returns:

Type Description
MolecularVerifierServerQuery

The validated instance

Raises:

Type Description
ValueError

If field lengths are not equal
Source code in mol_gen_docking/server_utils/utils.py 
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
@model_validator(mode="after")
def validate_equal_lengths(self) -> "MolecularVerifierServerQuery":
    """Validate that all fields have equal length.

    Ensures that metadata, query, and prompts lists all have the same length.
    This validator runs after all fields have been set and converted.

    Returns:
        The validated instance

    Raises:
        ValueError: If field lengths are not equal
    """
    # Extract lengths of non-None fields
    lengths = {}
    lengths["metadata"] = len(self.metadata)
    lengths["query"] = len(self.query)
    if self.prompts is not None:
        lengths["prompts"] = len(self.prompts)
    unique_lengths = set(lengths.values())
    if len(unique_lengths) > 1:
        raise ValueError(f"All query fields must have equal length. Got: {lengths}")
    return self

MolecularVerifierServerMetadata

Bases: BaseModel

Metadata returned with each scored molecule.

Aggregates detailed scoring information from all verifier types (Generation, Molecular Property, and Reaction). Each field may be populated or empty depending on which verifier was used.

Attributes:

Name Type Description
parsed_answer str

The parsed answer extracted from the model completion.
generation_verifier_metadata Optional[GenerationVerifierMetadataModel]

Metadata from the generation verifier containing:

  • properties: List of evaluated property names
  • individual_rewards: Individual reward for each property
  • all_smi_rewards: Rewards for all SMILES found in completion
  • all_smi: List of all extracted SMILES strings
  • smiles_extraction_failure: Error message if SMILES extraction failed
mol_prop_verifier_metadata Optional[MolPropVerifierMetadataModel]

Metadata from the molecular property verifier containing:

  • extracted_answer: The numerical answer extracted from the completion
  • extraction_success: Whether the answer extraction was successful
reaction_verifier_metadata Optional[ReactionVerifierMetadataModel]

Metadata from the reaction verifier containing:

  • valid: Proportion of valid reaction steps (0.0 to 1.0)
  • correct_product: Product correctness or similarity to target molecule
  • correct_reactant: Whether all building blocks used are valid
Source code in mol_gen_docking/server_utils/utils.py 
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
class MolecularVerifierServerMetadata(BaseModel):
    """Metadata returned with each scored molecule.

    Aggregates detailed scoring information from all verifier types (Generation,
    Molecular Property, and Reaction). Each field may be populated or empty
    depending on which verifier was used.

    Attributes:
        parsed_answer: The parsed answer extracted from the model completion.
        generation_verifier_metadata: Metadata from the generation verifier containing:

            - properties: List of evaluated property names
            - individual_rewards: Individual reward for each property
            - all_smi_rewards: Rewards for all SMILES found in completion
            - all_smi: List of all extracted SMILES strings
            - smiles_extraction_failure: Error message if SMILES extraction failed

        mol_prop_verifier_metadata: Metadata from the molecular property verifier containing:

            - extracted_answer: The numerical answer extracted from the completion
            - extraction_success: Whether the answer extraction was successful

        reaction_verifier_metadata: Metadata from the reaction verifier containing:

            - valid: Proportion of valid reaction steps (0.0 to 1.0)
            - correct_product: Product correctness or similarity to target molecule
            - correct_reactant: Whether all building blocks used are valid
    """

    parsed_answer: str = Field(
        "",
        description="The parsed answer extracted from the model completion.",
    )
    generation_verifier_metadata: Optional[GenerationVerifierMetadataModel] = Field(
        None,
        description="Metadata from the generation verifier, if applicable.",
    )
    mol_prop_verifier_metadata: Optional[MolPropVerifierMetadataModel] = Field(
        None,
        description="Metadata from the molecular property verifier, if applicable.",
    )
    reaction_verifier_metadata: Optional[ReactionVerifierMetadataModel] = Field(
        None,
        description="Metadata from the reaction verifier, if applicable.",
    )

MolecularVerifierServerResponse

Bases: BaseModel

Response from the Molecular Verifier server when the sever is in single-item mode (default).

Contains the computed reward score and detailed metadata for a single evaluated completion.

Attributes:

Name Type Description
reward float

Overall reward score combining all evaluated properties. Typically normalized to [0.0, 1.0] range when rescaling is enabled.
error Optional[str]

Error message if scoring failed. None if successful.
meta MolecularVerifierServerMetadata

Metadata dictionary with detailed scoring information. Contains extraction failures, property rewards, and verification results for the completion.
next_turn_feedback Optional[str]

Optional feedback for multi-turn conversations. Can be used to guide subsequent model generations.
Example 
{
  "reward": 0.75,
  "error": null,
  "meta": {
      "smiles_extraction_failure": "",
      "all_smi": ["CC(C)Cc1ccc(cc1)"],
      "all_smi_rewards": [0.75],
      "properties": ["GSK3B", "CalcLogP"],
      "individual_rewards": [1.0, 0.5]
    },
  "next_turn_feedback": null
}
Source code in mol_gen_docking/server_utils/utils.py 
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
class MolecularVerifierServerResponse(BaseModel):
    """Response from the Molecular Verifier server when the sever is in single-item mode (default).

    Contains the computed reward score and detailed metadata for a single
    evaluated completion.

    Attributes:
        reward: Overall reward score combining all evaluated properties.
            Typically normalized to [0.0, 1.0] range when rescaling is enabled.

        error: Error message if scoring failed. None if successful.

        meta: Metadata dictionary with detailed scoring information.
            Contains extraction failures, property rewards, and verification
            results for the completion.

        next_turn_feedback: Optional feedback for multi-turn conversations.
            Can be used to guide subsequent model generations.

    Example:
        ```json
        {
          "reward": 0.75,
          "error": null,
          "meta": {
              "smiles_extraction_failure": "",
              "all_smi": ["CC(C)Cc1ccc(cc1)"],
              "all_smi_rewards": [0.75],
              "properties": ["GSK3B", "CalcLogP"],
              "individual_rewards": [1.0, 0.5]
            },
          "next_turn_feedback": null
        }
        ```
    """

    reward: float = Field(
        ..., description="Overall reward score combining all evaluated properties."
    )
    error: Optional[str] = Field(None, description="Error message if scoring failed.")
    meta: MolecularVerifierServerMetadata = Field(
        default_factory=MolecularVerifierServerMetadata,
        description="Metadata dictionary with detailed scoring information.",
    )
    next_turn_feedback: Optional[str] = Field(
        None, description="Optional feedback for multi-turn conversations."
    )

    @field_validator("reward")
    @classmethod
    def validate_reward(cls, v: Any) -> float:
        """Validate and normalize reward value.

        If reward is None or NaN, sets it to 0.0.
        Ensures reward is a valid float.

        Args:
            v: The reward value to validate

        Returns:
            The validated reward value (float), or 0.0 if None/NaN

        Raises:
            ValueError: If reward is not a valid numeric type
        """
        # Handle None case
        if v is None:
            return 0.0
        # Handle NaN case
        if math.isnan(float(v)):
            return 0.0
        # Check if it's a numeric type
        if not isinstance(v, (int, float)):
            raise ValueError(f"reward must be a float, got {type(v).__name__}")
        return float(v)

validate_reward(v) classmethod

Validate and normalize reward value.

If reward is None or NaN, sets it to 0.0. Ensures reward is a valid float.

Parameters:

Name Type Description Default
v Any

The reward value to validate

 required

Returns:

Type Description
float

The validated reward value (float), or 0.0 if None/NaN

Raises:

Type Description
ValueError

If reward is not a valid numeric type
Source code in mol_gen_docking/server_utils/utils.py 
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
@field_validator("reward")
@classmethod
def validate_reward(cls, v: Any) -> float:
    """Validate and normalize reward value.

    If reward is None or NaN, sets it to 0.0.
    Ensures reward is a valid float.

    Args:
        v: The reward value to validate

    Returns:
        The validated reward value (float), or 0.0 if None/NaN

    Raises:
        ValueError: If reward is not a valid numeric type
    """
    # Handle None case
    if v is None:
        return 0.0
    # Handle NaN case
    if math.isnan(float(v)):
        return 0.0
    # Check if it's a numeric type
    if not isinstance(v, (int, float)):
        raise ValueError(f"reward must be a float, got {type(v).__name__}")
    return float(v)

BatchMolecularVerifierServerResponse

Bases: BaseModel

Response from the Molecular Verifier server when in batch mode.

Contains the computed reward scores and detailed metadata for each evaluated completion.

Attributes:

Name Type Description
rewards List[float]

List of individual reward scores, one per evaluated completion in the request.
error Optional[str]

Error message if scoring failed. None if successful.
metas List[MolecularVerifierServerMetadata]

List of metadata dictionaries with detailed scoring information. Contains extraction failures, property rewards, and verification results for each completion. One metadata object per query item.
next_turn_feedback Optional[str]

Optional feedback for multi-turn conversations. Can be used to guide subsequent model generations.
Example 
{
  "rewards": [0.75],
  "error": null,
  "metas": [
     "generation_verifier_metadata": {
            "smiles_extraction_failure": "",
            "all_smi": ["CC(C)Cc1ccc(cc1)"],
            "all_smi_rewards": [0.75],
            "properties": ["GSK3B", "CalcLogP"],
            "individual_rewards": [1.0, 0.5]
        }
    }
  ],
  "next_turn_feedback": null
}
Source code in mol_gen_docking/server_utils/utils.py 
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
class BatchMolecularVerifierServerResponse(BaseModel):
    """Response from the Molecular Verifier server when in batch mode.

    Contains the computed reward scores and detailed metadata for each
    evaluated completion.

    Attributes:
        rewards: List of individual reward scores, one per evaluated
            completion in the request.

        error: Error message if scoring failed. None if successful.

        metas: List of metadata dictionaries with detailed scoring information.
            Contains extraction failures, property rewards, and verification
            results for each completion. One metadata object per query item.

        next_turn_feedback: Optional feedback for multi-turn conversations.
            Can be used to guide subsequent model generations.

    Example:
        ```json
        {
          "rewards": [0.75],
          "error": null,
          "metas": [
             "generation_verifier_metadata": {
                    "smiles_extraction_failure": "",
                    "all_smi": ["CC(C)Cc1ccc(cc1)"],
                    "all_smi_rewards": [0.75],
                    "properties": ["GSK3B", "CalcLogP"],
                    "individual_rewards": [1.0, 0.5]
                }
            }
          ],
          "next_turn_feedback": null
        }
        ```
    """

    rewards: List[float] = Field(
        ...,
        description="List of individual reward scores, one per evaluated completion.",
    )
    error: Optional[str] = Field(None, description="Error message if scoring failed.")
    metas: List[MolecularVerifierServerMetadata] = Field(
        default_factory=list,
        description="List of metadata with detailed scoring information.",
    )
    next_turn_feedback: Optional[str] = Field(
        None, description="Optional feedback for multi-turn conversations."
    )

    @field_validator("rewards")
    @classmethod
    def validate_rewards(cls, v: Any) -> List[float]:
        """Validate and normalize reward value.

        If any reward is None or NaN, sets it to 0.0.
        Ensures reward is a valid float.

        Args:
            v: The reward value to validate

        Returns:
            The validated reward value (float), or 0.0 if None/NaN

        Raises:
            ValueError: If reward is not a valid numeric type
        """
        assert isinstance(v, list)
        # Handle None case
        for i, val in enumerate(v):
            if val is None:
                val = 0.0
            # Handle NaN case
            if math.isnan(float(val)):
                val = 0.0
            # Check if it's a numeric type
            if not isinstance(val, (int, float)):
                raise ValueError(f"reward must be a float, got {type(val).__name__}")
            v[i] = float(val)
        return v

validate_rewards(v) classmethod

Validate and normalize reward value.

If any reward is None or NaN, sets it to 0.0. Ensures reward is a valid float.

Parameters:

Name Type Description Default
v Any

The reward value to validate

 required

Returns:

Type Description
List[float]

The validated reward value (float), or 0.0 if None/NaN

Raises:

Type Description
ValueError

If reward is not a valid numeric type
Source code in mol_gen_docking/server_utils/utils.py 
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
@field_validator("rewards")
@classmethod
def validate_rewards(cls, v: Any) -> List[float]:
    """Validate and normalize reward value.

    If any reward is None or NaN, sets it to 0.0.
    Ensures reward is a valid float.

    Args:
        v: The reward value to validate

    Returns:
        The validated reward value (float), or 0.0 if None/NaN

    Raises:
        ValueError: If reward is not a valid numeric type
    """
    assert isinstance(v, list)
    # Handle None case
    for i, val in enumerate(v):
        if val is None:
            val = 0.0
        # Handle NaN case
        if math.isnan(float(val)):
            val = 0.0
        # Check if it's a numeric type
        if not isinstance(val, (int, float)):
            raise ValueError(f"reward must be a float, got {type(val).__name__}")
        v[i] = float(val)
    return v