reranker-ms-marco-sympathy-docs

Reranker model trained on Sympathy Documentation

This is a Cross Encoder model finetuned from cross-encoder/ms-marco-MiniLM-L6-v2 on the query-doc, anc-pos-neg and anc-pos-neg-2 datasets using the sentence-transformers library. It computes scores for pairs of texts, which can be used for text reranking and semantic search.

Model Details

Model Description

• Model Type: Cross Encoder
• Base model: cross-encoder/ms-marco-MiniLM-L6-v2 <!-- at revision c5ee24cb16019beea0893ab7796b1df96625c6b8 -->
• Maximum Sequence Length: 512 tokens
• Number of Output Labels: 1 label
• Training Datasets:
  • query-doc
  • anc-pos-neg
  • anc-pos-neg-2
• Language: en
• License: apache-2.0

Model Sources

• Documentation: Sentence Transformers Documentation
• Documentation: Cross Encoder Documentation
• Repository: Sentence Transformers on GitHub
• Hugging Face: Cross Encoders on Hugging Face

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

from sentence_transformers import CrossEncoder

# Download from the 🤗 Hub
model = CrossEncoder("emilwin/reranker-ms-marco-sympathy-docs")
# Get scores for pairs of texts
pairs = [
    ['What are the three primary parts of an ADAF, and how does the meta data differ functionally from the results and timeseries in terms of reproducibility?', '# API\n\nADAF API\n========\n\nAPI for working with the ADAF type.\n\nImport this module like this:\n\n```\nfrom sympathy.api import adaf\n\n```\n\nThe ADAF structure\n------------------\n\nAn ADAF consists of three parts: meta data, results, and timeseries.\n\nMeta data contains information about the data in the ADAF. Stuff like when,\nwhere and how it was measured or what parameter values were used to generated\nit. A general guideline is that the meta data should be enough to (at least in\ntheory) reproduce the data in the ADAF.\n\nResults and timeseries contain the actual data. Results are always scalar\nwhereas the timeseries can have any number of values.\n\nTimeseries can come in several systems and each system can contain several\nrasters. Each raster in turn has one basis and any number of timeseries. So\nfor example an experiment where some signals are sampled at 100Hz and others\nare sampled only once per second would have (at least) two rasters. A basis\ndoesn’t have to be uniform but can have samples only every now and then.\n\n'],
    ['What are the three primary parts of an ADAF, and how does the meta data differ functionally from the results and timeseries in terms of reproducibility?', '# Node\n\nTable to ADAF\n=============\n\nConvert a Table into an ADAF, placing its content in the specified container.\n\nDocumentation\n-------------\n\nThe target container in the ADAF is specified in the configuration GUI. If the\ntimeseries container is chosen it is necessary to specify the column in the\nTable which will be the time basis signal in the ADAF. You can also specify\nthe name of the system and raster containers.\n\nSee also Working with ADAF for tips about how to use these conversion\nnodes.\n\n'],
    ['What are the three primary parts of an ADAF, and how does the meta data differ functionally from the results and timeseries in terms of reproducibility?', '# Node\n\nSelect columns in ADAF with structure Table\n===========================================\n\nSelect the columns to keep in ADAF using selection table created by ADAF structure to table\n\nDocumentation\n-------------\n\nUse this node if you’re only interested in some of the data in an ADAF\ne.g. for performance reasons.\n\nThe Table/Tables argument should have four columns, which must be named\nType, System, Raster, and Parameter. These columns hold the names of the\ncorresponding fields in the ADAF/ADAFs.\n\nDefinition\n----------\n\n### Input ports\n\n> **selection**\n> :   Type: table\n> ADAF structure selection\n> \n> **data**\n> :   Type: adaf\n> ADAF data matched with selection\n\n### Output ports\n\n> **data**\n> :   Type: adaf\n> ADAF data after selection\n\n### Configuration\n\n> **Remove selected columns** (complement)\n> :   When enabled, the selected columns will be removed. When disabled, the non\\-selected columns will be removed.\n\n### Related nodes\n\n* Select columns in ADAFs with structure Table\n* Select columns in ADAFs with structure Tables\n* ADAF structure to Table\n* Select categories in ADAFs\n\n### Examples\n\nThis node is used in the following example flows:\n\n* SelectColumns.syx\n\n### Implementation\n\n*class* node\\_select\\_adaf\\_columns.SelectColumnsADAFWithTable\\[source]\n\n'],
    ['What are the three primary parts of an ADAF, and how does the meta data differ functionally from the results and timeseries in terms of reproducibility?', '# Node\n\nSelect categories in ADAFs\n==========================\n\nSelect what catgories to exist in the output ADAFs.\n\nDocumentation\n-------------\n\nA selector of categories in ADAFs can be used to drop parts of ADAFs.\nThe main reason to do this is when the ADAFs contain data that is no longer\nneeded further along a workflow. Dropping the unnecessary data can then be\nused as a way to try to optimize the workflow.\n\nDefinition\n----------\n\n### Input ports\n\n> **port1**\n> :   Type: \\[adaf]\n> Input ADAFs\n\n### Output ports\n\n> **port3**\n> :   Type: \\[adaf]\n> ADAFs with selected categories\n\n### Configuration\n\n> **Select meta group** (select\\_meta)\n> :   Select the meta group for inclusion in the output.\n> \n> **Select specific rasters:** (select\\_rasters)\n> :   Select specific rasters for inclusion in the output.\n> \n> **Select result group** (select\\_res)\n> :   Select the result group for inclusion in the output.\n\n### Related nodes\n\n* ADAF to Table\n\n### Examples\n\nThis node is used in the following example flows:\n\n* SelectCategoryInADAFs.syx\n\n### Implementation\n\n*class* node\\_category\\_selector.CategorySelectorMultiple\\[source]\n\n'],
    ['In Sympathy: What are the three primary parts of an ADAF, and how does the meta data differ functionally from the results and timeseries in terms of reproducibility?', '# API\n\nADAF API\n========\n\nAPI for working with the ADAF type.\n\nImport this module like this:\n\n```\nfrom sympathy.api import adaf\n\n```\n\nThe ADAF structure\n------------------\n\nAn ADAF consists of three parts: meta data, results, and timeseries.\n\nMeta data contains information about the data in the ADAF. Stuff like when,\nwhere and how it was measured or what parameter values were used to generated\nit. A general guideline is that the meta data should be enough to (at least in\ntheory) reproduce the data in the ADAF.\n\nResults and timeseries contain the actual data. Results are always scalar\nwhereas the timeseries can have any number of values.\n\nTimeseries can come in several systems and each system can contain several\nrasters. Each raster in turn has one basis and any number of timeseries. So\nfor example an experiment where some signals are sampled at 100Hz and others\nare sampled only once per second would have (at least) two rasters. A basis\ndoesn’t have to be uniform but can have samples only every now and then.\n\n'],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'What are the three primary parts of an ADAF, and how does the meta data differ functionally from the results and timeseries in terms of reproducibility?',
    [
        '# API\n\nADAF API\n========\n\nAPI for working with the ADAF type.\n\nImport this module like this:\n\n```\nfrom sympathy.api import adaf\n\n```\n\nThe ADAF structure\n------------------\n\nAn ADAF consists of three parts: meta data, results, and timeseries.\n\nMeta data contains information about the data in the ADAF. Stuff like when,\nwhere and how it was measured or what parameter values were used to generated\nit. A general guideline is that the meta data should be enough to (at least in\ntheory) reproduce the data in the ADAF.\n\nResults and timeseries contain the actual data. Results are always scalar\nwhereas the timeseries can have any number of values.\n\nTimeseries can come in several systems and each system can contain several\nrasters. Each raster in turn has one basis and any number of timeseries. So\nfor example an experiment where some signals are sampled at 100Hz and others\nare sampled only once per second would have (at least) two rasters. A basis\ndoesn’t have to be uniform but can have samples only every now and then.\n\n',
        '# Node\n\nTable to ADAF\n=============\n\nConvert a Table into an ADAF, placing its content in the specified container.\n\nDocumentation\n-------------\n\nThe target container in the ADAF is specified in the configuration GUI. If the\ntimeseries container is chosen it is necessary to specify the column in the\nTable which will be the time basis signal in the ADAF. You can also specify\nthe name of the system and raster containers.\n\nSee also Working with ADAF for tips about how to use these conversion\nnodes.\n\n',
        '# Node\n\nSelect columns in ADAF with structure Table\n===========================================\n\nSelect the columns to keep in ADAF using selection table created by ADAF structure to table\n\nDocumentation\n-------------\n\nUse this node if you’re only interested in some of the data in an ADAF\ne.g. for performance reasons.\n\nThe Table/Tables argument should have four columns, which must be named\nType, System, Raster, and Parameter. These columns hold the names of the\ncorresponding fields in the ADAF/ADAFs.\n\nDefinition\n----------\n\n### Input ports\n\n> **selection**\n> :   Type: table\n> ADAF structure selection\n> \n> **data**\n> :   Type: adaf\n> ADAF data matched with selection\n\n### Output ports\n\n> **data**\n> :   Type: adaf\n> ADAF data after selection\n\n### Configuration\n\n> **Remove selected columns** (complement)\n> :   When enabled, the selected columns will be removed. When disabled, the non\\-selected columns will be removed.\n\n### Related nodes\n\n* Select columns in ADAFs with structure Table\n* Select columns in ADAFs with structure Tables\n* ADAF structure to Table\n* Select categories in ADAFs\n\n### Examples\n\nThis node is used in the following example flows:\n\n* SelectColumns.syx\n\n### Implementation\n\n*class* node\\_select\\_adaf\\_columns.SelectColumnsADAFWithTable\\[source]\n\n',
        '# Node\n\nSelect categories in ADAFs\n==========================\n\nSelect what catgories to exist in the output ADAFs.\n\nDocumentation\n-------------\n\nA selector of categories in ADAFs can be used to drop parts of ADAFs.\nThe main reason to do this is when the ADAFs contain data that is no longer\nneeded further along a workflow. Dropping the unnecessary data can then be\nused as a way to try to optimize the workflow.\n\nDefinition\n----------\n\n### Input ports\n\n> **port1**\n> :   Type: \\[adaf]\n> Input ADAFs\n\n### Output ports\n\n> **port3**\n> :   Type: \\[adaf]\n> ADAFs with selected categories\n\n### Configuration\n\n> **Select meta group** (select\\_meta)\n> :   Select the meta group for inclusion in the output.\n> \n> **Select specific rasters:** (select\\_rasters)\n> :   Select specific rasters for inclusion in the output.\n> \n> **Select result group** (select\\_res)\n> :   Select the result group for inclusion in the output.\n\n### Related nodes\n\n* ADAF to Table\n\n### Examples\n\nThis node is used in the following example flows:\n\n* SelectCategoryInADAFs.syx\n\n### Implementation\n\n*class* node\\_category\\_selector.CategorySelectorMultiple\\[source]\n\n',
        '# API\n\nADAF API\n========\n\nAPI for working with the ADAF type.\n\nImport this module like this:\n\n```\nfrom sympathy.api import adaf\n\n```\n\nThe ADAF structure\n------------------\n\nAn ADAF consists of three parts: meta data, results, and timeseries.\n\nMeta data contains information about the data in the ADAF. Stuff like when,\nwhere and how it was measured or what parameter values were used to generated\nit. A general guideline is that the meta data should be enough to (at least in\ntheory) reproduce the data in the ADAF.\n\nResults and timeseries contain the actual data. Results are always scalar\nwhereas the timeseries can have any number of values.\n\nTimeseries can come in several systems and each system can contain several\nrasters. Each raster in turn has one basis and any number of timeseries. So\nfor example an experiment where some signals are sampled at 100Hz and others\nare sampled only once per second would have (at least) two rasters. A basis\ndoesn’t have to be uniform but can have samples only every now and then.\n\n',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

<!-- ### Direct Usage (Transformers) <details><summary>Click to see the direct usage in Transformers</summary> </details> --> <!-- ### Downstream Usage (Sentence Transformers) You can finetune this model on your own dataset. <details><summary>Click to expand</summary> </details> --> <!-- ### Out-of-Scope Use *List how the model may foreseeably be misused and address what users ought not to do with the model.* -->

Evaluation

Metrics

Cross Encoder Reranking

• Dataset: sydoc-tester
• Evaluated with <code>CrossEncoderRerankingEvaluator</code> with these parameters:

  {
      "at_k": 10,
      "always_rerank_positives": false
  }
  

Cross Encoder Reranking

• Datasets: NanoMSMARCO_R100, NanoNFCorpus_R100 and NanoNQ_R100
• Evaluated with <code>CrossEncoderRerankingEvaluator</code> with these parameters:

  {
      "at_k": 10,
      "always_rerank_positives": true
  }
  

Cross Encoder Nano BEIR

• Dataset: NanoBEIR_R100_mean
• Evaluated with <code>CrossEncoderNanoBEIREvaluator</code> with these parameters:

  {
      "dataset_names": [
          "msmarco",
          "nfcorpus",
          "nq"
      ],
      "rerank_k": 100,
      "at_k": 10,
      "always_rerank_positives": true
  }
  

<!-- ## Bias, Risks and Limitations *What are the known or foreseeable issues stemming from this model? You could also flag here known failure cases or weaknesses of the model.* --> <!-- ### Recommendations *What are recommendations with respect to the foreseeable issues? For example, filtering explicit content.* -->

Training Details

Training Datasets

query-doc

• Dataset: query-doc
• Size: 15,230 training samples
• Columns: <code>query</code>, <code>document</code>, and <code>label</code>
• Approximate statistics based on the first 1000 samples:

  	query	document	label
  type	string	string	int
  details	<ul><li>min: 31 characters</li><li>mean: 150.61 characters</li><li>max: 242 characters</li></ul>	<ul><li>min: 163 characters</li><li>mean: 1891.69 characters</li><li>max: 12851 characters</li></ul>	<ul><li>0: ~74.10%</li><li>1: ~25.90%</li></ul>

• Samples:

  query	document	label
  <code>What are the three primary parts of an ADAF, and how does the meta data differ functionally from the results and timeseries in terms of reproducibility?</code>	<code># API<br><br>ADAF API<br>========<br><br>API for working with the ADAF type.<br><br>Import this module like this:<br><br><br>from sympathy.api import adaf<br><br><br><br>The ADAF structure<br>------------------<br><br>An ADAF consists of three parts: meta data, results, and timeseries.<br><br>Meta data contains information about the data in the ADAF. Stuff like when,<br>where and how it was measured or what parameter values were used to generated<br>it. A general guideline is that the meta data should be enough to (at least in<br>theory) reproduce the data in the ADAF.<br><br>Results and timeseries contain the actual data. Results are always scalar<br>whereas the timeseries can have any number of values.<br><br>Timeseries can come in several systems and each system can contain several<br>rasters. Each raster in turn has one basis and any number of timeseries. So<br>for example an experiment where some signals are sampled at 100Hz and others<br>are sampled only once per second would have (at least) two rasters. A basis<br>doesn’t have to be uniform but can have samples on...</code>	<code>1</code>
  <code>What are the three primary parts of an ADAF, and how does the meta data differ functionally from the results and timeseries in terms of reproducibility?</code>	<code># Node<br><br>Table to ADAF<br>=============<br><br>Convert a Table into an ADAF, placing its content in the specified container.<br><br>Documentation<br>-------------<br><br>The target container in the ADAF is specified in the configuration GUI. If the<br>timeseries container is chosen it is necessary to specify the column in the<br>Table which will be the time basis signal in the ADAF. You can also specify<br>the name of the system and raster containers.<br><br>See also Working with ADAF for tips about how to use these conversion<br>nodes.<br><br></code>	<code>0</code>
  <code>What are the three primary parts of an ADAF, and how does the meta data differ functionally from the results and timeseries in terms of reproducibility?</code>	<code># Node<br><br>Select columns in ADAF with structure Table<br>===========================================<br><br>Select the columns to keep in ADAF using selection table created by ADAF structure to table<br><br>Documentation<br>-------------<br><br>Use this node if you’re only interested in some of the data in an ADAF<br>e.g. for performance reasons.<br><br>The Table/Tables argument should have four columns, which must be named<br>Type, System, Raster, and Parameter. These columns hold the names of the<br>corresponding fields in the ADAF/ADAFs.<br><br>Definition<br>----------<br><br>### Input ports<br><br>> selection<br>> : Type: table<br>> ADAF structure selection<br>> <br>> data<br>> : Type: adaf<br>> ADAF data matched with selection<br><br>### Output ports<br><br>> data<br>> : Type: adaf<br>> ADAF data after selection<br><br>### Configuration<br><br>> Remove selected columns (complement)<br>> : When enabled, the selected columns will be removed. When disabled, the non-selected columns will be removed.<br><br>### Related nodes<br><br>* Select columns in ADAFs with structure Table<br>* Sel...</code>	<code>0</code>

• Loss: <code>BinaryCrossEntropyLoss</code> with these parameters:

  {
      "activation_fn": "torch.nn.modules.linear.Identity",
      "pos_weight": 3
  }
  

anc-pos-neg

• Dataset: anc-pos-neg
• Size: 2,435 training samples
• Columns: <code>anchor</code>, <code>positive</code>, and <code>negative</code>
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative
  type	string	string	string
  details	<ul><li>min: 14 characters</li><li>mean: 158.34 characters</li><li>max: 317 characters</li></ul>	<ul><li>min: 163 characters</li><li>mean: 2474.63 characters</li><li>max: 20435 characters</li></ul>	<ul><li>min: 214 characters</li><li>mean: 1745.03 characters</li><li>max: 20435 characters</li></ul>

• Samples:

  anchor	positive	negative
  <code>- Retrieve the time (t) and signal (y) values of "Voltage"?</code>	<code># API ADAF API<br><br>Accessing the data<br>------------------<br><br>The adaf.ADAF object has two members called meta and rescontaining the meta data and results respectively. Both are sympathy.api.adaf.Groupobjects.<br><br>Example of how to use meta (res is completely analogous):<br>: <br>>>> from sympathy.api import adaf<br>>>> import numpy as np<br>>>> f = adaf.ADAF()<br>>>> f.meta.create_column(<br>... 'Duration', np.array([3]), {'unit': 'h'})<br>>>> f.meta.create_column(<br>... 'Relative humidity', np.array([63]), {'unit': '%'})<br>>>> print(f.meta['Duration'].value())<br>[3]<br>>>> print(f.meta['Duration'].attr['unit'])<br><br><br><br>Timeseries can be accessed in two different ways. Either via the membersys or via the member ts. Using sys is generally recommended sincets handles multiple timeseries with the same name across different rasters<br>poorly.<br><br>Example of how to use sys:<br>: ```<br>>>> f.sys.create('Measurement system')<br>>>> f.sys['Measurement system'].create('Raster1')<br>>>> f.sys['Measurement system']['Raster...</code>	<code># API ADAF API<br><br>Class sympathy.api.adaf.Timeseries<br>----------------------------------<br><br>class sympathy.api.adaf.Timeseries(node, data, name: str)<br>: Class representing a timeseries. The values in the timeseries can be<br>accessed as a numpy array via the member y. The timeseries is also<br>connected to a time basis whose values can be accessed as a numpy array<br>via the property t.<br><br>The timeseries can also have any number of attributes. The methodssympathy.api.adaf.Timeseries.unit and sympathy.api.adaf.Timeseries.description retrieve those two attributes. To get<br>all attributes use the method sympathy.api.adaf.Timeseries.get_attributes.<br><br>basis() → sympathy.typeutils.adaf.Column<br>: Return the timeseries data basis as a sympathy.api.adaf.Column.<br><br>description() → str<br>: Return the description attribute or an empty string if it is not set.<br><br>property dtype*: dtype*<br>: dtype of timeseries.<br><br>get_attributes() → Dict[str, int \
  <code>How can you add a custom attribute (e.g., {'description': 'Indicates system health'}) to a signal named "Status" in a raster, and how would you later retrieve this attribute?</code>	<code># API ADAF API<br><br>Accessing the data<br>------------------<br><br>The adaf.ADAF object has two members called meta and rescontaining the meta data and results respectively. Both are sympathy.api.adaf.Groupobjects.<br><br>Example of how to use meta (res is completely analogous):<br>: <br>>>> from sympathy.api import adaf<br>>>> import numpy as np<br>>>> f = adaf.ADAF()<br>>>> f.meta.create_column(<br>... 'Duration', np.array([3]), {'unit': 'h'})<br>>>> f.meta.create_column(<br>... 'Relative humidity', np.array([63]), {'unit': '%'})<br>>>> print(f.meta['Duration'].value())<br>[3]<br>>>> print(f.meta['Duration'].attr['unit'])<br><br><br><br>Timeseries can be accessed in two different ways. Either via the membersys or via the member ts. Using sys is generally recommended sincets handles multiple timeseries with the same name across different rasters<br>poorly.<br><br>Example of how to use sys:<br>: ```<br>>>> f.sys.create('Measurement system')<br>>>> f.sys['Measurement system'].create('Raster1')<br>>>> f.sys['Measurement system']['Raster...</code>	<code># API ADAF API<br><br>Class sympathy.api.adaf.RasterN<br>-------------------------------<br><br>class sympathy.api.adaf.RasterN(record, system: str, name: str)<br>: Represents a raster with a single time basis and any number of timeseries<br>columns.<br><br>property attr*: Attributes*<br>: Raster level attributes.<br><br>basis_column() → sympathy.typeutils.adaf.Column<br>: Return the time basis for this raster. The returned object is of typesympathy.api.adaf.Column.<br><br>create_basis(data: ndarray, *attributes: Dict[str, int \
  <code>How can you add a custom attribute (e.g., {'description': 'Indicates system health'}) to a signal named "Status" in a raster, and how would you later retrieve this attribute?</code>	<code># API ADAF API<br><br>Accessing the data<br>------------------<br><br>The adaf.ADAF object has two members called meta and rescontaining the meta data and results respectively. Both are sympathy.api.adaf.Groupobjects.<br><br>Example of how to use meta (res is completely analogous):<br>: <br>>>> from sympathy.api import adaf<br>>>> import numpy as np<br>>>> f = adaf.ADAF()<br>>>> f.meta.create_column(<br>... 'Duration', np.array([3]), {'unit': 'h'})<br>>>> f.meta.create_column(<br>... 'Relative humidity', np.array([63]), {'unit': '%'})<br>>>> print(f.meta['Duration'].value())<br>[3]<br>>>> print(f.meta['Duration'].attr['unit'])<br><br><br><br>Timeseries can be accessed in two different ways. Either via the membersys or via the member ts. Using sys is generally recommended sincets handles multiple timeseries with the same name across different rasters<br>poorly.<br><br>Example of how to use sys:<br>: ```<br>>>> f.sys.create('Measurement system')<br>>>> f.sys['Measurement system'].create('Raster1')<br>>>> f.sys['Measurement system']['Raster...</code>	<code># API ADAF API<br><br>Class sympathy.api.adaf.Timeseries<br>----------------------------------<br><br>class sympathy.api.adaf.Timeseries(node, data, name: str)<br>: Class representing a timeseries. The values in the timeseries can be<br>accessed as a numpy array via the member y. The timeseries is also<br>connected to a time basis whose values can be accessed as a numpy array<br>via the property t.<br><br>The timeseries can also have any number of attributes. The methodssympathy.api.adaf.Timeseries.unit and sympathy.api.adaf.Timeseries.description retrieve those two attributes. To get<br>all attributes use the method sympathy.api.adaf.Timeseries.get_attributes.<br><br>basis() → sympathy.typeutils.adaf.Column<br>: Return the timeseries data basis as a sympathy.api.adaf.Column.<br><br>description() → str<br>: Return the description attribute or an empty string if it is not set.<br><br>property dtype*: dtype*<br>: dtype of timeseries.<br><br>get_attributes() → Dict[str, int \

• Loss: <code>CachedMultipleNegativesRankingLoss</code> with these parameters:

  {
      "scale": 10.0,
      "num_negatives": 4,
      "activation_fn": "torch.nn.modules.activation.Sigmoid",
      "mini_batch_size": 32
  }
  

anc-pos-neg-2

• Dataset: anc-pos-neg-2
• Size: 1,219 training samples
• Columns: <code>anchor</code>, <code>positive</code>, and <code>negative</code>
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative
  type	string	string	string
  details	<ul><li>min: 0 characters</li><li>mean: 89.95 characters</li><li>max: 143 characters</li></ul>	<ul><li>min: 496 characters</li><li>mean: 1890.31 characters</li><li>max: 7315 characters</li></ul>	<ul><li>min: 160 characters</li><li>mean: 2026.14 characters</li><li>max: 12851 characters</li></ul>

• Samples:

  anchor	positive	negative
  <code>Is it possible to run a Node.js script or environment from within a Python program?</code>	<code># Nodes in python<br><br>Working with nodes<br>------------------<br><br>Nodes store the changes made during configure and when the parameters are<br>changed. They produce a list of data elements when executed and expect a list of<br>data elements as input, this makes it possible to easily connect the data<br>between nodes. Note that the ordering of inputs and outputs is important and<br>should match the declaration order in the node definition.<br><br>The code example below demonstrates how to use the result produced by one node as<br>input for another.<br><br><br>random_table = library.node('Random Table')<br>rt_output = random_table.execute()<br><br>item_to_list = library.node('Item to List')<br>itl_output = item_to_list.execute(rt_output)<br><br>assert itl_output[0][0].equal_to(rt_output[0])<br><br><br>The code example below demonstrates how to use the result produced by multiple<br>nodes as input for another.<br><br>```<br>random_table0 = library.node('Random Table')<br>rt_output0 = random_table.execute()<br><br>random_table1 = library.node('Random Table')<br>rt_outpu...</code>	<code><br><br>Nodes<br>=====<br><br>A node is defined as a Python class which inherits fromsympathy.api.node.Node. All node definitions should be in files with<br>filenames matching node_*.py and be placed in the nodes folder of a node<br>library. See Libraries for information about where to put nodes in your<br>library. Nodes can be placed in subfolders and multiple nodes can be defined in<br>the same file.<br><br>Node definition<br>---------------<br><br>The following class variables make up the definition of a node.<br><br>Note<br><br>The fields name and nodeid are needed to generate the node. If any<br>of these two are missing any attempt at creating this node stops immediately<br>without any error message. This can be a good way of e.g. creating a<br>superclass for multiple node classes.<br><br>name<br>: Required.<br><br>The name of the node, is what the user will rely on to identify the node. It<br>will show in the library view and in the node’s tooltip. It will also be used<br>as the default label of any instance of the node in a flow.<br><br>Try to keep the ...</code>
  <code>Is it possible to run a Node.js script or environment from within a Python program?</code>	<code># Nodes in python<br><br>Working with nodes<br>------------------<br><br>Nodes store the changes made during configure and when the parameters are<br>changed. They produce a list of data elements when executed and expect a list of<br>data elements as input, this makes it possible to easily connect the data<br>between nodes. Note that the ordering of inputs and outputs is important and<br>should match the declaration order in the node definition.<br><br>The code example below demonstrates how to use the result produced by one node as<br>input for another.<br><br><br>random_table = library.node('Random Table')<br>rt_output = random_table.execute()<br><br>item_to_list = library.node('Item to List')<br>itl_output = item_to_list.execute(rt_output)<br><br>assert itl_output[0][0].equal_to(rt_output[0])<br><br><br>The code example below demonstrates how to use the result produced by multiple<br>nodes as input for another.<br><br>```<br>random_table0 = library.node('Random Table')<br>rt_output0 = random_table.execute()<br><br>random_table1 = library.node('Random Table')<br>rt_outpu...</code>	<code># Nodes in python<br><br>Reference<br>---------<br><br>exception sympathy.app.interactive.InteractiveNotNodeError[source]<br><br>class sympathy.app.interactive.SyiLibrary(context, library, name_library, paths)[source]<br>: A library of nodes that can be configured and executed in Python code.<br><br>Should not be instantiated directly. Instead call sympathy.app.interactive.load_library.<br><br>node(nid, fuzzy_names=True) → sympathy.app.interactive.SyiNode[source]<br>: Attempt to find nid in the library.<br><br>Argument nid can be either a node id or a node name. If no matching<br>node can be found a KeyError is raised.<br><br>If fuzzy_names is True (the default) and nid doesn’t match any<br>node exactly, it is used as a pattern that the node name must match.<br>The characters of the pattern must appear in the node name in the same<br>order as in the pattern, but must not be of the same case, and may have<br>other characters in between them. If multiple nodes match the pattern a<br>KeyError is raised.<br><br>nodeids() ...</code>
  <code>Is it possible to run a Node.js script or environment from within a Python program?</code>	<code># Nodes in python<br><br>Working with nodes<br>------------------<br><br>Nodes store the changes made during configure and when the parameters are<br>changed. They produce a list of data elements when executed and expect a list of<br>data elements as input, this makes it possible to easily connect the data<br>between nodes. Note that the ordering of inputs and outputs is important and<br>should match the declaration order in the node definition.<br><br>The code example below demonstrates how to use the result produced by one node as<br>input for another.<br><br><br>random_table = library.node('Random Table')<br>rt_output = random_table.execute()<br><br>item_to_list = library.node('Item to List')<br>itl_output = item_to_list.execute(rt_output)<br><br>assert itl_output[0][0].equal_to(rt_output[0])<br><br><br>The code example below demonstrates how to use the result produced by multiple<br>nodes as input for another.<br><br>```<br>random_table0 = library.node('Random Table')<br>rt_output0 = random_table.execute()<br><br>random_table1 = library.node('Random Table')<br>rt_outpu...</code>	<code># API<br><br>Datasource API<br>==============<br><br>API for working with the Datasource type.<br><br>Import this module like this:<br><br><br>from sympathy.api import datasource<br><br><br>Class datasource.Datasource<br>-----------------------------<br><br>class sympathy.api.datasource.Datasource(*filename: str \

• Loss: <code>CachedMultipleNegativesRankingLoss</code> with these parameters:

  {
      "scale": 10.0,
      "num_negatives": 4,
      "activation_fn": "torch.nn.modules.activation.Sigmoid",
      "mini_batch_size": 32
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_train_batch_size: 64
• per_device_eval_batch_size: 64
• learning_rate: 0.0001
• num_train_epochs: 10
• warmup_ratio: 0.1
• dataloader_num_workers: 4
• load_best_model_at_end: True
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

<details><summary>Click to expand</summary>

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: epoch
• prediction_loss_only: True
• per_device_train_batch_size: 64
• per_device_eval_batch_size: 64
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 0.0001
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• num_train_epochs: 10
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.1
• warmup_steps: 0
• log_level: passive
• log_level_replica: warning
• log_on_each_node: True
• logging_nan_inf_filter: True
• save_safetensors: True
• save_on_each_node: False
• save_only_model: False
• restore_callback_states_from_checkpoint: False
• no_cuda: False
• use_cpu: False
• use_mps_device: False
• seed: 42
• data_seed: None
• jit_mode_eval: False
• use_ipex: False
• bf16: False
• fp16: False
• fp16_opt_level: O1
• half_precision_backend: auto
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• local_rank: 0
• ddp_backend: None
• tpu_num_cores: None
• tpu_metrics_debug: False
• debug: []
• dataloader_drop_last: False
• dataloader_num_workers: 4
• dataloader_prefetch_factor: None
• past_index: -1
• disable_tqdm: False
• remove_unused_columns: True
• label_names: None
• load_best_model_at_end: True
• ignore_data_skip: False
• fsdp: []
• fsdp_min_num_params: 0
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
• fsdp_transformer_layer_cls_to_wrap: None
• accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch
• optim_args: None
• adafactor: False
• group_by_length: False
• length_column_name: length
• ddp_find_unused_parameters: None
• ddp_bucket_cap_mb: None
• ddp_broadcast_buffers: False
• dataloader_pin_memory: True
• dataloader_persistent_workers: False
• skip_memory_metrics: True
• use_legacy_prediction_loop: False
• push_to_hub: False
• resume_from_checkpoint: None
• hub_model_id: None
• hub_strategy: every_save
• hub_private_repo: None
• hub_always_push: False
• hub_revision: None
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• include_for_metrics: []
• eval_do_concat_batches: True
• fp16_backend: auto
• push_to_hub_model_id: None
• push_to_hub_organization: None
• mp_parameters:
• auto_find_batch_size: False
• full_determinism: False
• torchdynamo: None
• ray_scope: last
• ddp_timeout: 1800
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• include_tokens_per_second: False
• include_num_input_tokens_seen: False
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• eval_on_start: False
• use_liger_kernel: False
• liger_kernel_config: None
• eval_use_gather_object: False
• average_tokens_across_devices: False
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: round_robin
• router_mapping: {}
• learning_rate_mapping: {}

</details>

Training Logs

<details><summary>Click to expand</summary>

The model card is truncated. Read the rest on Hugging Face.