inoid commited on
Commit
a4c8bb4
1 Parent(s): 945b864

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +50 -53
README.md CHANGED
@@ -59,9 +59,9 @@ Para más información sobre la dataset card metadata ver: https://github.com/hu
59
 
60
  <!-- Si queréis incluir una versión de la Dataset Card en español, enlazarla aquí al principio (e.g. `README_es.md`).-->
61
 
62
- Este dataset agrupa y organiza varios dataset presentes en hugginface (p.ej: PlanTL-GOB-ES/cantemist-ner, PlanTL-GOB-ES/pharmaconer)
63
- y otros recursos públicos creados por investigadores con distintos formatos (p.ej.; MedLexSp )
64
- para permitir ser fuente de conocimiento de grandes modelos de lenguaje en idioma español para el dominio médico.
65
 
66
  <!--
67
  This dataset groups and organizes several datasets present in hugginface (e.g.: PlanTL-GOB-ES/cantemist-ner, PlanTL-GOB-ES/pharmaconer)
@@ -92,7 +92,7 @@ para permitir ser fuente de conocimiento de grandes modelos de lenguaje en idiom
92
  ## Uses
93
 
94
  <!-- Address questions around how the dataset is intended to be used. -->
95
- Se sugiere el uso de este dataset para lograr el autojuste y prentrenamiento de LLM para el dominio médico con información en idioma español.
96
 
97
  ### Direct Use
98
 
@@ -102,8 +102,8 @@ Fine Tuning an LLM instruction in Spanish language with question prompts and ans
102
  ### Out-of-Scope Use
103
 
104
  <!-- This section addresses misuse, malicious use, and uses that the dataset will not work well for. -->
105
- Los creadores del dataset no se hacen responsable de resultados nocivos que puedan generar los modelos al ser entrenados con esta información.
106
- Se sugiere un proceso de evaluación riguroso con especialistas de los resultados generados por modelos de LLM entrenados.
107
 
108
  ## Dataset Structure
109
 
@@ -111,17 +111,17 @@ Se sugiere un proceso de evaluación riguroso con especialistas de los resultado
111
 
112
  <!-- En esta sección podéis enumerar y explicar cada columna del corpus. Para cada columna que sea de tipo "categoría" podéis indicar el porcentaje de ejemplos. -->
113
 
114
- Por cada entrada o documento en la fuente de información organizarla en un dataset de hugginface de la siguiente forma:
115
-
116
- - **question (raw_text)**: Texto asociado al documento, pregunta, caso clínico u otro tipo de información.
117
- - **answer (topic)**: (Texto asociado al tratamiento médico (healthcare_treatment), diagnóstico (healthcare_diagnosis),
118
- tópico de salud (topic), respuesta de una pregunta (answer), other, o estar vacío p.ej en el texto abierto)
119
- - **speciality**: (Especialidad médica a la que se relaciona el raw_text p.ej: cardiología, cirugía, otros)
120
- - **raw_text_type**: (Puede ser clinic_case, open_text, question o vacio)
121
- - **topic_type**: (Puede ser medical_topic, medical_diagnostic,answer,natural_medicine_topic, other, o vacio)
122
- - **source**: Identificador de la fuente asociada al documento que aparece en el README y descripción del dataset.
123
- - **country**: Identificador del país de procedencia de la fuente (p.ej.; ch, es) usando el estándar ISO 3166-1 alfa-2 (Códigos de país de dos letras.).
124
- - **document_id**: Identificador del documento en el dataset de procedencia, este valor puede estar vacio en caso que no se conozca
125
  <!-- - **idioma**: (Variedad geográfica) código ISO del idioma -->
126
  <!--- **registro** (Variedad funcional): Siempre es `medio`. -->
127
  <!-- - **periodo** (Variedad histórica): Siempre es `actual`. -->
@@ -129,36 +129,35 @@ Por cada entrada o documento en la fuente de información organizarla en un data
129
  <!-- - **tarea**: `pregunta` | `resumen` | `open_text` | `clinic_case`. -->
130
  <!-- - **país_origen**: País de origen de los datos. -->
131
 
132
- Al inicio de este proceso de construcción se debe actualizar en la tabla de la sección [Source Data](#source_data) la
133
- descripción de la fuente de información con los siguientes datos:
134
-
135
- - **Id**: Este será un número para que la fuente de información pueda ser referenciada en cada entrada del conjunto de datos.
136
- - **Nombre**: Nombre de la fuente de donde procede.
137
- - **Tokens**: Cantidad de tokens que contiene.
138
- - **Memoria**: Tamaño en memoria del dataset generado para hugginface
139
- - **Licencia**: En este caso si es solo para investigación o si posee otra licencia como MIT,
140
- Apache 2 u otras
141
- - **Dirección**: URL de donde se puede descargar o consultar la información.
142
- - **País**: País de procedencia de la información usando el [standar ISO 3166-1](https://es.wikipedia.org/wiki/ISO_3166-1) código alfa-2: Código ISO de 2 letras asignado a ese país o territorio.
143
-
144
 
145
  ## Dataset Creation
146
 
147
  ### Curation Rationale
148
 
149
  <!-- Motivation for the creation of this dataset. -->
150
- Más de 600 millones de personas hablantes del idioma español necesitan recursos, como los LLMs, para la obtención de información médica de forma libre
151
- y segura, cumpliendo con los objetivo del milenio: Salud y Bienestar, Educación y Calidad, Fin de la Pobreza propuestos por la ONU.
152
- Existen pocos recursos o conjuntos de datos del dominio médico para el entrenamiento o autoajuste de un LLM en idioma español.
153
 
154
- Para entrenar un autoajustar un LLM en el dominio de la medicina y los cuidados de salud se necesitan gran cantidad de datos de este contexto.
155
- Para crear un conjunto de datos en el dominio médico es necesaria alguna certificación por especialistas en la construcción del corpus.
156
 
157
  ### Source Data
158
 
159
  <!-- This section describes the source data (e.g. news text and headlines, social media posts, translated sentences, ...). -->
160
 
161
- | Id | Nombre | Tokens | Memoria | Licencia | Dirección | País |
162
  | --- | ----------- | ----------- | ----------- | ----------- | ----------- | ----------- |
163
  | 1 | Cantemist corpus: gold standard of oncology clinical cases annotated with CIE-O 3 terminology | 349287 | 9157 kB | [CC Attribution 4.0 International](https://creativecommons.org/licenses/by/4.0/) | https://huggingface.co/datasets/bigbio/cantemist/viewer/cantemist_bigbio_kb | es |
164
  | 2 | MedlinePlus Spanish (National Library of Medicine, NLM) | 7757337 | 35 MB | | https://medlineplus.gov/spanish/ | es |
@@ -197,16 +196,15 @@ Para crear un conjunto de datos en el dominio médico es necesaria alguna certif
197
 
198
 
199
  **Sugerencias:**
200
- - En el caso de [BioMistral/BioInstructQA](https://huggingface.co/datasets/BioMistral/BioInstructQA) se utilizó la información en idioma español. Para más información consultar el artículo [BioMistral: A Collection of Open-Source Pretrained Large Language Models for Medical Domains](https://arxiv.org/pdf/2402.10373.pdf?trk=public_post_comment-text).
201
- - Para [Cantemist](https://huggingface.co/datasets/bigbio/cantemist/viewer/cantemist_bigbio_kb) se hizo una búsqueda del código asociado a la patología y se estableció como tópico.
202
- - En [CARES](https://huggingface.co/datasets/chizhikchi/CARES) se busco el tipo asociado en la tabla de códigos establecido.
203
 
204
  #### Who are the source data producers?
205
 
206
  <!-- This section describes the people or systems who originally created the data. It should also include self-reported demographic or identity information for the source data creators if this information is available. -->
207
 
208
- Diferentes eventos, competencias de PLN o la construcción de los conjuntos de datos para LLM como BioMistral. Ver [tabla en la sección Source Data](#Source)
209
-
210
 
211
  #### Annotation process
212
 
@@ -214,18 +212,18 @@ Diferentes eventos, competencias de PLN o la construcción de los conjuntos de d
214
 
215
  <!-- Enlazar aquí el notebook utilizado para crear el espacio de anotación de Argilla y la guía de anotación. -->
216
 
217
- El proceso de anotacion fue automático, realizando la converción de las fuentes de datos a los atributos del nuevo conjunto de datos.
218
 
219
  #### Who are the annotators?
220
 
221
  <!-- This section describes the people or systems who created the annotations. -->
222
- Ver la sección [Team](#Team)
223
 
224
  #### Personal and Sensitive Information
225
 
226
  <!-- State whether the dataset contains data that might be considered personal, sensitive, or private (e.g., data that reveals addresses, uniquely identifiable names or aliases, racial or ethnic origins, sexual orientations, religious beliefs, political opinions, financial or health data, etc.). If efforts were made to anonymize the data, describe the anonymization process. -->
227
 
228
- En el proceso de construcción se tuvo en cuenta que no se incluyeran datos sensibles de los usuarios en ninguno de los casos (p.ej.; casos clinicos).
229
 
230
  <!-- In the construction process, it was taken into account that sensitive user data was not included in any of the cases (e.g., clinical cases). -->
231
 
@@ -236,11 +234,10 @@ En el proceso de construcción se tuvo en cuenta que no se incluyeran datos sens
236
 
237
  <!-- Aquí podéis mencionar los posibles sesgos heredados según el origen de los datos y de las personas que lo han anotado, hablar del balance de las categorías representadas, los esfuerzos que habéis hecho para intentar mitigar sesgos y riesgos. -->
238
 
239
- Se sugiere tener en cuenta el alcance de las licencia de cada una de las fuentes (e.d., revisar el campo source y Licencia de la tabla anterior).
240
-
241
- En el caso de necesitar filtrar por fuente de datos u otro criterio usted puede auxiliarse de las propiedades de la estructura de datos `Dataset` del marco de trabajo
242
- Hugginface. En el siguiente ejemplo de código se obtienen del conjunto de datos las entradas que tienen un tipo de tópico sobre diagnóstico medico o un tópico médico:
243
 
 
 
244
 
245
  ```
246
 
@@ -259,14 +256,14 @@ Example:
259
 
260
  Users should be made aware of the risks, biases and limitations of the dataset. More information needed for further recommendations. -->
261
 
262
- Los usuarios deben ser conscientes de los riesgos, sesgos y limitaciones del conjunto de datos.
263
 
264
- Para el autojuste de un LLM se sugiere tener en cuenta las filas donde el tipo del tópico (ed., campo topic_type) tenga valores: `medical_topic`,
265
- `medical_diagnostic`, `answer`, `natural_medicine_topic`. Porque indica que este campo no está vacio y tiene valor para la creación de intrucciones de la
266
- forma pregunta y respuestas.
267
 
268
- Para el prentrenamiento de un LLM se sugiere tener en cuenta cuando el campo `raw_text_type` sea igual a `open_text`. Esto indica que el texto
269
- presente no forma parte de un formato pregunta/respuesta pero tiene un valor importante para el prentrenamiento de un LLM.
270
 
271
  <!--
272
  In case of using this dataset for the LLM training or finetuning for natural language generating with a production
 
59
 
60
  <!-- Si queréis incluir una versión de la Dataset Card en español, enlazarla aquí al principio (e.g. `README_es.md`).-->
61
 
62
+ This dataset groups and organizes several datasets present in hugginface (e.g.: PlanTL-GOB-ES/cantemist-ner, PlanTL-GOB-ES/pharmaconer)
63
+ and other public resources created by researchers with different formats (e.g.; MedLexSp )
64
+ to allow it to be a source of knowledge of large language models in Spanish for the medical domain.
65
 
66
  <!--
67
  This dataset groups and organizes several datasets present in hugginface (e.g.: PlanTL-GOB-ES/cantemist-ner, PlanTL-GOB-ES/pharmaconer)
 
92
  ## Uses
93
 
94
  <!-- Address questions around how the dataset is intended to be used. -->
95
+ The use of this dataset is suggested to achieve self-tuning and pre-training of LLM for the medical domain with information in Spanish.
96
 
97
  ### Direct Use
98
 
 
102
  ### Out-of-Scope Use
103
 
104
  <!-- This section addresses misuse, malicious use, and uses that the dataset will not work well for. -->
105
+ The creators of the dataset are not responsible for harmful results that the models may generate when trained with this information.
106
+ A rigorous evaluation process with specialists of the results generated by trained LLM models is suggested.
107
 
108
  ## Dataset Structure
109
 
 
111
 
112
  <!-- En esta sección podéis enumerar y explicar cada columna del corpus. Para cada columna que sea de tipo "categoría" podéis indicar el porcentaje de ejemplos. -->
113
 
114
+ For each entry or document in the information source, organize it in a Hugginface dataset as follows:
115
+
116
+ - **question (raw_text)**: Text associated with the document, question, clinical case or other type of information.
117
+ - **answer (topic)**: (Text associated with medical treatment (healthcare_treatment), diagnosis (healthcare_diagnosis),
118
+ health topic (topic), answer to a question (answer), other, or be empty e.g. in the open text)
119
+ - **speciality**: (Medical specialty to which the raw_text relates, e.g. cardiology, surgery, others)
120
+ - **raw_text_type**: (Can be clinic_case, open_text, question or empty)
121
+ - **topic_type**: (It can be medical topic, medical diagnosis, answer, natural medicine topic, other, or empty)
122
+ - **source**: Identifier of the source associated with the document that appears in the README and description of the dataset.
123
+ - **country**: Identifier of the country of origin of the source (e.g.; ch, es) using the ISO 3166-1 alpha-2 standard (Two-letter country codes).
124
+ - **document_id**: Document identifier in the source dataset, this value can be empty in case it is not known.
125
  <!-- - **idioma**: (Variedad geográfica) código ISO del idioma -->
126
  <!--- **registro** (Variedad funcional): Siempre es `medio`. -->
127
  <!-- - **periodo** (Variedad histórica): Siempre es `actual`. -->
 
129
  <!-- - **tarea**: `pregunta` | `resumen` | `open_text` | `clinic_case`. -->
130
  <!-- - **país_origen**: País de origen de los datos. -->
131
 
132
+ At the beginning of this construction process, the table in the [Source Data](#source_data) section must be updated.
133
+ description of the source of information with the following data:
134
+
135
+ - **Id**: This will be a number so that the source of information can be referenced in each entry of the data set.
136
+ - **Name**: Name of the source from which it comes.
137
+ - **Tokens**: Number of tokens it contains.
138
+ - **Memory**: Memory size of the dataset generated for huggingface
139
+ - **Licencia**: In this case, if it is only for research or if you have another license such as MIT,
140
+ Apache 2 or others
141
+ - **Address**: URL from where the information can be downloaded or consulted.
142
+ - **Country**: Information source country of the using the [ISO 3166-1 standard](https://es.wikipedia.org/wiki/ISO_3166-1) alpha-2 code: 2-letter ISO code assigned to that country or territory.
 
143
 
144
  ## Dataset Creation
145
 
146
  ### Curation Rationale
147
 
148
  <!-- Motivation for the creation of this dataset. -->
149
+ More than 600 million Spanish speakers need resources, such as LLMs, to obtain medical information freely
150
+ and safe, complying with the millennium objectives: Health and Wellbeing, Education and Quality, End of Poverty proposed by the UN.
151
+ There are few resources or data sets from the medical domain for training or self-tuning for an LLM in the Spanish language.
152
 
153
+ To train an LLM autotuner in the domain of medicine and healthcare, a large amount of data from this context is needed.
154
+ To create a data set in the medical domain, some certification by specialists in corpus construction is necessary.
155
 
156
  ### Source Data
157
 
158
  <!-- This section describes the source data (e.g. news text and headlines, social media posts, translated sentences, ...). -->
159
 
160
+ | Id | Name | Tokens | Memory | Licencia | Address | Country |
161
  | --- | ----------- | ----------- | ----------- | ----------- | ----------- | ----------- |
162
  | 1 | Cantemist corpus: gold standard of oncology clinical cases annotated with CIE-O 3 terminology | 349287 | 9157 kB | [CC Attribution 4.0 International](https://creativecommons.org/licenses/by/4.0/) | https://huggingface.co/datasets/bigbio/cantemist/viewer/cantemist_bigbio_kb | es |
163
  | 2 | MedlinePlus Spanish (National Library of Medicine, NLM) | 7757337 | 35 MB | | https://medlineplus.gov/spanish/ | es |
 
196
 
197
 
198
  **Sugerencias:**
199
+ - In [BioMistral/BioInstructQA](https://huggingface.co/datasets/BioMistral/BioInstructQA) the information was used in Spanish. For more information consult the article [BioMistral: A Collection of Open-Source Pretrained Large Language Models for Medical Domains](https://arxiv.org/pdf/2402.10373.pdf?trk=public_post_comment-text).
200
+ - In [Cantemist](https://huggingface.co/datasets/bigbio/cantemist/viewer/cantemist_bigbio_kb) a search was made for the code associated with the pathology and it was established as a topic.
201
+ - In [CARES](https://huggingface.co/datasets/chizhikchi/CARES) the associated type was searched in the established code table.
202
 
203
  #### Who are the source data producers?
204
 
205
  <!-- This section describes the people or systems who originally created the data. It should also include self-reported demographic or identity information for the source data creators if this information is available. -->
206
 
207
+ Different events, NLP competitions or the construction of data sets for LLM such as BioMistral. See [table in Source Data section](#Source)
 
208
 
209
  #### Annotation process
210
 
 
212
 
213
  <!-- Enlazar aquí el notebook utilizado para crear el espacio de anotación de Argilla y la guía de anotación. -->
214
 
215
+ The annotation process was automatic, converting the data sources to the attributes of the new data set.
216
 
217
  #### Who are the annotators?
218
 
219
  <!-- This section describes the people or systems who created the annotations. -->
220
+ See the section [Team](#Team)
221
 
222
  #### Personal and Sensitive Information
223
 
224
  <!-- State whether the dataset contains data that might be considered personal, sensitive, or private (e.g., data that reveals addresses, uniquely identifiable names or aliases, racial or ethnic origins, sexual orientations, religious beliefs, political opinions, financial or health data, etc.). If efforts were made to anonymize the data, describe the anonymization process. -->
225
 
226
+ In the construction process, it was taken into account that sensitive user data was not included in any of the cases (e.g., clinical cases).
227
 
228
  <!-- In the construction process, it was taken into account that sensitive user data was not included in any of the cases (e.g., clinical cases). -->
229
 
 
234
 
235
  <!-- Aquí podéis mencionar los posibles sesgos heredados según el origen de los datos y de las personas que lo han anotado, hablar del balance de las categorías representadas, los esfuerzos que habéis hecho para intentar mitigar sesgos y riesgos. -->
236
 
237
+ It is suggested to take into account the scope of the license of each of the sources (e.g., review the source and License field in the previous table).
 
 
 
238
 
239
+ If you need to filter by data source or other criteria, you can use the properties of the `Dataset` data structure of the framework.
240
+ Hugginface. In the following code example, the entries that have a topic type about medical diagnosis or a medical topic are obtained from the data set:
241
 
242
  ```
243
 
 
256
 
257
  Users should be made aware of the risks, biases and limitations of the dataset. More information needed for further recommendations. -->
258
 
259
+ Personnel using this dataset must be aware of the risks, biases and limitations of the dataset.
260
 
261
+ For the autotuning of an LLM, it is suggested to take into account the rows where the topic type (ed., topic_type field) has values: `medical_topic`,
262
+ `medical_diagnostic`, `answer`, `natural_medicine_topic`. Because it indicates that this field is not empty and has value for the creation of instructions of the
263
+ question and answer form.
264
 
265
+ For LLM pre-training, it is suggested to take into account when the `raw_text_type` field is equal to `open_text`. This indicates that the text
266
+ is not part of a question/answer format but has important value for LLM pre-training.
267
 
268
  <!--
269
  In case of using this dataset for the LLM training or finetuning for natural language generating with a production