Saltar al contenido
    Ertas
    MedErtas

    MedErtas

    Modelos abiertos para IA médica.

    MedErtas es una familia de modelos de lenguaje de pesos abiertos para empresas de IA médica, diseñados para una inferencia rápida y de bajo costo.

    ~8x

    Más rápido que la línea base.

    +25%

    Más precisión que MedGemma 27B.

    Desde $0.10

    Por millón de tokens de entrada.

    Presentamos
    MedErtas v0.1.

    31B

    MedErtas-v0.1-31B-LS

    Basado en Gemma 4 31B

    Ideal para

    • Conocimiento médico y códigos
    • Revisión de respuestas
    • Detección de preguntas sin respuesta
    Velocidad de salida
    189 tokens/s
    Pesos
    24.2 GB, NVFP4
    Modelo borrador
    Perigee
    Licencia
    Apache-2.0
    27B

    MedErtas-v0.1-27B-LS

    Basado en Qwen3.8-27B

    Ideal para

    • Cálculo clínico, con o sin herramientas
    • Historias clínicas largas con respuestas cortas
    Velocidad de salida
    207 tokens/s
    Pesos
    22 GB, NVFP4
    Modelo borrador
    Perigee
    Licencia
    Apache-2.0

    Comparación detallada

    Qué modelo elegir para cada tarea. Precisión en porcentaje por tarea. La columna Lidera compara los dos modelos de MedErtas; MedGemma 27B se incluye como referencia.
    Tarea31B LS27B LSLideraMedGemma 27Breferencia
    Cálculo clínicoMedCalc-Bench 1.263.977.327B LS48.6
    Cálculo con herramientasMedCalc with tools61.176.327B LSN/D
    Conocimiento médicoMedQA93.086.731B LS87.4
    Preguntas sin respuestaMetaMedQA70.465.431B LS36.2
    Códigos médicosMedConceptsQA, promedio de los niveles medio y difícil69.863.231B LS46.4
    Revisión de respuestasMedHallu, promedio de 3 niveles76.772.331B LS64.7
    Historias clínicas largasLongHealth, promedio de 2 tareas90.089.0Cualquiera27B cuesta menos con entradas largas81.5

    Posibles casos de uso*

    Los desarrolladores pueden adaptar MedErtas a flujos de trabajo clínicos y ejecutarlo de forma privada dentro de sistemas de agentes.

    Consulta el uso previsto y las limitaciones

    Mira los modelos
    en acción.

    Los ejemplos grabados se muestran en inglés, el idioma en el que se ejecutaron.

    Cálculo clínico

    App de historia clínica · Nota de ingreso
    Nota de ingreso

    An 87-year-old man was admitted to our hospital for anorexia for several days, high-grade fever from the previous day, and liver dysfunction. Of note, he had a history of hypertension, diabetes mellitus (DM), and angina. Physical examination findings included: clear consciousness; height, 163 cm; weight, 48 kg; blood pressure, 66/40 mmHg; heart rate, 75/min; respiratory rate, 22/min; oxygen saturation of peripheral artery, 96%; and body temperature, 38.1 °C. He had no surface lymphadenopathy. Laboratory findings included: white blood cell (WBC) count, 4.2 × 109/L; hemoglobin, 9.6 g/dL; platelet count, 106 × 109/L; lactate dehydrogenase (LDH), 1662 IU/L; aspartate aminotransferase (AST), 6562 IU/L; alanine aminotransferase (ALT), 1407 IU/L; alkaline phosphatase (ALP), 509 IU/L; γ-glutamyl transpeptidase (γ-GTP), 130 IU/L; total bilirubin, 2.7 mg/dL; prothrombin time (PT), 20.3 s; international normalized ratio (INR), 1.73; blood urea nitrogen (BUN), 34.4 mg/dL; and creatinine, 1.4 mg/dL. Serologic tests were all negative for hepatitis A, B, C, and E. Computed tomography (CT) performed at admission revealed no distinct nodules or masses in the liver, but mild splenomegaly, multiple hypodense splenic lesions, and para-aortic lymph node swelling (Fig. a–c). Antibiotics were administered intravenously for suspected acute cholangitis and sepsis, but the fever persisted. Abdominal ultrasonography (US) and magnetic resonance imaging (MRI) performed 2 days later newly depicted gallbladder wall thickening and ascites (Figs. , ). On the same night, the patient presented with grade II hepatic encephalopathy, with blood tests showing a decrease in platelet count and worsening liver function suggestive of “acute type” ALF with hepatic coma and disseminated intravascular coagulation (DIC). While consideration was being given to treating acute liver failure, the patient suddenly developed ventricular fibrillation and died despite immediate resuscitation measures on day 3 of hospitalization. Given the rapid clinical course, an autopsy was performed in this patient with suspected ALF likely due to drug-induced liver injury, autoimmune hepatitis, hepatitis of indeterminate etiology, circulatory disturbance, metabolic disease, or malignant cells including malignant lymphoma. This led to the patient being diagnosed with malignant, diffuse large B-cell lymphoma, characterized by multiple lymph-node, as well as multi-organ (i.e., spleen, liver, lung, and prostate) involvement: (1) general swelling of lymph nodes in the para-abdominal aorta (Fig. a), peripancreatic, peribronchial, pulmonary hilar, subclavicular regions, as well as in the pelvis, especially around the iliac arteries, measuring up to 60 mm in diameter; (2) presence of multiple whitish splenic nodules (Fig. b) with the entire spleen shown to be microscopically infiltrated by tumor cells (Fig. a); (3) irregular geographical morphology of the liver (Fig. c) with the lymphoma cells mainly infiltrating the portal areas (Fig. b); (4) involvement of the right lobe of the lung; (5) involvement of both lobes of the prostate consequent to outer connective tissue involvement (Fig. c). Microscopically, the tumor cells in the lymph nodes were shown to have medium-sized irregular nuclei and small nucleoli with a high nuclear cytoplasmic ratio, thus showing diffuse proliferation characterized by the presence of large numbers of small T-cells (Fig. ). Immunohistochemistry revealed the tumor cells to be positive for CD20 (Fig. a), CD79a (Fig. b), bcl-2 (Fig. c), and MUM1 (Fig. d), but negative for CD3 (Fig. e), CD10, bcl-6, and cyclin D1, with a Ki-67 labeling index (Ki-67 LI) of 16% (Fig. f). The causes of death were thus determined as (1) DIC and (2) ALF associated with diffuse large B-cell lymphoma.

    Tomado de la nota
    • 87-year-old
    • weight, 48 kg
    • height, 163 cm
    • creatinine, 1.4 mg/dL
    Solicitud

    What is the patient's Creatinine Clearance using the Cockroft-Gault Equation in terms of mL/min?

    Asistente de cálculoMedErtas-v0.1-27B-LS
    1. I'll calculate the Creatinine Clearance using the Cockcroft-Gault equation.

      Step 1: Determine BMI and weight category

      Let me extract the values from the patient note: - Age: 87 years - Sex: Male - Height: 163 cm - Weight: 48 kg - Creatinine: 1.4 mg/dL

    2. Llamó a calculator48 / (1.63 * 1.63)= 18.0662
    3. BMI ≈ 18.07, which is underweight (BMI < 18.5).

      Per the instructions: "If the patient is underweight, please set their adjusted body weight to their actual body weight."

      So adjusted body weight = 48 kg (actual weight).

      Step 2: Apply Cockcroft-Gault equation for males:

      CrCl = [(140 − age) × weight] / (72 × serum creatinine)

    4. Llamó a calculator((140 - 87) * 48) / (72 * 1.4)= 25.2381
    5. Calculation: - (140 − 87) = 53 - 53 × 48 = 2544 - 72 × 1.4 = 100.8 - 2544 / 100.8 ≈ 25.24 mL/min

    Aclaramiento de creatinina25.24mL/min
    2 llamadas a herramientas360 tokens de salida
    Prompt original y ejecución completa grabada
    Entrada del modelo
    You are a helpful assistant. Your task is to read a patient note and compute a medical value based on the following question. You have access to a calculator tool and a python code execution tool. Use them to do any math based on values in the patient note. If using the python tool, make sure the code returns or prints all of its outputs. If there are any errors with compiling or running your script, you may need to re-write your code to obtain the output. Note that all of the necessary information is provided in the patient note and you should not need to prompt the user for any information. When you are finished with all computations, please write your final answer value without any units, using the following formats inside the indicated answer tag: - Decimal Answer Format: 17.29 - Score-Based Answer Format: 5 - Estimated Date Answer Format: 5/21/2021 - Estimated Age Answer Format: (4 weeks, 3 days) Patient Note: An 87-year-old man was admitted to our hospital for anorexia for several days, high-grade fever from the previous day, and liver dysfunction. Of note, he had a history of hypertension, diabetes mellitus (DM), and angina. Physical examination findings included: clear consciousness; height, 163 cm; weight, 48 kg; blood pressure, 66/40 mmHg; heart rate, 75/min; respiratory rate, 22/min; oxygen saturation of peripheral artery, 96%; and body temperature, 38.1 °C. He had no surface lymphadenopathy. Laboratory findings included: white blood cell (WBC) count, 4.2 × 109/L; hemoglobin, 9.6 g/dL; platelet count, 106 × 109/L; lactate dehydrogenase (LDH), 1662 IU/L; aspartate aminotransferase (AST), 6562 IU/L; alanine aminotransferase (ALT), 1407 IU/L; alkaline phosphatase (ALP), 509 IU/L; γ-glutamyl transpeptidase (γ-GTP), 130 IU/L; total bilirubin, 2.7 mg/dL; prothrombin time (PT), 20.3 s; international normalized ratio (INR), 1.73; blood urea nitrogen (BUN), 34.4 mg/dL; and creatinine, 1.4 mg/dL. Serologic tests were all negative for hepatitis A, B, C, and E. Computed tomography (CT) performed at admission revealed no distinct nodules or masses in the liver, but mild splenomegaly, multiple hypodense splenic lesions, and para-aortic lymph node swelling (Fig. a–c). Antibiotics were administered intravenously for suspected acute cholangitis and sepsis, but the fever persisted. Abdominal ultrasonography (US) and magnetic resonance imaging (MRI) performed 2 days later newly depicted gallbladder wall thickening and ascites (Figs. , ). On the same night, the patient presented with grade II hepatic encephalopathy, with blood tests showing a decrease in platelet count and worsening liver function suggestive of “acute type” ALF with hepatic coma and disseminated intravascular coagulation (DIC). While consideration was being given to treating acute liver failure, the patient suddenly developed ventricular fibrillation and died despite immediate resuscitation measures on day 3 of hospitalization. Given the rapid clinical course, an autopsy was performed in this patient with suspected ALF likely due to drug-induced liver injury, autoimmune hepatitis, hepatitis of indeterminate etiology, circulatory disturbance, metabolic disease, or malignant cells including malignant lymphoma. This led to the patient being diagnosed with malignant, diffuse large B-cell lymphoma, characterized by multiple lymph-node, as well as multi-organ (i.e., spleen, liver, lung, and prostate) involvement: (1) general swelling of lymph nodes in the para-abdominal aorta (Fig. a), peripancreatic, peribronchial, pulmonary hilar, subclavicular regions, as well as in the pelvis, especially around the iliac arteries, measuring up to 60 mm in diameter; (2) presence of multiple whitish splenic nodules (Fig. b) with the entire spleen shown to be microscopically infiltrated by tumor cells (Fig. a); (3) irregular geographical morphology of the liver (Fig. c) with the lymphoma cells mainly infiltrating the portal areas (Fig. b); (4) involvement of the right lobe of the lung; (5) involvement of both lobes of the prostate consequent to outer connective tissue involvement (Fig. c). Microscopically, the tumor cells in the lymph nodes were shown to have medium-sized irregular nuclei and small nucleoli with a high nuclear cytoplasmic ratio, thus showing diffuse proliferation characterized by the presence of large numbers of small T-cells (Fig. ). Immunohistochemistry revealed the tumor cells to be positive for CD20 (Fig. a), CD79a (Fig. b), bcl-2 (Fig. c), and MUM1 (Fig. d), but negative for CD3 (Fig. e), CD10, bcl-6, and cyclin D1, with a Ki-67 labeling index (Ki-67 LI) of 16% (Fig. f). The causes of death were thus determined as (1) DIC and (2) ALF associated with diffuse large B-cell lymphoma. Question: What is the patient's Creatinine Clearance using the Cockroft-Gault Equation in terms of mL/min? You should use the patient's adjusted body weight in kg instead of the patient's actual body weight if the patient is overweight or obese based on their BMI. If the patient's BMI's normal, set their adjusted body weight to the minimum of the ideal body and actual weight. If the patient is underweight, please set their adjusted body weight to their actual body weight. You should use the patient's medical values and health status when they were first admitted to the hospital prior to any treatment.
    Salida grabada del modelo · MedErtas-v0.1-27B-LS
    I'll calculate the Creatinine Clearance using the Cockcroft-Gault equation. **Step 1: Determine BMI and weight category** Let me extract the values from the patient note: - Age: 87 years - Sex: Male - Height: 163 cm - Weight: 48 kg - Creatinine: 1.4 mg/dL
    {"id": "chatcmpl-tool-b4cd295e4a01108c", "name": "calculator", "arguments": "{\"expression\": \"48 / (1.63 * 1.63)\"}"}
    18.06616733787497
    BMI ≈ 18.07, which is **underweight** (BMI < 18.5). Per the instructions: "If the patient is underweight, please set their adjusted body weight to their actual body weight." So adjusted body weight = 48 kg (actual weight). **Step 2: Apply Cockcroft-Gault equation for males:** CrCl = [(140 − age) × weight] / (72 × serum creatinine)
    {"id": "chatcmpl-tool-a00ee6f6473d87f6", "name": "calculator", "arguments": "{\"expression\": \"((140 - 87) * 48) / (72 * 1.4)\"}"}
    25.238095238095237
    **Calculation:** - (140 − 87) = 53 - 53 × 48 = 2544 - 72 × 1.4 = 100.8 - 2544 / 100.8 ≈ **25.24 mL/min** <answer>25.24</answer>
    Respuesta de referencia25.2381
    Solicitud a la API para este ejemplo

    Prompt grabado, con la temperatura del protocolo de evaluación. Confirma el identificador del modelo y las opciones de solicitud que admite tu despliegue de la API. Tu aplicación debe ejecutar el bucle de la herramienta de cálculo.

    JSON
    {
      "model": "MedErtas-v0.1-27B-LS",
      "messages": [
        {
          "role": "system",
          "content": "Please reason step by step, use tools as needed, then give your final answer within <answer>...</answer> XML tags."
        },
        {
          "role": "user",
          "content": "You are a helpful assistant. Your task is to read a patient note and compute a medical value based on the following question.\nYou have access to a calculator tool and a python code execution tool. Use them to do any math based on values in the patient note. If using the python tool, make sure the code returns or prints all of its outputs. If there are any errors with compiling or running your script, you may need to re-write your code to obtain the output.\nNote that all of the necessary information is provided in the patient note and you should not need to prompt the user for any information.\nWhen you are finished with all computations, please write your final answer value without any units, using the following formats inside the indicated answer tag:\n\n- Decimal Answer Format: 17.29\n- Score-Based Answer Format: 5\n- Estimated Date Answer Format: 5/21/2021\n- Estimated Age Answer Format: (4 weeks, 3 days)\n\nPatient Note:\n\nAn 87-year-old man was admitted to our hospital for anorexia for several days, high-grade fever from the previous day, and liver dysfunction. Of note, he had a history of hypertension, diabetes mellitus (DM), and angina. Physical examination findings included: clear consciousness; height, 163 cm; weight, 48 kg; blood pressure, 66/40 mmHg; heart rate, 75/min; respiratory rate, 22/min; oxygen saturation of peripheral artery, 96%; and body temperature, 38.1 °C. He had no surface lymphadenopathy. Laboratory findings included: white blood cell (WBC) count, 4.2 × 109/L; hemoglobin, 9.6 g/dL; platelet count, 106 × 109/L; lactate dehydrogenase (LDH), 1662 IU/L; aspartate aminotransferase (AST), 6562 IU/L; alanine aminotransferase (ALT), 1407 IU/L; alkaline phosphatase (ALP), 509 IU/L; γ-glutamyl transpeptidase (γ-GTP), 130 IU/L; total bilirubin, 2.7 mg/dL; prothrombin time (PT), 20.3 s; international normalized ratio (INR), 1.73; blood urea nitrogen (BUN), 34.4 mg/dL; and creatinine, 1.4 mg/dL. Serologic tests were all negative for hepatitis A, B, C, and E. Computed tomography (CT) performed at admission revealed no distinct nodules or masses in the liver, but mild splenomegaly, multiple hypodense splenic lesions, and para-aortic lymph node swelling (Fig. a–c). Antibiotics were administered intravenously for suspected acute cholangitis and sepsis, but the fever persisted. Abdominal ultrasonography (US) and magnetic resonance imaging (MRI) performed 2 days later newly depicted gallbladder wall thickening and ascites (Figs. , ). On the same night, the patient presented with grade II hepatic encephalopathy, with blood tests showing a decrease in platelet count and worsening liver function suggestive of “acute type” ALF with hepatic coma and disseminated intravascular coagulation (DIC). While consideration was being given to treating acute liver failure, the patient suddenly developed ventricular fibrillation and died despite immediate resuscitation measures on day 3 of hospitalization.\nGiven the rapid clinical course, an autopsy was performed in this patient with suspected ALF likely due to drug-induced liver injury, autoimmune hepatitis, hepatitis of indeterminate etiology, circulatory disturbance, metabolic disease, or malignant cells including malignant lymphoma. This led to the patient being diagnosed with malignant, diffuse large B-cell lymphoma, characterized by multiple lymph-node, as well as multi-organ (i.e., spleen, liver, lung, and prostate) involvement: (1) general swelling of lymph nodes in the para-abdominal aorta (Fig. a), peripancreatic, peribronchial, pulmonary hilar, subclavicular regions, as well as in the pelvis, especially around the iliac arteries, measuring up to 60 mm in diameter; (2) presence of multiple whitish splenic nodules (Fig. b) with the entire spleen shown to be microscopically infiltrated by tumor cells (Fig. a); (3) irregular geographical morphology of the liver (Fig. c) with the lymphoma cells mainly infiltrating the portal areas (Fig. b); (4) involvement of the right lobe of the lung; (5) involvement of both lobes of the prostate consequent to outer connective tissue involvement (Fig. c). Microscopically, the tumor cells in the lymph nodes were shown to have medium-sized irregular nuclei and small nucleoli with a high nuclear cytoplasmic ratio, thus showing diffuse proliferation characterized by the presence of large numbers of small T-cells (Fig. ). Immunohistochemistry revealed the tumor cells to be positive for CD20 (Fig. a), CD79a (Fig. b), bcl-2 (Fig. c), and MUM1 (Fig. d), but negative for CD3 (Fig. e), CD10, bcl-6, and cyclin D1, with a Ki-67 labeling index (Ki-67 LI) of 16% (Fig. f). The causes of death were thus determined as (1) DIC and (2) ALF associated with diffuse large B-cell lymphoma.\n\nQuestion:\n\nWhat is the patient's Creatinine Clearance using the Cockroft-Gault Equation in terms of mL/min? You should use the patient's adjusted body weight in kg instead of the patient's actual body weight if the patient is overweight or obese based on their BMI. If the patient's BMI's normal, set their adjusted body weight to the minimum of the ideal body and actual weight. If the patient is underweight, please set their adjusted body weight to their actual body weight. You should use the patient's medical values and health status when they were first admitted to the hospital prior to any treatment."
        }
      ],
      "temperature": 1,
      "chat_template_kwargs": {
        "enable_thinking": false
      },
      "tools": [
        {
          "type": "function",
          "function": {
            "name": "calculator",
            "description": "Evaluate a mathematical expression safely.\n\nSupports basic arithmetic (+, -, *, /, %, **), parentheses, and common math functions. Use ^ or ** for exponentiation.\n\nAvailable functions: sqrt, sin, cos, tan, log, log10, exp, abs, round\nAvailable constants: pi, e",
            "parameters": {
              "properties": {
                "expression": {
                  "description": "A mathematical expression to evaluate.",
                  "title": "Expression",
                  "type": "string"
                }
              },
              "required": [
                "expression"
              ],
              "title": "calculator_args",
              "type": "object",
              "additionalProperties": false
            },
            "strict": null
          }
        },
        {
          "type": "function",
          "function": {
            "name": "code_interpreter",
            "description": "Execute Python code in a sandboxed REPL environment. Variables persist across calls.\n\nThis is a restricted environment - file I/O and network access are disabled.\nAllowed imports: math, numpy, scipy. Pre-loaded: numpy (as np), math, scipy. Use print() to display results.",
            "parameters": {
              "properties": {
                "code": {
                  "description": "A block of Python code to execute.",
                  "title": "Code",
                  "type": "string"
                }
              },
              "required": [
                "code"
              ],
              "title": "code_interpreter_args",
              "type": "object",
              "additionalProperties": false
            },
            "strict": null
          }
        }
      ]
    }

    La misma respuesta,
    rápida como un rayo.

    App de historia clínica · Asistente de cálculo

    El mismo prompt llega a ambos modelos. A precisión completa y sin modelo borrador, Qwen3.8-27B tarda 5.54 segundos en dar la respuesta de 144 tokens, a 26 tokens/s. MedErtas-v0.1-27B-LS con el modelo borrador Perigee tarda 0.70 segundos, a 207 tokens/s.

    Qwen3.8-27BQwen3.8-27B26 tokens/s
    MedErtas-27B-LSMedErtas-27B-LS207 tokens/s

    De dónde viene
    la velocidad.

    Velocidad de salida en tokens por segundo tras cada paso, para cada modelo de MedErtas v0.1
    MedErtas v0.1 31B LSMedErtas v0.1 27B LS
    Línea base22.5 tokens/s26 tokens/s
    Pesos calibrados para medicina51 tokens/s2.3x60 tokens/s2.3x
    Modelo borrador especulativo168 tokens/s3.3x188 tokens/s3.1x
    Modelo borrador médico Perigee189 tokens/s+13%8.4x la línea base207 tokens/s+10%8x la línea base
    Consulta el desglose completo de velocidad en el whitepaper

    Precios de inferencia
    sin servidor

    Por 1M de tokensBetaBetaLa oferta de producción llegará pronto. Todavía no se recomienda para un uso intensivo en producción.
    Precios de inferencia sin servidor por 1M de tokens
    ModeloEntradaLectura de cachéSalida
    MedErtas-v0.1-31B-LS$0.20$0.05$0.60
    MedErtas-v0.1-27B-LS$0.10$0.05$1.00

    Preguntas frecuentes sobre MedErtas

    ¿Qué es MedErtas?

    MedErtas es una familia de modelos de lenguaje médicos de pesos abiertos de Ertas AI. El lanzamiento v0.1 se basa en Gemma 4 31B y Qwen3.8-27B, y combina una cuantización de 4 bits adaptada al dominio con Perigee, nuestros modelos borrador médicos, para una inferencia más rápida.

    ¿Qué modelo de MedErtas debería usar?

    Empieza con MedErtas-v0.1-27B-LS para cálculo clínico y aplicaciones con historias clínicas largas y entradas grandes. Empieza con MedErtas-v0.1-31B-LS para conocimiento médico, conocimiento de códigos y revisión de respuestas. El 27B tiene un precio de entrada más bajo; el 31B, un precio de salida más bajo. Evalúa la precisión, la latencia y el uso total de tokens con tu propia carga de trabajo.

    ¿MedErtas es de código abierto?

    Los pesos del modelo y los modelos borrador médicos Perigee se publican bajo Apache-2.0 en Hugging Face. Puedes descargarlos y ejecutar tu propio despliegue. La API alojada es una forma opcional de acceder a los modelos.

    ¿En qué se diferencia MedErtas de MedGemma?

    En los 11 benchmarks médicos que la clasificación de Medmarks comparte con nuestras ejecuciones, MedErtas 31B y 27B promedian 79.6 y 77.1, 16.0 y 13.5 puntos por encima de los 63.6 de MedGemma 27B en la instantánea de la clasificación del 26 de octubre de 2026. Son resultados de benchmarks, no evidencia de idoneidad clínica. MedErtas v0.1 solo trabaja con texto y está diseñado para una inferencia rápida, con cuantización adaptada al dominio y modelos borrador médicos Perigee; MedGemma también admite la comprensión de imágenes médicas. MedErtas usa Apache-2.0, y MedGemma usa los términos de Health AI Developer Foundations. Ertas ofrece una API beta con precio por token, mientras que el despliegue de Google en Vertex AI usa un endpoint dedicado. Consulta los precios vigentes del proveedor para tu despliegue.

    ¿Puedo ejecutar MedErtas on-premise?

    Sí. Los pesos y los modelos borrador Perigee se pueden descargar. Los pesos usan NVFP4, que funciona en GPUs NVIDIA Blackwell. Los pesos del modelo 31B ocupan 24.2 GB, más los 1.0 GB del modelo borrador Perigee; los del modelo 27B ocupan 22 GB. La memoria en ejecución depende de tu configuración. Hablemos de tu despliegue en tu propia nube, on-premise o air-gapped.

    ¿Es segura la cuantización de 4 bits para modelos médicos?

    Todo modelo médico, incluso a precisión completa, debe integrarse en un sistema cuidadosamente diseñado y revisado antes de usarse en producción. MedErtas busca igualar en capacidad médica a su base de precisión completa, a la vez que hace el modelo más rápido y más económico de servir, y cada lanzamiento publica la comparación con la precisión completa en cada conjunto que evaluamos. Las versiones comprimidas de terceros todavía pueden perder precisión en determinadas tareas médicas. Prueba cualquier modelo frente a la precisión completa en tus propias tareas clínicas antes de confiar en él.

    ¿Qué significa LS en el nombre del modelo?

    LS, de Lightweight Study (estudio ligero), es el perfil de la versión. Una versión LS comprime los pesos con cuantización adaptada al dominio y los combina con un modelo borrador entrenado en el dominio, sin cambiar el entrenamiento original del modelo. Los perfiles futuros abarcarán modelos con entrenamiento posterior en datos del dominio.

    ¿Qué es Perigee?

    Perigee es el nombre de los modelos borrador médicos de MedErtas. Un modelo borrador es un pequeño modelo auxiliar que propone varios tokens por adelantado para la decodificación especulativa; el modelo principal los verifica en una sola pasada y conserva solo los tokens que él mismo produciría, de modo que las respuestas siguen al modelo principal y llegan más rápido. MedErtas-v0.1-31B-LS-Perigee es el modelo borrador asistente de Gemma 4 31B de Google, ajustado con respuestas médicas. MedErtas-v0.1-27B-LS-Perigee es el modelo borrador DSpark de Red Hat para Qwen3.8-27B, con entrenamiento adicional en respuestas médicas. Ambos se publican bajo Apache-2.0 en Hugging Face y se ejecutan con vLLM. El nombre viene del punto de una órbita más cercano al cuerpo que rodea: cuanto más de cerca sigue un modelo borrador al modelo principal, más tokens suyos se aceptan.

    ¿Puedo usar la API de MedErtas en producción?

    Todavía no para un uso intensivo o estable en producción. La API está en beta mientras terminamos la oferta de producción. Los pesos están en Hugging Face si quieres ejecutarlos en tu propio hardware, y si planificas volúmenes de producción, hablemos.

    Lanzamientos de MedErtas

    Uso previsto
    y limitaciones

    MedErtas es un punto de partida para desarrolladores que crean aplicaciones de salud sobre texto médico. Cada caso de uso necesita su propia validación, adaptación o modificación significativa por parte del desarrollador antes de confiar en el modelo.

    Las salidas de los modelos MedErtas no están destinadas a informar directamente el diagnóstico clínico, las decisiones de manejo del paciente, las recomendaciones de tratamiento ni ninguna otra práctica clínica directa. Los resultados de los benchmarks muestran la capacidad de base en las pruebas que nombramos, y las salidas pueden ser inexactas. Trata cada salida como preliminar: necesita verificación independiente y revisión clínica.

    Empieza a construir
    con MedErtas.

    ¿Planificas volúmenes de producción? Hablemos