MedLM

MedLM is a family of foundation models fine-tuned for the healthcare industry. Med-PaLM 2 is one of the text-based models developed by Google Research that powers MedLM, and was the first AI system to reach human expert level on answering US Medical Licensing Examination (USMLE)-style questions. The development of these models has been informed by specific customer needs such as answering medical questions and drafting summaries.

MedLM model card

The MedLM model card outlines the model details, such as MedLM's intended use, data overview, and safety information. Click the following link to download a PDF version of the MedLM model card:

Download the MedLM model card

Use cases

  • Question answering: Provide draft answers to medically-related questions, given as text.
  • Summarization: Draft a shorter version of a document (such as an After Visit Summary or History and Physical Examination note) that incorporates pertinent information from the original text.

For more information on designing text prompts, see Design text prompts.

HTTP request

MedLM-medium (medlm-medium):

POST https://us-central1-aiplatform.googleapis.com/v1/projects/{PROJECT_ID}/locations/us-central1/publishers/google/models/medlm-medium:predict

MedLM-large (medlm-large):

POST https://us-central1-aiplatform.googleapis.com/v1/projects/{PROJECT_ID}/locations/us-central1/publishers/google/models/medlm-large:predict

See the predict method for more information.

Model versions

MedLM provides the following models:

  • MedLM-medium (medlm-medium)
  • MedLM-large (medlm-large)

The following table contains the available stable model versions:

medlm-medium model Release date
medlm-medium December 13, 2023
medlm-large model Release date
medlm-large December 13, 2023

MedLM-medium and MedLM-large have separate endpoints and provide customers with additional flexibility for their use cases. MedLM-medium provides customers with better throughputs and includes more recent data. MedLM-large is the same model from the preview phase. Both models will continue to be refreshed over the product lifecycle. In this page, "MedLM" refers to both models.

For more information, see Model versions and lifecycle.

MedLM safety filters and attributes

Content processed through the MedLM API is assessed against a list of safety attributes, including "harmful categories" and topics that may be considered sensitive. If you are seeing a fallback response, like "I'm not able to help with that, as I'm only a language model," it means that either the prompt or the response is triggering a safety filter.

Safety thresholds

When using Vertex AI Studio, you can use an adjustable safety filter threshold to determine how likely you are to see responses that could be harmful. Model responses are blocked based on the probability that it contains harassment, hate speech, dangerous content, or sexually explicit content. The safety filter setting is located on the right side of the prompt field in Vertex AI Studio. You can choose from three options: block most, block some, and block few.

Console-image

Testing your confidence and severity thresholds

You can test Google's safety filters and define confidence thresholds that are right for your business. By using these thresholds, you can take comprehensive measures to detect content that violates Google's usage policies or terms of service and take appropriate action.

The confidence scores are predictions only, and you shouldn't depend on the scores for reliability or accuracy. Google is not responsible for interpreting or using these scores for business decisions.

To utilize this technology safely and responsibly, it's important to consider other risks specific to your use case, users, and business context in addition to built-in technical safeguards.

We recommend taking the following steps:

  1. Assess your application's security risks.
  2. Consider adjustments to mitigate safety risks.
  3. Perform safety testing appropriate to your use case.
  4. Solicit user feedback and monitor content.

To learn more, see Google's recommendations for Responsible AI.

Request body

{
  "instances": [
    {
      "content": string
    }
  ],
  "parameters": {
    "temperature": number,
    "maxOutputTokens": integer,
    "topK": integer,
    "topP": number
  }
}

Use the following parameters for the medlm-medium and medlm-large models. For more information, see Design text prompts.

Parameter Description Acceptable values

content

 Text input to generate model response. Prompts can include preamble, questions, suggestions, instructions, or examples. Text

temperature

 The temperature is used for sampling during response generation, which occurs when topP and topK are applied. Temperature controls the degree of randomness in token selection. Lower temperatures are good for prompts that require a less open-ended or creative response, while higher temperatures can lead to more diverse or creative results. A temperature of 0 means that the highest probability tokens are always selected. In this case, responses for a given prompt are mostly deterministic, but a small amount of variation is still possible.

If the model returns a response that's too generic, too short, or the model gives a fallback response, try increasing the temperature.

0.0–1.0

Default: 0.2

maxOutputTokens

 Maximum number of tokens that can be generated in the response. A token is approximately four characters. 100 tokens correspond to roughly 60-80 words.

Specify a lower value for shorter responses and a higher value for potentially longer responses.

1–8192 for medlm-medium

1–1024 for medlm-large

topK

 Top-K changes how the model selects tokens for output. A top-K of 1 means the next selected token is the most probable among all tokens in the model's vocabulary (also called greedy decoding), while a top-K of 3 means that the next token is selected from among the three most probable tokens by using temperature.

For each token selection step, the top-K tokens with the highest probabilities are sampled. Then tokens are further filtered based on top-P with the final token selected using temperature sampling.

Specify a lower value for less random responses and a higher value for more random responses.

1–40

Default: 40

topP

 Top-P changes how the model selects tokens for output. Tokens are selected from the most (see top-K) to least probable until the sum of their probabilities equals the top-P value. For example, if tokens A, B, and C have a probability of 0.3, 0.2, and 0.1 and the top-P value is 0.5, then the model will select either A or B as the next token by using temperature and excludes C as a candidate.

Specify a lower value for less random responses and a higher value for more random responses.

0.0–1.0

Default: 0.8

Sample request

When using the MedLM API, it's important to incorporate prompt engineering. For example, we strongly recommend providing appropriate, task-specific instructions at the beginning of each prompt. For more information, see Introduction to prompt design.

REST

Before using any of the request data, make the following replacements:

  • PROJECT_ID: Your project ID.
  • MEDLM_MODEL: The MedLM model, either medlm-medium or medlm-large.

HTTP method and URL: 

POST https://us-central1-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/us-central1/publishers/google/models/MEDLM_MODEL:predict

Request JSON body: 

{
  "instances": [
    {
      "content": "Question: What causes you to get ringworm?"
    }
  ],
  "parameters": {
    "temperature": 0,
    "maxOutputTokens": 256,
    "topK": 40,
    "topP": 0.95
  }
}

To send your request, choose one of these options:

curl

Save the request body in a file named request.json. Run the following command in the terminal to create or overwrite this file in the current directory: 

cat > request.json << 'EOF'
{
  "instances": [
    {
      "content": "Question: What causes you to get ringworm?"
    }
  ],
  "parameters": {
    "temperature": 0,
    "maxOutputTokens": 256,
    "topK": 40,
    "topP": 0.95
  }
}
EOF

Then execute the following command to send your REST request: 

curl -X POST \
     -H "Authorization: Bearer $(gcloud auth print-access-token)" \
     -H "Content-Type: application/json; charset=utf-8" \
     -d @request.json \
     "https://us-central1-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/us-central1/publishers/google/models/MEDLM_MODEL:predict"

PowerShell

Save the request body in a file named request.json. Run the following command in the terminal to create or overwrite this file in the current directory: 

@'
{
  "instances": [
    {
      "content": "Question: What causes you to get ringworm?"
    }
  ],
  "parameters": {
    "temperature": 0,
    "maxOutputTokens": 256,
    "topK": 40,
    "topP": 0.95
  }
}
'@  | Out-File -FilePath request.json -Encoding utf8

Then execute the following command to send your REST request: 

$cred = gcloud auth print-access-token
$headers = @{ "Authorization" = "Bearer $cred" }

Invoke-WebRequest `
    -Method POST `
    -Headers $headers `
    -ContentType: "application/json; charset=utf-8" `
    -InFile request.json `
    -Uri "https://us-central1-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/us-central1/publishers/google/models/MEDLM_MODEL:predict" | Select-Object -Expand Content
 

Response body

{
  "predictions": [
    {
      "content": string,
      "citationMetadata": {
        "citations": [
          {
            "startIndex": integer,
            "endIndex": integer,
            "url": string,
            "title": string,
            "license": string,
            "publicationDate": string
          }
        ]
      },
      "logprobs": {
        "tokenLogProbs": [ float ],
        "tokens": [ string ],
        "topLogProbs": [ { map<string, float> } ]
      },
      "safetyAttributes": {
        "categories": [ string ],
        "blocked": boolean,
        "scores": [ float ],
        "errors": [ int ]
      }
    }
  ],
  "metadata": {
    "tokenMetadata": {
      "input_token_count": {
        "total_tokens": integer,
        "total_billable_characters": integer
      },
      "output_token_count": {
        "total_tokens": integer,
        "total_billable_characters": integer
      }
    }
  }
}
Response element Description
content The result generated from input text.
categories The display names of Safety Attribute categories associated with the generated content. Order matches the Scores.
scores The confidence scores of the each category, higher value means higher confidence.
blocked A flag indicating if the model's input or output was blocked.
errors An error code that identifies why the input or output was blocked. For a list of error codes, see Safety filters and attributes.
startIndex Index in the prediction output where the citation starts (inclusive). Must be greater than or equal to 0 and less than end_index.
endIndex Index in the prediction output where the citation ends (exclusive). Must be greater than start_index and less than len(output).
url URL associated with this citation. If present, this URL links to the web page of the source of this citation. Possible URLs include news websites, GitHub repos, and so forth.
title Title associated with this citation. If present, it refers to the title of the source of this citation. Possible titles include news titles, book titles, and so forth.
license License associated with this recitation. If present, it refers to the license of the source of this citation. Possible licenses include code licenses, such as MIT license.
publicationDate Publication date associated with this citation. If present, it refers to the date at which the source of this citation was published. Possible formats are YYYY, YYYY-MM, YYYY-MM-DD.
input_token_count Number of input tokens. This is the total number of tokens across all prompts, prefixes, and suffixes.
output_token_count Number of output tokens. This is the total number of tokens in content across all predictions.
tokens The sampled tokens.
tokenLogProbs The sampled tokens' log probabilities.
topLogProb The most likely candidate tokens and their log probabilities at each step.
logprobs Results of the `logprobs` parameter. 1-1 mapping to `candidates`.

Sample response

{
  "predictions": [
    {
      "citationMetadata": {
        "citations": []
      },
      "content": "\n\nAnswer and Explanation:\nRingworm is a fungal infection of the skin that is caused by a type of fungus called dermatophyte. Dermatophytes can live on the skin, hair, and nails, and they can be spread from person to person through direct contact or through contact with contaminated objects.\n\nRingworm can cause a variety of symptoms, including:\n\n* A red, itchy rash\n* A raised, circular border\n* Blisters or scales\n* Hair loss\n\nRingworm is most commonly treated with antifungal medications, which can be applied to the skin or taken by mouth. In some cases, surgery may be necessary to remove infected hair or nails.",
      "safetyAttributes": {
        "scores": [
          1
        ],
        "blocked": false,
        "categories": [
          "Health"
        ]
      }
    }
  ],
  "metadata": {
    "tokenMetadata": {
      "outputTokenCount": {
        "totalTokens": 140,
        "totalBillableCharacters": 508
      },
      "inputTokenCount": {
        "totalTokens": 10,
        "totalBillableCharacters": 36
      }
    }
  }
}

Stream response from Generative AI models

The parameters are the same for streaming and non-streaming requests to the APIs.

To view sample code requests and responses using the REST API, see Examples using the REST API.

To view sample code requests and responses using the Vertex AI SDK for Python, see Examples using Vertex AI SDK for Python.