Neuronpedia API

https://www.neuronpedia.org/api-doc
Admin
Neuronpedia API provides programmatic access to neural network feature analysis, including...
Tokens:2,528
Snippets:21
Trust Score:9.1
Update:4 months ago
Show doc for...
Tokens:
Raw
### Get Activations Request Body Example

Source: https://www.neuronpedia.org/api-doc#description/introduction

Example JSON body for the GET Activations API endpoint. Used to specify parameters for retrieving activations.

```JSON
{
  "modelId": "",
  "source": "",
  "index": ""
}
```

--------------------------------

### List User's Vectors Request Example

Source: https://www.neuronpedia.org/api-doc#description/introduction

Example of how to list all vectors owned by the authenticated user. Requires an API key for authentication.

```Python
requests.post("/api/vector/list-owned",
    headers={
      "x-api-key": "YOUR_SECRET_TOKEN"
    }
)
```

--------------------------------

### Get Vector Details Request Example

Source: https://www.neuronpedia.org/api-doc#description/introduction

Example of how to retrieve details for a specific vector. Requires the vector's model ID, source, and index.

```Python
requests.post("/api/vector/get",
    headers={
      "Content-Type": "application/json"
    },
    json={
      "modelId": "",
      "source": "",
      "index": ""
    }
)
```

--------------------------------

### All Feature Activations in a Source/SAE

Source: https://www.neuronpedia.org/api-doc#description/introduction

Gets activation values for all features in a source/SAE when processing custom input text(s). It returns the activations and tokenized prompts only, without any of the dashboards or explanations.

```APIDOC
## POST /api/activation/source

### Description
Gets activation values for all features in a source/SAE when processing custom input text(s). It returns the activations and tokenized prompts only, without any of the dashboards or explanations.

### Method
POST

### Endpoint
/api/activation/source

### Parameters
#### Request Body
- **customText** (string) - Required - The custom text to process. Either a single string or an array of strings. If it's a single string, the max is a 1024 token text. If it's an array, the max is 4 strings of 256 tokens max each.
- **modelId** (string) - Required - The model the source is in. Default: gpt2-small
- **source** (string) - Required - The source/SAE ID. Default: 9-res-jb

### Responses
#### Success Response (200)
Successful response, with activation data
#### Error Response
- **400**: Bad request, missing or invalid input
- **401**: Unauthorized, user doesn't have access to the model or SAE set
- **500**: Internal server error
```

--------------------------------

### Python Requests Example

Source: https://www.neuronpedia.org/api-doc#description/introduction

This snippet demonstrates how to send a request to the Neuronpedia API using the Python Requests library. It shows how to access all cookies, headers, and the body of the response.

```python
import requests

url = "YOUR_API_ENDPOINT"
headers = {
    "Content-Type": "application/json"
}

try:
    response = requests.get(url, headers=headers)
    response.raise_for_status()  # Raise an exception for bad status codes

    # Access all cookies
    all_cookies = response.cookies.get_dict()
    print("All Cookies:", all_cookies)

    # Access all headers
    all_headers = response.headers
    print("All Headers:", all_headers)

    # Access the response body
    response_body = response.json()  # Assuming the response is JSON
    print("Response Body:", response_body)

except requests.exceptions.RequestException as e:
    print(f"An error occurred: {e}")
except ValueError as e:
    print(f"Error decoding JSON response: {e}")

```

--------------------------------

### Delete Vector Request Example

Source: https://www.neuronpedia.org/api-doc#description/introduction

Example of how to delete a vector using the Neuronpedia API. Requires authentication and specifies the vector's model ID, source, and index.

```Python
requests.post("/api/vector/delete",
    headers={
      "Content-Type": "application/json",
      "x-api-key": "YOUR_SECRET_TOKEN"
    },
    json={
      "modelId": "",
      "source": "",
      "index": ""
    }
)
```

--------------------------------

### Feature Activation for Text

Source: https://www.neuronpedia.org/api-doc#description/introduction

Gets activation values for a given feature when processing custom input text. Equivalent to going to a feature page and entering custom text.

```APIDOC
## POST /api/activation/new

### Description
Gets activation values for a given feature when processing custom input text. Equivalent to going to a feature page and entering custom text.

### Method
POST

### Endpoint
/api/activation/new

### Parameters
#### Request Body
- **customText** (string) - Required - The custom text to process. Either a single string or an array of strings.
- **feature** (object) - Required - For feature details including modelId, source, and index.
  - **modelId** (string) - Required - The model the source is in. Default: gpt2-small
  - **source** (string) - Required - The source/SAE ID. Default: 9-res-jb
  - **index** (string) - Required - The feature index.

### Responses
#### Success Response (200)
Successful response, with activation data. Response schema includes tokens, activations, and error fields.
#### Error Response
- **400**: Bad request, missing or invalid input
- **401**: Unauthorized, user doesn't have access to the model or SAE set
- **500**: Internal server error
```

--------------------------------

### Get Vector Details

Source: https://www.neuronpedia.org/api-doc#description/introduction

Returns details for a specific vector by model ID, source, and index.

```APIDOC
## POST /api/vector/get

### Description
Returns details for a specific vector by model ID, source and index.

### Method
POST

### Endpoint
/api/vector/get

### Parameters
#### Request Body
- **index** (string) - Required - Vector index
- **modelId** (string) - Required - ID of the model
- **source** (string) - Required - Source/layer identifier

### Request Example
```python
requests.post("/api/vector/get",
    headers={
      "Content-Type": "application/json"
    },
    json={
      "modelId": "",
      "source": "",
      "index": ""
    }
)
```

### Response
#### Success Response (200)
- **vector** (object) - Vector details
  - **modelId** (string)
  - **layer** (string)
  - **index** (string)
  - **vectorLabel** (string)
  - **vectorDefaultSteerStrength** (number)
  - **hookName** (string)
  - **vector** (array of numbers)
  - **createdAt** (string)
```

--------------------------------

### Get Feature Details

Source: https://www.neuronpedia.org/api-doc#description/introduction

Retrieves details for a specific feature identified by modelId, layer, and index.

```APIDOC
## GET /api/feature/{modelId}/{layer}/{index}

### Description
Retrieves details for a specific feature identified by modelId, layer, and index.

### Method
GET

### Endpoint
/api/feature/{modelId}/{layer}/{index}

### Parameters
#### Path Parameters
- **modelId** (string) - Required - The ID of the model.
- **layer** (string) - Required - The layer of the feature.
- **index** (string) - Required - The index of the feature.
```

--------------------------------

### Post Feature Activation Request

Source: https://www.neuronpedia.org/api-doc#description/introduction

Sends a POST request to the /api/activation/new endpoint to get activation values for a specific feature when processing custom input text. The request body includes feature details and the custom text.

```python
requests.post("/api/activation/new",
    headers={
      "Content-Type": "application/json"
    },
    json={
      "feature": {
        "modelId": "gpt2-small",
        "source": "9-res-jb",
        "index": "200"
      },
      "customText": {}
    }
)
```

--------------------------------

### Post Activation Source Request

Source: https://www.neuronpedia.org/api-doc#description/introduction

Sends a POST request to the /api/activation/source endpoint to retrieve activation values for all features in a source/SAE for custom input text. Ensure 'Content-Type' is set to 'application/json'.

```python
requests.post("/api/activation/source",
    headers={
      "Content-Type": "application/json"
    },
    json={
      "modelId": "gpt2-small",
      "source": "9-res-jb",
      "customText": {}
    }
)
```

--------------------------------

### List User's Vectors

Source: https://www.neuronpedia.org/api-doc#description/introduction

Returns all vectors owned by the authenticated user. Requires authentication.

```APIDOC
## POST /api/vector/list-owned

### Description
Returns all vectors owned by the authenticated user.

### Method
POST

### Endpoint
/api/vector/list-owned

### Request Example
```python
requests.post("/api/vector/list-owned",
    headers={
      "x-api-key": "YOUR_SECRET_TOKEN"
    }
)
```

### Response
#### Success Response (200)
- **vectors** (array of objects) - List of user's vectors
  - **modelId** (string)
  - **layer** (string)
  - **index** (string)
  - **vectorLabel** (string)
  - **createdAt** (string)
  - **vectorDefaultSteerStrength** (number)
  - **hookName** (string)
  - **vector** (array of numbers)
```

--------------------------------

### Activation Response Schema

Source: https://www.neuronpedia.org/api-doc#description/introduction

Defines the structure of a successful response from activation-related API endpoints, including tokens, activation data, and potential error messages.

```json
{
  "tokens": [
    "string"
  ],
  "activations": {
    "layer": "string",
    "index": 1,
    "values": [
      1
    ],
    "maxValue": 1,
    "maxValueIndex": 1,
    "dfaValues": [
      1
    ],
    "dfaTargetIndex": 1,
    "dfaMaxValue": 1
  },
  "error": "string"
}
```

--------------------------------

### Delete Vector

Source: https://www.neuronpedia.org/api-doc#description/introduction

Deletes an existing vector on Neuronpedia. Users can only delete vectors they have created. Requires authentication.

```APIDOC
## POST /api/vector/delete

### Description
Deletes an existing vector on Neuronpedia. You can only delete vectors you created.

### Method
POST

### Endpoint
/api/vector/delete

### Parameters
#### Request Body
- **index** (string) - Required - Index of the vector to delete
- **modelId** (string) - Required - ID of the model
- **source** (string) - Required - Source identifier for the vector

### Request Example
```python
requests.post("/api/vector/delete",
    headers={
      "Content-Type": "application/json",
      "x-api-key": "YOUR_SECRET_TOKEN"
    },
    json={
      "modelId": "",
      "source": "",
      "index": ""
    }
)
```

### Response
#### Success Response (200)
- **message** (string) - Vector deleted successfully
```
Tokens:
Raw