EvolutionaryScale Sampling API

openapi: 3.1.0
info:
  title: EvolutionaryScale Forge ESM3 Embeddings Sampling API
  description: 'Hosted inference API for the ESM3 multimodal protein language model from EvolutionaryScale. ESM3 reasons jointly across sequence, structure, and function tracks. The Forge API exposes generate, batch_generate, encode, decode, forward_and_sample, and logits operations across small (1.4B), medium (7B), and large (98B) parameter checkpoints. All requests authenticate with a Forge API token obtained from forge.evolutionaryscale.ai. This OpenAPI is reconstructed from the open-source `esm` Python SDK (Biohub/esm) which is the canonical client for the Forge service.

    '
  version: 2024-08-01
  contact:
    name: EvolutionaryScale Forge
    url: https://forge.evolutionaryscale.ai
  license:
    name: Cambrian Inference Clickthrough License Agreement
    url: https://www.evolutionaryscale.ai/policies/cambrian-inference-clickthrough-license-agreement
servers:
- url: https://forge.evolutionaryscale.ai
  description: EvolutionaryScale Forge Production
security:
- BearerAuth: []
tags:
- name: Sampling
  description: Low-level forward passes with logits and sampling control.
paths:
  /api/v1/forward_and_sample:
    post:
      summary: Forward And Sample
      description: 'Executes a single forward pass of ESM3 against a tokenized protein and samples tokens according to a `SamplingConfig`. Intended for power users who need fine-grained control over each iteration (for example, chain-of-thought protein design or custom sampling schedules).

        '
      operationId: forwardAndSample
      tags:
      - Sampling
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/ForwardAndSampleRequest'
      responses:
        '200':
          description: Sampled output with optional logits and entropy.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/ForwardAndSampleOutput'
        4XX:
          $ref: '#/components/responses/ErrorResponse'
  /api/v1/logits:
    post:
      summary: Get Logits And Embeddings
      description: 'Returns raw per-position logits, embeddings, and optional hidden states for a tokenized protein. The SDK discourages routine use because logits payloads are large; prefer encode + forward_and_sample for typical workflows.

        '
      operationId: logits
      tags:
      - Sampling
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/LogitsRequest'
      responses:
        '200':
          description: Logits and embeddings response.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/LogitsOutput'
        4XX:
          $ref: '#/components/responses/ErrorResponse'
  /api/v1/esmc/logits:
    post:
      summary: Get ESM C Logits And Embeddings
      description: 'Returns per-residue sequence logits and (optionally) embeddings or hidden states for an `ESMProteinTensor` produced by `esmc/encode`. Use for representation learning, classification heads, mutational effect prediction, and similar downstream tasks.

        '
      operationId: esmcLogits
      tags:
      - Sampling
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/ESMCLogitsRequest'
      responses:
        '200':
          description: Logits and embedding output.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/LogitsOutput_2'
        4XX:
          $ref: '#/components/responses/ErrorResponse'
components:
  schemas:
    ForwardAndSampleOutput:
      type: object
      properties:
        protein_tensor:
          $ref: '#/components/schemas/ESMProteinTensor'
        per_residue_entropy:
          type: array
          items:
            type: number
        topk_logprobs:
          type: object
        embeddings:
          type: array
          items:
            type: array
            items:
              type: number
    ESMCLogitsRequest:
      type: object
      required:
      - model
      - protein_tensor
      properties:
        model:
          type: string
          example: esmc-300m-2024-12
        protein_tensor:
          $ref: '#/components/schemas/ESMProteinTensor_2'
        config:
          $ref: '#/components/schemas/LogitsConfig'
    Error:
      type: object
      properties:
        error:
          type: object
          properties:
            type:
              type: string
            message:
              type: string
            code:
              type: string
    SamplingConfig:
      type: object
      description: Per-track sampling configuration for forward_and_sample.
      properties:
        sequence:
          $ref: '#/components/schemas/SamplingTrackConfig'
        structure:
          $ref: '#/components/schemas/SamplingTrackConfig'
        secondary_structure:
          $ref: '#/components/schemas/SamplingTrackConfig'
        sasa:
          $ref: '#/components/schemas/SamplingTrackConfig'
        function:
          $ref: '#/components/schemas/SamplingTrackConfig'
    LogitsConfig:
      type: object
      properties:
        sequence:
          type: boolean
          default: true
        return_embeddings:
          type: boolean
          default: false
        return_hidden_states:
          type: boolean
          default: false
        ith_hidden_layer:
          type: integer
          default: -1
    LogitsOutput:
      type: object
      description: Logits, embeddings, and optional hidden states.
      properties:
        logits:
          type: object
          properties:
            sequence:
              type: array
              items:
                type: array
                items:
                  type: number
            structure:
              type: array
              items:
                type: array
                items:
                  type: number
        embeddings:
          type: array
          description: Per-residue embedding vectors.
          items:
            type: array
            items:
              type: number
        hidden_states:
          type: array
          items:
            type: array
            items:
              type: array
              items:
                type: number
    SamplingTrackConfig:
      type: object
      properties:
        temperature:
          type: number
        top_p:
          type: number
        invalid_ids:
          type: array
          items:
            type: integer
        only_sample_masked_tokens:
          type: boolean
          default: true
    ESMProteinTensor_2:
      type: object
      properties:
        sequence:
          type: array
          items:
            type: integer
    ForwardAndSampleRequest:
      type: object
      required:
      - model
      - protein_tensor
      - sampling_config
      properties:
        model:
          type: string
          example: esm3-medium-2024-08
        protein_tensor:
          $ref: '#/components/schemas/ESMProteinTensor'
        sampling_config:
          $ref: '#/components/schemas/SamplingConfig'
    ESMProteinTensor:
      type: object
      description: Tokenized representation of a protein across all ESM3 tracks.
      properties:
        sequence:
          type: array
          items:
            type: integer
        structure:
          type: array
          items:
            type: integer
        secondary_structure:
          type: array
          items:
            type: integer
        sasa:
          type: array
          items:
            type: integer
        function:
          type: array
          items:
            type: integer
        residue_annotations:
          type: array
          items:
            type: integer
    LogitsRequest:
      type: object
      required:
      - model
      - protein_tensor
      properties:
        model:
          type: string
          example: esm3-medium-2024-08
        protein_tensor:
          $ref: '#/components/schemas/ESMProteinTensor'
        config:
          type: object
          properties:
            sequence:
              type: boolean
              default: false
            structure:
              type: boolean
              default: false
            secondary_structure:
              type: boolean
              default: false
            sasa:
              type: boolean
              default: false
            function:
              type: boolean
              default: false
            return_embeddings:
              type: boolean
              default: false
            return_hidden_states:
              type: boolean
              default: false
            ith_hidden_layer:
              type: integer
              default: -1
    LogitsOutput_2:
      type: object
      properties:
        logits:
          type: object
          properties:
            sequence:
              type: array
              items:
                type: array
                items:
                  type: number
        embeddings:
          type: array
          items:
            type: array
            items:
              type: number
        hidden_states:
          type: array
          items:
            type: array
            items:
              type: array
              items:
                type: number
  responses:
    ErrorResponse:
      description: Error returned by the Forge API.
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/Error'
  securitySchemes:
    BearerAuth:
      type: http
      scheme: bearer
      description: Forge API token issued via forge.evolutionaryscale.ai.