> ## Documentation Index > Fetch the complete documentation index at: https://docs.revilico.bio/llms.txt > Use this file to discover all available pages before exploring further. # Start a RevTrain run > Fine-tunes the from-scratch molecule generator on your own set of molecules, so RevDenovo and RevOptimization runs that use the trained model generate molecules resembling that set. The output is a model, not molecules. Starts a run and returns it with status `queued`; poll GET /v1/runs/{runId} until it finishes. Runs spend credits. **Limits:** `maxTrainingMolecules`: 100000, `maxEpochs`: 1000. How the engine works and how to read its results: [RevTrain documentation](/docs/revtrain). **Scope:** `runs:write`. ## OpenAPI ````yaml /api-reference/openapi.json post /v1/engines/revtrain/runs openapi: 3.1.0 info: title: Revilico API version: 1.0.0 description: >- Run Revilico's computational chemistry and biology engines from your own code: upload input files, start runs, follow them and read their results. servers: - url: https://public-api.revilico.bio security: - bearerAuth: [] tags: - name: Account description: Who you are, your credits and what your runs have used. - name: Engines description: The engines you can run, their parameter schemas and examples. - name: Files description: 'Upload input files: create an upload, then PUT the bytes.' - name: Runs description: Start, follow and cancel engine runs. - name: Results and outputs description: 'Read what a run produced: results, files and outputs.' - name: Projects description: Group runs into the projects you see in the web app. - name: Sharing description: >- Share runs with people in your organization, as in the web app's Pipeline sharing. - name: Notes description: >- RevNotes: read, write and share the rich-text notes and note folders of the web app's RevNotes panel, in markdown or HTML. - name: Drive description: >- Your RevDrive files and shared drives, as in the web app: browse, read, use files as run inputs, save outputs, and organize. Ids: `dfile_…` and `dfld_…` in a personal Drive, `drive_…` for a shared drive, `sdfile_…` and `sdfld_…` in one. - name: Engine runs description: >- Start a run of one engine, with that engine's parameters and input files. Same endpoint as **Start a run**. - name: Health description: Service status. Needs no API key. paths: /v1/engines/revtrain/runs: post: tags: - Engine runs summary: Start a RevTrain run description: >- Fine-tunes the from-scratch molecule generator on your own set of molecules, so RevDenovo and RevOptimization runs that use the trained model generate molecules resembling that set. The output is a model, not molecules. Starts a run and returns it with status `queued`; poll GET /v1/runs/{runId} until it finishes. Runs spend credits. **Limits:** `maxTrainingMolecules`: 100000, `maxEpochs`: 1000. How the engine works and how to read its results: [RevTrain documentation](/docs/revtrain). **Scope:** `runs:write`. operationId: engineRuns.revtrain requestBody: required: true content: application/json: schema: type: object properties: name: description: A name for the run, shown in the web app. type: string maxLength: 200 parameters: type: object properties: generator: default: reinvent description: >- The generator to fine-tune. reinvent (the only one for now) makes molecules from scratch; use the trained model with RevDenovo's reinvent generator. type: string enum: - reinvent training_smiles: default: [] description: >- The molecules the model should learn to resemble, one SMILES per item. Give them here or upload them as the training_set file (or both); for more than about 2,000 molecules use the file. Hundreds to thousands of related molecules work best. maxItems: 100000 type: array items: type: string pattern: ^[!-~]{1,1000}$ epochs: default: 10 description: >- How many passes over the training molecules. More fits the set more closely and takes longer; too many makes the model copy the set. type: integer minimum: 1 maximum: 1000 save_every_n_epochs: default: 5 description: >- Save a snapshot of the model every this many passes (in the archive output). The final model is always saved. type: integer minimum: 1 maximum: 1000 batch_size: default: 128 description: Training molecules per learning step. type: integer minimum: 1 maximum: 1024 num_reference_molecules: default: 50 description: >- Training molecules picked at random to compare generated molecules against while training (for monitoring only). type: integer minimum: 1 maximum: 10000 sample_batch_size: default: 100 description: >- Molecules generated after each saved pass to check how training is going (for monitoring only). type: integer minimum: 100 maximum: 10000 additionalProperties: false inputs: type: object properties: training_set: type: string description: >- Training molecules, for sets too large to send inline: SMI or TXT with one SMILES per line (anything after the first space, such as a name, is ignored), or CSV with a column named smiles (otherwise the first column is used). Lines starting with # are skipped. A file id: an upload (`file_…`) or a Drive file (`dfile_…`, `sdfile_…`). Formats: smi, txt, csv; up to 20 MB. additionalProperties: false budget: description: Most credits the run may spend; it is stopped beyond that. type: number exclusiveMinimum: 0 projectId: description: >- Add the run to this project (`proj_…`); needs editor or owner on it. type: string additionalProperties: false examples: inline: summary: >- Fine-tune the from-scratch generator on a few molecules given inline (a real set would have hundreds or more). value: name: COX-2 inhibitor generator parameters: training_smiles: - Cc1ccc(-c2cc(C(F)(F)F)nn2-c2ccc(S(N)(=O)=O)cc2)cc1 - CS(=O)(=O)c1ccc(C2=C(c3ccccc3)C(=O)OC2)cc1 - Cc1onc(-c2ccccc2)c1-c1ccc(S(N)(=O)=O)cc1 inputs: {} file: summary: >- Fine-tune on a larger set uploaded as a file, with more passes over it. value: name: Kinase series A generator parameters: epochs: 30 save_every_n_epochs: 10 inputs: training_set: file_… responses: '202': description: OK content: application/json: schema: $ref: '#/components/schemas/Run' '400': $ref: '#/components/responses/400' '401': $ref: '#/components/responses/401' '402': $ref: '#/components/responses/402' '403': $ref: '#/components/responses/403' '404': $ref: '#/components/responses/404' '422': $ref: '#/components/responses/422' '500': $ref: '#/components/responses/500' '502': $ref: '#/components/responses/502' components: schemas: Run: type: object properties: id: type: string description: Run id, `run_…`. engine: type: string description: Engine slug, e.g. `revredox`. name: type: - string - 'null' access: type: string enum: - owner - shared - project description: >- Why you can see the run: you started it, it was shared with you, or it is in a project you belong to. Only the owner can cancel it or read its inputs. owner: type: object properties: name: type: - string - 'null' email: type: string required: - name - email description: Who started the run. status: type: string enum: - queued - running - succeeded - failed - cancelled description: >- queued until the engine picks the run up; succeeded, failed and cancelled are final. statusReason: description: >- Why a cancelled run stopped: `user`, `insufficient_credits` or `budget_exceeded`. Null otherwise. type: - string - 'null' progress: type: number description: Progress the engine reports; 0 until it reports any. createdAt: type: string format: date-time startedAt: anyOf: - type: string format: date-time - type: 'null' endedAt: anyOf: - type: string format: date-time - type: 'null' budget: description: Most credits the run may spend, if one was set. type: - number - 'null' creditsUsed: type: number description: Credits charged to the run so far. error: anyOf: - type: object properties: message: type: string required: - message - type: 'null' description: Set when status is failed. children: type: object properties: total: type: number queued: type: number running: type: number succeeded: type: number failed: type: number cancelled: type: number required: - total - queued - running - succeeded - failed - cancelled description: >- For engines that run each input as its own job: how many children are in each status. GET /v1/runs/{runId}/children lists them. links: type: object propertyNames: type: string additionalProperties: type: string description: >- Related endpoints: self, inputs (owner only), results, outputs, files. warnings: description: >- Set when the run was created but a follow-up step, such as adding it to a project, failed. type: array items: type: string required: - id - engine - name - access - owner - status - statusReason - progress - createdAt - startedAt - endedAt - budget - creditsUsed - error - links Error: type: object properties: error: type: object properties: code: type: string enum: - VALIDATION_FAILED - INVALID_ID - INVALID_PARAMETERS - INPUT_LIMIT_EXCEEDED - NOT_FOUND - FORBIDDEN - METHOD_NOT_ALLOWED - CONFLICT - PAYLOAD_TOO_LARGE - MISSING_CREDENTIALS - INVALID_CREDENTIALS - WRONG_ENVIRONMENT - CREDENTIALS_EXPIRED - CREDENTIALS_REVOKED - ACCOUNT_INACTIVE - API_ACCESS_NOT_ENABLED - INSUFFICIENT_SCOPE - SESSION_REQUIRED - TOO_MANY_AUTH_FAILURES - API_KEY_LIMIT_REACHED - ENGINE_NOT_ENABLED - DOCS_NOT_AVAILABLE - INSUFFICIENT_CREDITS - RUN_NOT_COMPLETE - ENGINE_SUBMIT_FAILED - RATE_LIMITED - SERVICE_UNAVAILABLE - INTERNAL_ERROR description: Stable within /v1; branch on this, not on the message. message: type: string description: Readable explanation, for people. details: description: >- Extra context, e.g. `issues` ({ path, message }[]) for validation errors. type: object propertyNames: type: string additionalProperties: {} required: - code - message required: - error responses: '400': description: >- The request is malformed. `error.code` is one of `VALIDATION_FAILED`, `INVALID_ID`. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: code: VALIDATION_FAILED message: The request is malformed. '401': description: >- The API key is missing or not valid. `error.code` is one of `MISSING_CREDENTIALS`, `INVALID_CREDENTIALS`, `WRONG_ENVIRONMENT`, `CREDENTIALS_EXPIRED`, `CREDENTIALS_REVOKED`. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: code: MISSING_CREDENTIALS message: The API key is missing or not valid. '402': description: >- Your organization doesn't have enough credits. `error.code` is one of `INSUFFICIENT_CREDITS`. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: code: INSUFFICIENT_CREDITS message: Your organization doesn't have enough credits. '403': description: >- The key may not do this. `error.code` is one of `INSUFFICIENT_SCOPE`, `API_ACCESS_NOT_ENABLED`, `ENGINE_NOT_ENABLED`, `ACCOUNT_INACTIVE`, `FORBIDDEN`. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: code: INSUFFICIENT_SCOPE message: The key may not do this. '404': description: Not found, or not visible to you. `error.code` is one of `NOT_FOUND`. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: code: NOT_FOUND message: Not found, or not visible to you. '422': description: >- The parameters or input files don't fit the engine. `error.code` is one of `INVALID_PARAMETERS`, `INPUT_LIMIT_EXCEEDED`. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: code: INVALID_PARAMETERS message: The parameters or input files don't fit the engine. '500': description: >- Something went wrong on our side. `error.code` is one of `INTERNAL_ERROR`. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: code: INTERNAL_ERROR message: Something went wrong on our side. '502': description: >- The engine service rejected or failed the request. `error.code` is one of `ENGINE_SUBMIT_FAILED`. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: code: ENGINE_SUBMIT_FAILED message: The engine service rejected or failed the request. securitySchemes: bearerAuth: type: http scheme: bearer description: >- An API key from the web app (**Settings → Developer**), sent as `Authorization: Bearer rvl_live_…`. ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.