> ## Documentation Index > Fetch the complete documentation index at: https://docs.coreweave.com/llms.txt > Use this file to discover all available pages before exploring further. # Créer une évaluation ## OpenAPI ````yaml fr/openapi/model-distillation/management.openapi.yaml POST /evals openapi: 3.1.0 info: title: CoreWeave Model Distillation Management API version: 1.0.0 description: > Gérez les fournisseurs, les projets (appelés tasks dans les chemins d’API), les versions d’acheminement, les jeux de données, le réétiquetage, les modèles affinés, les évaluations et les analyses. Il s’agit de la même API que celle utilisée par Model Distillation. servers: - url: https://distillation.training.wandb.ai/v1 description: Production security: - WandbCredential: [] tags: - name: Providers description: >- Points de terminaison compatibles avec OpenAI, identifiants d’authentification, catalogues de modèles et tarification. - name: Tasks description: Identité stable des tâches et informations propres à chaque tâche. - name: Routing description: Modèles cibles, pondérations et paramètres de requête versionnés. - name: Datasets description: >- Instantanés de données reproductibles, entrées, réétiquetage et sorties de modèles réutilisables. - name: Fine-tunes description: Tâches de fine-tuning supervisé et artifacts de modèles hébergés. - name: Evaluations description: >- Évaluations par comparaison directe, correspondance exacte et catégorisation. - name: Analytics description: >- Modèles de lecture des données de trafic, de consommation de jetons, d’erreurs et de coûts estimés des tâches. - name: UI preferences description: Préférences d’affichage par tâche utilisées par l’interface utilisateur. paths: /evals: parameters: - $ref: '#/components/parameters/WandbEntity' post: tags: - Evaluations summary: Créer une évaluation operationId: createEval requestBody: required: true content: application/json: schema: type: object properties: name: type: string minLength: 1 maxLength: 128 description: >- Un nom d’évaluation court et facile à repérer. Privilégiez 2 à 5 mots et évitez d’inclure le nom du jeu de données, la liste de tous les modèles comparés ou des détails de configuration. dataset_id: type: string format: uuid pattern: >- ^([0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[1-8][0-9a-fA-F]{3}-[89abAB][0-9a-fA-F]{3}-[0-9a-fA-F]{12}|00000000-0000-0000-0000-000000000000|ffffffff-ffff-ffff-ffff-ffffffffffff)$ spec: oneOf: - type: object properties: type: type: string const: h2h_judge judge_model_ref: type: string pattern: ^[a-z0-9][a-z0-9_-]{0,63}\/\S+$ judge_prompt: default: >- You are an intelligent and fair judge of chatbots. Evaluate the following two responses and choose which one is better. If the outputs are of similar quality, you can mark them as a tie. type: string minLength: 1 reference_conversation_count: default: 0 type: integer minimum: 0 maximum: 50 required: - type - judge_model_ref additionalProperties: false - type: object additionalProperties: false required: - type properties: type: type: string const: exact_match - type: object additionalProperties: false required: - type - field properties: type: type: string const: categorization field: type: string minLength: 1 maxLength: 256 participants: minItems: 1 type: array items: anyOf: - oneOf: - type: object additionalProperties: false required: - kind properties: kind: type: string const: original - type: object properties: kind: type: string const: relabel relabel_run_id: type: string format: uuid pattern: >- ^([0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[1-8][0-9a-fA-F]{3}-[89abAB][0-9a-fA-F]{3}-[0-9a-fA-F]{12}|00000000-0000-0000-0000-000000000000|ffffffff-ffff-ffff-ffff-ffffffffffff)$ required: - kind - relabel_run_id additionalProperties: false - type: object properties: kind: type: string const: model model_ref: type: string pattern: ^[a-z0-9][a-z0-9_-]{0,63}\/\S+$ required: - kind - model_ref additionalProperties: false - type: object properties: kind: type: string const: provider model_ref: type: string pattern: ^[a-z0-9][a-z0-9_-]{0,63}\/\S+$ required: - kind - model_ref additionalProperties: false - oneOf: - type: object properties: kind: type: string const: model model_ref: type: string pattern: ^[a-z0-9][a-z0-9_-]{0,63}\/\S+$ required: - kind - model_ref additionalProperties: false - type: object properties: kind: type: string const: provider model_ref: type: string pattern: ^[a-z0-9][a-z0-9_-]{0,63}\/\S+$ required: - kind - model_ref additionalProperties: false reference: oneOf: - type: object additionalProperties: false required: - kind properties: kind: type: string const: original - type: object properties: kind: type: string const: model model_ref: type: string pattern: ^[a-z0-9][a-z0-9_-]{0,63}\/\S+$ required: - kind - model_ref additionalProperties: false - type: object properties: kind: type: string const: provider model_ref: type: string pattern: ^[a-z0-9][a-z0-9_-]{0,63}\/\S+$ required: - kind - model_ref additionalProperties: false sample_size: type: integer exclusiveMinimum: 0 maximum: 9007199254740991 required: - name - dataset_id - spec - participants - sample_size additionalProperties: false responses: '202': description: Les cas d’évaluation ont été mis en file d’attente. content: application/json: schema: $ref: '#/components/schemas/Eval' default: $ref: '#/components/responses/Error' components: parameters: WandbEntity: name: Wandb-Entity in: header required: false description: >- entity personnel ou d’équipe accessible. Omettez ce champ pour utiliser l’entity par défaut de l’utilisateur authentifié. schema: type: string schemas: Eval: type: object additionalProperties: true required: - id - dataset_id - name - spec - participants - sample_size - status - progress - created_at - updated_at properties: id: type: string format: uuid dataset_id: type: string format: uuid name: type: string spec: $ref: '#/components/schemas/EvalSpec' participants: type: array items: $ref: '#/components/schemas/EvalParticipant' reference: oneOf: - $ref: '#/components/schemas/EvalParticipant' - type: 'null' sample_size: type: integer status: type: string enum: - queued - running - completed - failed - stale progress: type: object required: - total - queued - running - completed - failed properties: total: type: integer queued: type: integer running: type: integer completed: type: integer failed: type: integer results: oneOf: - $ref: '#/components/schemas/JsonObject' - type: 'null' failure_summary: type: object additionalProperties: true created_at: type: string format: date-time updated_at: type: string format: date-time EvalSpec: oneOf: - type: object additionalProperties: false required: - type - judge_model_ref properties: type: type: string const: h2h_judge judge_model_ref: $ref: '#/components/schemas/ModelRef' judge_prompt: type: string minLength: 1 - type: object additionalProperties: false required: - type properties: type: type: string const: exact_match - type: object additionalProperties: false required: - type - field properties: type: type: string const: categorization field: type: string minLength: 1 maxLength: 256 EvalParticipant: oneOf: - type: object additionalProperties: false required: - kind properties: kind: type: string const: original - type: object additionalProperties: false required: - kind - relabel_run_id properties: kind: type: string const: relabel relabel_run_id: type: string format: uuid - type: object additionalProperties: false required: - kind - model_ref properties: kind: type: string enum: - model - provider model_ref: $ref: '#/components/schemas/ModelRef' JsonObject: type: object additionalProperties: true ErrorResponse: type: object required: - error properties: error: type: object required: - message - type properties: message: type: string type: type: string example: error: message: Task 'missing' not found in entity 'your-team' type: not_found ModelRef: type: string pattern: ^[a-z0-9][a-z0-9_-]{0,63}/\S+$ example: openai/gpt-5.6-sol responses: Error: description: La requête a échoué. content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' securitySchemes: WandbCredential: type: http scheme: bearer bearerFormat: W&B API key description: Une clé API W&B. ```` ## Related topics - [Créer une évaluation](/fr/products/wandb/weave/tutorial-eval.md)