# CoreWeave Docs: English Forge Post-Training > Learn how to deploy, manage, and observe your AI training and inference workloads on CoreWeave. ## Post-Training - [Post-Training](https://docs.coreweave.com/products/post-training.md): Adapt models to your tasks with distillation, reinforcement learning, and supervised fine-tuning on managed infrastructure. ### Model Distillation - [Model Distillation](https://docs.coreweave.com/model-distillation.md): Train specialized models from real application traffic, compare them with the models you use today, and deploy the best result. #### Get started - [Overview](https://docs.coreweave.com/model-distillation/overview.md): Use one workflow to collect training data, train specialized models, prove their quality, and deploy them safely. - [Quick start](https://docs.coreweave.com/model-distillation/quickstart.md): Connect a project, collect data, train candidate models, evaluate them, and deploy a winner. - [How Model Distillation works](https://docs.coreweave.com/model-distillation/get-started/how-it-works.md): See how real product usage becomes a trained, evaluated, and safely deployed model. - [Train and deploy your first model](https://docs.coreweave.com/model-distillation/get-started/first-model.md): Turn real application examples into a specialized model and compare it before deployment. #### Features ##### Providers - [Providers](https://docs.coreweave.com/model-distillation/studio/providers.md): Register inference endpoints, manage callable model IDs, and configure pricing. - [Providers quick start](https://docs.coreweave.com/model-distillation/studio/providers-quickstart.md): Connect an inference provider and make one of its models available in the UI. ##### Projects and traffic - [Projects and traffic](https://docs.coreweave.com/model-distillation/studio/projects-and-traffic.md): Create a project with a stable proxy model name and connect your production Chat Completions application. - [Connect your application](https://docs.coreweave.com/model-distillation/get-started/connect-your-app.md): Send your first verified request through the Model Distillation proxy. - [Projects and versions](https://docs.coreweave.com/model-distillation/concepts/projects-and-versions.md): Use stable proxy model names and isolated routing versions to evolve behavior safely. ##### Datasets - [Datasets](https://docs.coreweave.com/model-distillation/studio/datasets.md): Build reproducible train and test data from project traces or an existing W&B project. - [Datasets Quick Start](https://docs.coreweave.com/model-distillation/studio/datasets-quickstart.md): Create a reproducible dataset from project traffic and inspect its training and test rows. - [Upload a Dataset](https://docs.coreweave.com/model-distillation/studio/datasets-upload.md): Create a dataset from examples you already have by writing entries or uploading a JSONL file through the Management API. - [Relabeling](https://docs.coreweave.com/model-distillation/studio/relabeling.md): Generate a stronger or more consistent training target without modifying original data. ##### Fine-tuning - [Fine-tuning](https://docs.coreweave.com/model-distillation/studio/fine-tuning.md): Train supported base models from original or relabeled dataset outputs. - [Fine-tuning Quick Start](https://docs.coreweave.com/model-distillation/studio/fine-tuning-quickstart.md): Train your first specialized model from a ready dataset. - [Models you can train](https://docs.coreweave.com/model-distillation/supported-models.md): Train several model families and find the best quality, cost, and latency tradeoff for your task. ##### Evaluations - [Evaluations](https://docs.coreweave.com/model-distillation/studio/evaluations.md): Compare outputs on stable validation samples and understand live, provisional results. - [Evaluations Quick Start](https://docs.coreweave.com/model-distillation/studio/evaluations-quickstart.md): Run a head-to-head evaluation between your reference output and one or more candidate models. ##### Routing - [Routing and versions](https://docs.coreweave.com/model-distillation/studio/routing-and-versions.md): Control target weights and parameters, and isolate changes before production rollout. - [Routing quick start](https://docs.coreweave.com/model-distillation/studio/routing-quickstart.md): Route a small share of project traffic to a candidate model and verify the rollout. ##### Analytics - [Analytics](https://docs.coreweave.com/model-distillation/studio/analytics.md): Monitor traffic, tokens, errors, and model cost from project-attributed traces. - [Analytics quick start](https://docs.coreweave.com/model-distillation/studio/analytics-quickstart.md): Check traffic health and estimate the savings from routing to a candidate model. #### Inference proxy - [Inference proxy overview](https://docs.coreweave.com/model-distillation/proxy/overview.md): Use one OpenAI-compatible endpoint for project routing, provider credentials, and trace capture. - [Chat Completions](https://docs.coreweave.com/model-distillation/proxy/chat-completions.md): Send non-streaming, streaming, tool, multimodal, and structured-output requests. - [Response caching](https://docs.coreweave.com/model-distillation/proxy/response-caching.md): Reuse the exact response for a repeated non-streaming request instead of calling the provider again. - [Model routing](https://docs.coreweave.com/model-distillation/proxy/model-routing.md): Choose project versions or registered provider models through the model field. - [Metadata and traces](https://docs.coreweave.com/model-distillation/proxy/metadata-and-traces.md): Select a W&B entity, preserve conversation identity, and attach searchable context. - [Proxy errors](https://docs.coreweave.com/model-distillation/proxy/errors.md): Diagnose authentication, entity, routing, provider, and upstream failures. #### API reference - [Model Distillation API reference](https://docs.coreweave.com/model-distillation/reference/overview.md): Choose between the management API and the inference proxy. - [Authentication and entities](https://docs.coreweave.com/model-distillation/reference/authentication.md): Authenticate with W&B and select the correct personal or team entity. - [API conventions](https://docs.coreweave.com/model-distillation/reference/conventions.md): Understand model references, asynchronous status, errors, pagination, and idempotency. ##### Management API - [Management API](https://docs.coreweave.com/model-distillation/reference/management.md): Manage Model Distillation providers, projects, datasets, training jobs, evaluations, routing, and analytics. ###### Providers - [List providers](https://docs.coreweave.com/model-distillation/reference/management/providers/list-providers.md): Lists entity providers, masked credentials, deployment state, and model entries. Lazily provisions the built-in `wandb-inference` provider. - [Get a provider](https://docs.coreweave.com/model-distillation/reference/management/providers/get-a-provider.md) - [Validate a saved provider](https://docs.coreweave.com/model-distillation/reference/management/providers/validate-a-saved-provider.md): Sends a minimal completion through a saved provider and persists the resulting validation status. - [Create or rotate a provider](https://docs.coreweave.com/model-distillation/reference/management/providers/create-or-rotate-a-provider.md): Creates or rotates a provider credential. When validation_model is supplied, the configured backend must complete a minimal request to that model before any changes are saved. Omit validation_model to skip the test. Publishes remote provider material unless local_only is true, waiting up to 10 secon… - [Delete a provider](https://docs.coreweave.com/model-distillation/reference/management/providers/delete-a-provider.md): Fails with 409 while routing, relabeling, or an evaluation still references this provider. - [List resources using a provider](https://docs.coreweave.com/model-distillation/reference/management/providers/list-resources-using-a-provider.md) - [Add a custom provider model](https://docs.coreweave.com/model-distillation/reference/management/providers/add-a-custom-provider-model.md) - [Update a provider model](https://docs.coreweave.com/model-distillation/reference/management/providers/update-a-provider-model.md): Enables or disables the model in Model Distillation and sets optional per-million-token pricing. - [Delete a custom provider model](https://docs.coreweave.com/model-distillation/reference/management/providers/delete-a-custom-provider-model.md): Predefined catalog entries cannot be deleted; disable them instead. ###### Projects - [List projects](https://docs.coreweave.com/model-distillation/reference/management/projects/list-projects.md) - [List accessible Forge projects](https://docs.coreweave.com/model-distillation/reference/management/projects/list-accessible-forge-projects.md): Returns the names of Forge projects accessible to the caller in the resolved entity. To check whether a specific project can be linked to a new Model Distillation project, use GET /projects/availability. - [Check project setup availability](https://docs.coreweave.com/model-distillation/reference/management/projects/check-project-setup-availability.md): Checks whether the proxy model name and Forge project can be connected using the selected create or import mode. - [List project summaries](https://docs.coreweave.com/model-distillation/reference/management/projects/list-project-summaries.md): Returns project cards with dataset, fine-tuned model, evaluation, and approximate logged trace counts. Trace counts may be unavailable when Weave cannot be queried. - [Get a project](https://docs.coreweave.com/model-distillation/reference/management/projects/get-a-project.md) - [Create a Model Distillation project](https://docs.coreweave.com/model-distillation/reference/management/projects/create-a-model-distillation-project.md): Creates project version 1 and publishes its required initial routing configuration. Set project_mode to create for a new Forge project or import for an accessible existing one. Each Forge project can belong to exactly one Model Distillation project per entity. The path alias is the proxy model name… - [Delete a project](https://docs.coreweave.com/model-distillation/reference/management/projects/delete-a-project.md): Removes Model Distillation resources and applied routing, but never deletes the Forge project, source traces, runs, or artifacts. ###### Routing - [Get version 1 routing](https://docs.coreweave.com/model-distillation/reference/management/routing/get-version-1-routing.md) - [Replace version 1 routing](https://docs.coreweave.com/model-distillation/reference/management/routing/replace-version-1-routing.md): Saves the routing config, then waits up to 10 seconds for its publication to settle. The returned deployment reports the outcome: applied, failed with last_error, or still pending or applying. - [Delete version 1 routing](https://docs.coreweave.com/model-distillation/reference/management/routing/delete-version-1-routing.md) - [Get routing for a project version](https://docs.coreweave.com/model-distillation/reference/management/routing/get-routing-for-a-project-version.md) - [Replace routing for a project version](https://docs.coreweave.com/model-distillation/reference/management/routing/replace-routing-for-a-project-version.md): Saves the routing config, then waits up to 10 seconds for its publication to settle. The returned deployment reports the outcome: applied, failed with last_error, or still pending or applying. - [Delete routing for a project version](https://docs.coreweave.com/model-distillation/reference/management/routing/delete-routing-for-a-project-version.md) - [List project versions](https://docs.coreweave.com/model-distillation/reference/management/routing/list-project-versions.md) - [Create a project version](https://docs.coreweave.com/model-distillation/reference/management/routing/create-a-project-version.md): Creates the next sequential version and optionally publishes its initial targets. - [Delete a project version](https://docs.coreweave.com/model-distillation/reference/management/routing/delete-a-project-version.md): Version 1 cannot be deleted. Callers pinned to a deleted version begin receiving 404 responses. ###### Datasets - [List datasets](https://docs.coreweave.com/model-distillation/reference/management/datasets/list-datasets.md) - [Create a dataset](https://docs.coreweave.com/model-distillation/reference/management/datasets/create-a-dataset.md): With `params`, queues a build from project traces and returns 202. Without `params`, creates an empty ready dataset for the entries API and returns 201. - [Preview dataset deletion](https://docs.coreweave.com/model-distillation/reference/management/datasets/preview-dataset-deletion.md): Returns an informational preview of current dependencies and blockers plus operation status. DELETE computes its own scope; fetching this preview is optional. - [Get a dataset](https://docs.coreweave.com/model-distillation/reference/management/datasets/get-a-dataset.md) - [Rename a dataset](https://docs.coreweave.com/model-distillation/reference/management/datasets/rename-a-dataset.md) - [Delete a dataset](https://docs.coreweave.com/model-distillation/reference/management/datasets/delete-a-dataset.md): Deletes with one request; no preview is required. The server records the current dependency scope. Set `cascade=true` to also delete dependent fine-tunes, their training artifacts, and affected evaluations. Active fine-tunes, evaluations, and routed models still block deletion. Imports and relabelin… - [Append traces to a dataset](https://docs.coreweave.com/model-distillation/reference/management/datasets/append-traces-to-a-dataset.md): Runs another fixed query and adds only unseen trace/turn pairs while preserving the original split configuration. - [List relabel runs](https://docs.coreweave.com/model-distillation/reference/management/datasets/list-relabel-runs.md) - [Start a relabel run](https://docs.coreweave.com/model-distillation/reference/management/datasets/start-a-relabel-run.md) - [Select the active relabel run](https://docs.coreweave.com/model-distillation/reference/management/datasets/select-the-active-relabel-run.md): Selects the relabeled outputs used by the dataset, or clears the selection with null. - [Delete a relabel run](https://docs.coreweave.com/model-distillation/reference/management/datasets/delete-a-relabel-run.md) - [Resume a failed relabel run](https://docs.coreweave.com/model-distillation/reference/management/datasets/resume-a-failed-relabel-run.md): Queues the same failed run again and generates only outputs that are still missing. - [Apply a relabeling strategy to new rows](https://docs.coreweave.com/model-distillation/reference/management/datasets/apply-a-relabeling-strategy-to-new-rows.md): Creates an execution for the current dataset revision, carries forward existing outputs, and generates only missing outputs. - [Publish relabeled outputs to Weave](https://docs.coreweave.com/model-distillation/reference/management/datasets/publish-relabeled-outputs-to-weave.md): Queues a Weave sidecar publication for a completed relabel run. - [List dataset entries](https://docs.coreweave.com/model-distillation/reference/management/datasets/list-dataset-entries.md) - [Add or replace dataset entries](https://docs.coreweave.com/model-distillation/reference/management/datasets/add-or-replace-dataset-entries.md): Writes up to 5,000 OpenAI chat rows in one transaction. Send a JSON array, or the same rows as NDJSON (one row per line) with `Content-Type: application/x-ndjson`. Any invalid row rejects the request and is reported by its index; NDJSON line N is index N - 1. `mode=append` rejects a `row_id` that al… - [Move dataset entries to a split](https://docs.coreweave.com/model-distillation/reference/management/datasets/move-dataset-entries-to-a-split.md): Moves every entry matching `filters` (the listing's filters, including metadata) and `row_ids` to `split`. `fraction` keeps that share of the matching groups, ranked by a digest of the group id, so a group is never divided and a later call can take its share from what remains. Use `val` to hold a po… - [Start a JSONL upload into a dataset](https://docs.coreweave.com/model-distillation/reference/management/datasets/start-a-jsonl-upload-into-a-dataset.md): Opens an upload of one JSONL file, up to 1 GiB and 100,000 rows, into an existing Postgres-plane dataset. Each line uses the row shape of `POST .../entries`. PUT the whole file to `upload.url` with its exact declared size (for example `curl -T rows.jsonl "$url"`), then call complete. The URL must be… - [Get a dataset upload](https://docs.coreweave.com/model-distillation/reference/management/datasets/get-a-dataset-upload.md) - [Cancel a dataset upload](https://docs.coreweave.com/model-distillation/reference/management/datasets/cancel-a-dataset-upload.md) - [Finish a dataset upload](https://docs.coreweave.com/model-distillation/reference/management/datasets/finish-a-dataset-upload.md): Confirms the stored file matches its declared size and queues the import worker, which validates every line and then commits all rows in one transaction with the rules of `POST .../entries`. An invalid line, an existing row_id in append mode, or a group split across splits fails the upload without c… - [Publish a dataset to Weave](https://docs.coreweave.com/model-distillation/reference/management/datasets/publish-a-dataset-to-weave.md): Queues publication of the current dataset revision to Weave. - [List comparison-output jobs](https://docs.coreweave.com/model-distillation/reference/management/datasets/list-comparison-output-jobs.md) - [Ensure comparison-model outputs](https://docs.coreweave.com/model-distillation/reference/management/datasets/ensure-comparison-model-outputs.md): Queues only missing validation outputs and reuses matching cached generations. - [Regenerate one comparison model's outputs](https://docs.coreweave.com/model-distillation/reference/management/datasets/regenerate-one-comparison-models-outputs.md): Deletes the cached validation outputs for a model on this dataset and queues complete regeneration using its current inference defaults. Evaluations that depend on those outputs become stale. - [Get a dataset entry](https://docs.coreweave.com/model-distillation/reference/management/datasets/get-a-dataset-entry.md) - [Delete a dataset entry](https://docs.coreweave.com/model-distillation/reference/management/datasets/delete-a-dataset-entry.md) ###### Fine-tunes - [List fine-tunes](https://docs.coreweave.com/model-distillation/reference/management/fine-tunes/list-fine-tunes.md) - [Create a fine-tune](https://docs.coreweave.com/model-distillation/reference/management/fine-tunes/create-a-fine-tune.md) - [Preview fine-tune deletion](https://docs.coreweave.com/model-distillation/reference/management/fine-tunes/preview-fine-tune-deletion.md): Returns an informational preview of evaluation impact and blockers plus operation status. Fetching this preview is optional before DELETE. - [Get a fine-tune](https://docs.coreweave.com/model-distillation/reference/management/fine-tunes/get-a-fine-tune.md) - [Delete a fine-tune](https://docs.coreweave.com/model-distillation/reference/management/fine-tunes/delete-a-fine-tune.md): Queues durable deletion with one request. Removes the training model, Forge artifacts, generated outputs, and Model Distillation record. The fine-tune is removed from comparison sets; evaluations using it as ground truth are deleted. Repeating DELETE returns the existing operation or retries failed… - [Retry a failed fine-tune](https://docs.coreweave.com/model-distillation/reference/management/fine-tunes/retry-a-failed-fine-tune.md): Requeues a failed fine-tune with fresh credentials. If a confirmed training job can be recovered, retry resumes monitoring that job and submits only the remaining epochs. Otherwise, if no remote training model exists, retry restarts preparation. Returns 409 Conflict if the training service reported… - [Set fine-tune inference pricing](https://docs.coreweave.com/model-distillation/reference/management/fine-tunes/set-fine-tune-inference-pricing.md) - [Update fine-tune inference defaults](https://docs.coreweave.com/model-distillation/reference/management/fine-tunes/update-fine-tune-inference-defaults.md): Sets the default sampling temperature used when this fine-tune serves traffic or generates comparison outputs. Existing cached outputs remain unchanged until explicitly regenerated. - [Regenerate all comparison outputs for a fine-tune](https://docs.coreweave.com/model-distillation/reference/management/fine-tunes/regenerate-all-comparison-outputs-for-a-fine-tune.md): Deletes cached comparison outputs backed by this fine-tune across the project and queues complete regeneration using its current inference defaults. Evaluations that depend on those outputs become stale. ###### Evaluations - [List evaluations](https://docs.coreweave.com/model-distillation/reference/management/evaluations/list-evaluations.md) - [Create an evaluation](https://docs.coreweave.com/model-distillation/reference/management/evaluations/create-an-evaluation.md) - [Get an evaluation](https://docs.coreweave.com/model-distillation/reference/management/evaluations/get-an-evaluation.md) - [Update and reconcile an evaluation](https://docs.coreweave.com/model-distillation/reference/management/evaluations/update-and-reconcile-an-evaluation.md): Preserves unchanged completed cases and queues only added or invalidated work. - [Delete an evaluation](https://docs.coreweave.com/model-distillation/reference/management/evaluations/delete-an-evaluation.md) - [Retry failed evaluation cases](https://docs.coreweave.com/model-distillation/reference/management/evaluations/retry-failed-evaluation-cases.md): Creates a failed-only batch after the current evaluation has settled and preserves completed cases. - [List Weave export attempts](https://docs.coreweave.com/model-distillation/reference/management/evaluations/list-weave-export-attempts.md): Lists export attempts for an evaluation, newest first. - [Publish an evaluation to Weave](https://docs.coreweave.com/model-distillation/reference/management/evaluations/publish-an-evaluation-to-weave.md): Queues publication of evaluation records to Weave. ###### Analytics - [Get project analytics](https://docs.coreweave.com/model-distillation/reference/management/analytics/get-project-analytics.md) ###### UI preferences - [List model color preferences](https://docs.coreweave.com/model-distillation/reference/management/ui-preferences/list-model-color-preferences.md) - [Set a model color preference](https://docs.coreweave.com/model-distillation/reference/management/ui-preferences/set-a-model-color-preference.md) - [List analytics comparison models](https://docs.coreweave.com/model-distillation/reference/management/ui-preferences/list-analytics-comparison-models.md) - [Replace analytics comparison models](https://docs.coreweave.com/model-distillation/reference/management/ui-preferences/replace-analytics-comparison-models.md) ##### Inference proxy API - [Inference proxy API](https://docs.coreweave.com/model-distillation/reference/proxy.md): Send OpenAI-compatible Chat Completions requests through Model Distillation project routing. ###### Chat Completions - [Create a routed chat completion](https://docs.coreweave.com/model-distillation/reference/proxy/chat-completions/create-a-routed-chat-completion.md): Send a standard Chat Completions request. `model` accepts a task alias, a version pin such as `alias@v2`, or a direct registered-provider reference such as `openai/gpt-5.6-sol`. Task requests are traced automatically. ### Serverless Training - [About serverless training](https://docs.coreweave.com/products/post-training/serverless-training.md): Post-train and fine-tune LLMs on managed serverless infrastructure with Serverless RL and Serverless SFT. - [Available models](https://docs.coreweave.com/products/post-training/serverless-training/available-models.md): Browse the foundation models available for training with Serverless Training, including supported model variants and capabilities. - [Usage information and limits](https://docs.coreweave.com/products/post-training/serverless-training/usage-limits.md): Understand pricing, usage limits, and account restrictions for W&B Serverless Training - [Use Serverless SFT](https://docs.coreweave.com/products/post-training/serverless-training/sft.md): Fine-tune models with Serverless SFT using OpenPipe's ART framework and the Serverless Training API for supervised learning. - [Use Serverless RL](https://docs.coreweave.com/products/post-training/serverless-training/rl.md): Post-train models with Serverless RL using OpenPipe's ART framework and the Serverless Training API, then send requests to them. #### API Reference - [Serverless Training API reference](https://docs.coreweave.com/products/post-training/serverless-training/api-reference.md): Browse the Serverless Training API endpoints for chat completions, models, training jobs, and health checks with authentication details. ##### chat-completions - [Create Chat Completion](https://docs.coreweave.com/products/post-training/serverless-training/api-reference/chat-completions/create-chat-completion.md): Create a new chat completion. ##### models - [List Base Models](https://docs.coreweave.com/products/post-training/serverless-training/api-reference/models/list-base-models.md): Models this deployment can train, including architecture validation. - [Create Model](https://docs.coreweave.com/products/post-training/serverless-training/api-reference/models/create-model.md): Create a new model. - [List Model Checkpoints](https://docs.coreweave.com/products/post-training/serverless-training/api-reference/models/list-model-checkpoints.md) - [Delete Model Checkpoints](https://docs.coreweave.com/products/post-training/serverless-training/api-reference/models/delete-model-checkpoints.md): Delete specific checkpoints for a model. - [Delete Model](https://docs.coreweave.com/products/post-training/serverless-training/api-reference/models/delete-model.md): Delete a model, all its checkpoints, artifacts, and the associated W&B run. - [Log](https://docs.coreweave.com/products/post-training/serverless-training/api-reference/models/log.md): Log trajectories and calculate metrics. ##### training-jobs - [Create RL Training Job](https://docs.coreweave.com/products/post-training/serverless-training/api-reference/training-jobs/create-rl-training-job.md): Create a new RL (Reinforcement Learning) training job. - [Create SFT Training Job](https://docs.coreweave.com/products/post-training/serverless-training/api-reference/training-jobs/create-sft-training-job.md): Create a new SFT (Supervised Fine-Tuning) training job. - [Get Training Job](https://docs.coreweave.com/products/post-training/serverless-training/api-reference/training-jobs/get-training-job.md): Get a training job by ID. - [Get Training Job Events](https://docs.coreweave.com/products/post-training/serverless-training/api-reference/training-jobs/get-training-job-events.md): Get events for a training job. ##### health - [Health Check](https://docs.coreweave.com/products/post-training/serverless-training/api-reference/health/health-check.md) - [System Check](https://docs.coreweave.com/products/post-training/serverless-training/api-reference/health/system-check.md): Check health of all system components. ## OpenAPI Specs - [management.openapi](/openapi/model-distillation/management.openapi.yaml) - [proxy.openapi](/openapi/model-distillation/proxy.openapi.yaml) - [openapi](/products/post-training/serverless-training/api-reference/openapi.json)