wandb · anastasiaguspan · May 18, 2026 · May 18, 2026 · May 18, 2026 · May 18, 2026
@@ -1261,7 +1261,10 @@
                   "support/models/tags/administrator",
                   "support/models/tags/alerts",
                   "support/models/tags/anonymous",
+                  "support/models/tags/api",
+                  "support/models/tags/api-keys",
                   "support/models/tags/artifacts",
+                  "support/models/tags/authentication",
                   "support/models/tags/aws",
                   "support/models/tags/billing",
                   "support/models/tags/charts",
@@ -1273,6 +1276,7 @@
                   "support/models/tags/logs",
                   "support/models/tags/metrics",
                   "support/models/tags/notebooks",
+                  "support/models/tags/org-management",
                   "support/models/tags/outage",
                   "support/models/tags/privacy",
                   "support/models/tags/projects",
@@ -1281,13 +1285,16 @@
                   "support/models/tags/resuming",
                   "support/models/tags/run-crashes",
                   "support/models/tags/runs",
+                  "support/models/tags/sdk",
                   "support/models/tags/security",
                   "support/models/tags/storage",
                   "support/models/tags/sweeps",
                   "support/models/tags/tables",
                   "support/models/tags/team-management",
+                  "support/models/tags/teams",
                   "support/models/tags/tensorboard",
                   "support/models/tags/user-management",
+                  "support/models/tags/workspace",
                   "support/models/tags/workspaces",
                   "support/models/tags/wysiwyg"
                 ]
@@ -1393,6 +1400,80 @@
               }
             ],
             "tab": "W&B Launch"
+          },
+          {
+            "tab": "Support: W&B Models",
+            "hidden": true,
+            "pages": [
+              "support/models",
+              "support/models/tags/academic",
+              "support/models/tags/administrator",
+              "support/models/tags/alerts",
+              "support/models/tags/anonymous",
+              "support/models/tags/api",
+              "support/models/tags/api-keys",
+              "support/models/tags/artifacts",
+              "support/models/tags/authentication",
+              "support/models/tags/aws",
+              "support/models/tags/billing",
+              "support/models/tags/charts",
+              "support/models/tags/connectivity",
+              "support/models/tags/environment-variables",
+              "support/models/tags/experiments",
+              "support/models/tags/hyperparameter",
+              "support/models/tags/inference",
+              "support/models/tags/logs",
+              "support/models/tags/metrics",
+              "support/models/tags/notebooks",
+              "support/models/tags/org-management",
+              "support/models/tags/outage",
+              "support/models/tags/privacy",
+              "support/models/tags/projects",
+              "support/models/tags/python",
+              "support/models/tags/reports",
+              "support/models/tags/resuming",
+              "support/models/tags/run-crashes",
+              "support/models/tags/runs",
+              "support/models/tags/sdk",
+              "support/models/tags/security",
+              "support/models/tags/storage",
+              "support/models/tags/sweeps",
+              "support/models/tags/tables",
+              "support/models/tags/team-management",
+              "support/models/tags/teams",
+              "support/models/tags/tensorboard",
+              "support/models/tags/user-management",
+              "support/models/tags/workspace",
+              "support/models/tags/workspaces",
+              "support/models/tags/wysiwyg"
+            ]
+          },
+          {
+            "tab": "Support: W&B Weave",
+            "hidden": true,
+            "pages": [
+              "support/weave",
+              "support/weave/tags/client-info",
+              "support/weave/tags/code-capture",
+              "support/weave/tags/data-capture",
+              "support/weave/tags/evaluation",
+              "support/weave/tags/performance",
+              "support/weave/tags/system-info",
+              "support/weave/tags/trace-data",
+              "support/weave/tags/ui-rendering"
+            ]
+          },
+          {
+            "tab": "Support: W&B Inference",
+            "hidden": true,
+            "pages": [
+              "support/inference",
+              "support/inference/tags/administrator",
+              "support/inference/tags/authentication-access",
+              "support/inference/tags/billing",
+              "support/inference/tags/quotas-rate-limits",
+              "support/inference/tags/server-errors"
+            ]
           }
         ]
       },
@@ -5725,6 +5806,10 @@
       "destination": "/support/models/articles/how-do-i-change-my-billing-address",
       "source": "/models/support/change_billing_address"
     },
+    {
+      "destination": "/support/models/articles/how-do-i-connect-to-wandb-self-managed",
+      "source": "/support/models/articles/how-do-i-connect-to-a-self-hosted-wb-server"
+    },
     {
       "destination": "/support/models/articles/how-can-i-change-the-directory-my-sweep-",
       "source": "/models/support/change_directory_sweep_logs_locally"
@@ -5922,9 +6007,13 @@
       "source": "/models/support/how_can_i_disable_logging_of_system_metrics_to_wb"
     },
     {
-      "destination": "/support/models/articles/how-can-i-log-in-to-wb-server",
+      "destination": "/support/models/articles/how-do-i-log-in-to-wandb-self-managed",
       "source": "/models/support/how_can_i_log_in_to_wb_server"
     },
+    {
+      "destination": "/support/models/articles/how-do-i-log-in-to-wandb-self-managed",
+      "source": "/support/models/articles/how-can-i-log-in-to-wb-server"
+    },
     {
       "destination": "/support/models/articles/how-can-i-overwrite-the-logs-from-previo",
       "source": "/models/support/how_can_i_overwrite_the_logs_from_previous_steps"

@@ -36,24 +36,53 @@ W&B captures three types of console logs and adds a prefix to indicate each log'
 
 ## Console log settings
 
-To control which types of console output W&B captures and displays, pass a `wandb.Settings` object to `wandb.init()` when you initialize a run. The relevant parameters are `show_errors`, `show_warnings`, `show_info`, and `silent`. For details on each parameter and its default value, see the [`wandb.Settings` reference](/models/ref/python/experiments/settings).
+To control which types of console output W&B captures and displays, pass a [`wandb.Settings`](/models/ref/python/experiments/settings) object to `wandb.init()` when you initialize a run. The relevant parameters include `show_errors`, `show_warnings`, `show_info`, and `silent`. The [Settings reference](/models/ref/python/experiments/settings) lists every parameter and is generated from the SDK, so it stays current as new options are added.
 
-The following example shows how to configure these settings:
+The example below toggles whether informational messages, warnings, and errors from W&B appear in the **Logs** tab. Set `silent=True` to suppress all W&B console output (useful when you want a quiet training script):
 
 ```python
 import wandb
 
 settings = wandb.Settings(
-    show_errors=True,  # Show error messages in the W&B App
-    silent=False,      # Disable all W&B console output
-    show_warnings=True # Show warning messages in the W&B App
+    show_errors=True,
+    silent=False,
+    show_warnings=True,
 )
 
 with wandb.init(settings=settings) as run:
-    # Your training code here
     run.log({"accuracy": 0.95})
 ```
 
+For stdout and stderr capture (`console`), multipart uploads (`console_multipart`, `console_chunk_max_bytes`, `console_chunk_max_seconds`), and troubleshooting, see the sections below and the Settings reference. Distributed training and other edge cases are covered in [Why is console output not captured for my run?](/support/models/articles/why-is-console-output-not-captured-for-my-run).
+
+### Multipart console logging
+
+By default, W&B stores your script's stdout and stderr as a single `output.log` file and uploads it *when the run finishes*. While a run is active, the **Logs** tab streams output for viewing, but `output.log` does not appear on the **Files** tab until the run completes.
+
+Enable multipart console logging when you need downloadable logs while a run is still active, when a run may crash before it finishes, or when you resume a run and want to preserve log output from earlier sessions. Set `console_multipart=True` (SDK v0.22.3 or later) so the SDK writes timestamped chunks under `logs/` and uploads each chunk when it closes. Use `console_chunk_max_bytes` and `console_chunk_max_seconds` to control rollover; see the [Settings reference](/models/ref/python/experiments/settings) for defaults and behavior when both are `0`.
+
+<Note>
+Uploaded chunks are immutable. Terminal control sequences that modify previous lines (for example, progress bars that use carriage returns) only affect the current chunk.
+</Note>
+
+```python
+import wandb
+
+with wandb.init(
+    project="my-project",
+    settings=wandb.Settings(
+        console_multipart=True,
+        console_chunk_max_bytes=1_000_000,  # rotate at ~1 MB
+        console_chunk_max_seconds=60,     # or every 60 seconds, whichever first
+    ),
+) as run:
+    print("Logs upload in chunks while this run is active.")
+```
+
+You must set `console_multipart` at `wandb.init` time. Upload cadence cannot be changed after a run has started.
+
+For troubleshooting (console capture disabled, distributed training, resumed runs, and display limits), see [Why is console output not captured for my run?](/support/models/articles/why-is-console-output-not-captured-for-my-run) and [How do I download the console log file from a run?](/support/models/articles/how-do-i-download-the-console-log-file-from-a-run).
+
 ## Custom logging
 
 If you already have your own logging setup, you can continue to use it alongside W&B. W&B captures console logs from your application, but it doesn't interfere with your own logging setup. You can use Python's built-in `print()` function or the `logging` module to log messages.

@@ -288,7 +288,7 @@ Two `wandb` functions come into play here: `watch` and `log`.
 `run.watch()` will log the gradients and the parameters of your model,
 every `log_freq` steps of training.
 
-All you need to do is call it before you start training.
+Call `run.watch()` before you start training. For log modes, multiple models, and performance tips, see [How do I log gradients and model weights with wandb.watch?](/support/models/articles/how-do-i-log-gradients-and-model-weights-with-wandb-watch).
 
 The rest of the training code remains the same:
 we iterate over epochs and batches,

@@ -73,7 +73,13 @@ wandb login --host https://my-wandb-server.example.com
 wandb login --relogin
 ```
 
+## API key formats
 
+W&B issues API keys in the `wandb_v1_` format (about 92 characters). Older SDK versions expect exactly 40 characters and reject the new format with `API key must be 40 characters long`. Upgrade the SDK (`pip install -U wandb`) and run `wandb login --relogin`.
+
+Keys generated at [wandb.ai/authorize](https://wandb.ai/authorize) are shown only once at creation. Copy them immediately; you cannot retrieve them again after closing the dialog.
+
+For pinned SDK versions, legacy keys, and related errors, see [Why does my API key fail with 'must be 40 characters long'?](/support/models/articles/why-does-my-api-key-fail-with-must-be-40-characters).
 
 ## Arguments
 

@@ -97,6 +97,8 @@ On preemption, the **training process** must receive the signal, mark the run as
 
 **Multi-agent coordination:** When many agents run at once (such as SLURM array jobs), they can race to claim the same preempted run. This is a known limitation. Stagger agent startup or use external coordination mechanisms like locks to help work around this potential issue.
 
+For multi-GPU SLURM jobs where only one process should call `wandb.agent()`, see [How do I run sweeps with distributed training on SLURM?](/support/models/articles/how-do-i-run-sweeps-with-distributed-training-on-slurm).
+
 ## `wandb sweep --cancel`
 
 You cancel a sweep using the W&B API, not an OS signal. Run a command like `wandb sweep --cancel entity/project/sweep_ID`. The server tells the agent to exit, and the agent then terminates running child processes and stops. There can be a short delay (on the order of the agent's API polling interval) before cancellation takes effect.

@@ -39,7 +39,10 @@ products:
       - Administrator
       - Alerts
       - Anonymous
+      - API
+      - API Keys
       - Artifacts
+      - Authentication
       - AWS
       - Billing
       - Charts
@@ -51,6 +54,7 @@ products:
       - Logs
       - Metrics
       - Notebooks
+      - Org Management
       - Outage
       - Privacy
       - Projects
@@ -59,13 +63,16 @@ products:
       - Resuming
       - Run Crashes
       - Runs
+      - SDK
       - Security
       - Storage
       - Sweeps
       - Tables
       - Team Management
+      - Teams
       - Tensorboard
       - User Management
+      - Workspace
       - Workspaces
       - Wysiwyg
 

@@ -516,7 +516,7 @@ export const kbArticleMap = [
     "tags": [
       "user management"
     ],
-    "title": "How can I log in to W&B Server?"
+    "title": "How do I log in to W&B Self-Managed?"
   },
   {
     "page": "/models/support/how_can_i_overwrite_the_logs_from_previous_steps",

@@ -47,17 +47,17 @@ and the W&B community.
 <CardGroup cols={3}>
 <Card title="W&B Models" href="/support/models" arrow="true" icon="/icons/cropped-models.svg">
   {/* AUTO-GENERATED: counts */}
-  190 articles &middot; 33 tags
+  210 articles &middot; 40 tags
   {/* END AUTO-GENERATED: counts */}
 </Card>
 <Card title="W&B Weave" href="/support/weave" arrow="true" icon="/icons/cropped-weave.svg">
   {/* AUTO-GENERATED: counts */}
-  16 articles &middot; 8 tags
+  17 articles &middot; 8 tags
   {/* END AUTO-GENERATED: counts */}
 </Card>
 <Card title="Serverless Inference" href="/support/inference" arrow="true" icon="/icons/cropped-inference.svg">
   {/* AUTO-GENERATED: counts */}
-  7 articles &middot; 5 tags
+  9 articles &middot; 5 tags
   {/* END AUTO-GENERATED: counts */}
 </Card>
 </CardGroup>

@@ -19,5 +19,5 @@ template: "scripts/knowledgebase-nav/templates/support_product_index.mdx.j2"
   2 articles
 </Card>
 <Card title="Server Errors" href="/support/inference/tags/server-errors" arrow="true">
-  2 articles
+  4 articles
 </Card>
@@ -0,0 +1,33 @@
+---
+title: "API error code 404 - Model not found"
+keywords: ["Server Errors"]
+---
+
+A 404 error with the message "Model not found" means the model identifier in your request does not match any model available through the W&B Inference API.
+
+## Why this happens
+
+- **Incorrect model ID**: The model name or path in your request is misspelled or formatted incorrectly.
+- **Model is not available on W&B Inference**: Not all models hosted on other providers are available through the W&B Inference API. The model you are requesting may not have been onboarded.
+- **Using the wrong base URL**: If you are pointing at a different provider's endpoint but using a W&B API key, the model ID will not resolve correctly.
+- **Model was removed**: A model that was previously available may have been deprecated or removed.
+
+## What you can do
+
+1. **Check the model ID**
+   - Verify the exact model identifier against the [W&B Inference supported models list](/inference/supported-models).
+   - Model IDs are case-sensitive and typically use the format `provider/model-name` (for example, `meta-llama/Llama-3.3-70B-Instruct`).
+
+2. **Confirm the base URL**
+   - The W&B Inference base URL is `https://api.wandb.ai/v1`. Ensure your client is pointed at this URL and not another provider's endpoint.
+
+3. **Request access to a new model**
+   - If the model you need is not currently supported, contact [W&B support](mailto:support@wandb.com) to request it be added.
+
+For more information, see [Serverless Inference](/inference) and the [list models API reference](/inference/api-reference/list-models).
+
+---
+
+{/* AUTO-GENERATED: tab badges */}
+<Badge stroke shape="pill" color="orange" size="md">[Server Errors](/support/inference/tags/server-errors)</Badge>
+{/* END AUTO-GENERATED: tab badges */}
@@ -0,0 +1,35 @@
+---
+title: "API error code 422 - Invalid request parameters"
+keywords: ["Server Errors"]
+---
+
+A 422 error means the request was understood by the server but could not be processed because one or more parameters are invalid, missing, or out of range.
+
+## Why this happens
+
+- **Unsupported parameter for the model**: Some parameters (such as `frequency_penalty`, `logprobs`, or `response_format`) are not supported by all models. Passing an unsupported parameter returns a 422.
+- **Parameter value out of range**: Values like `temperature` must fall within a valid range (typically 0–2). Passing a value outside that range causes a 422.
+- **Malformed messages array**: The `messages` field must be a list of objects with `role` and `content` keys. Missing or incorrectly typed fields return a 422.
+- **Invalid `response_format`**: Requesting JSON mode (`response_format: {"type": "json_object"}`) with a model that does not support structured outputs causes a 422.
+
+## What you can do
+
+1. **Check the error message body**
+   - The 422 response body contains a `detail` field that identifies which parameter is invalid and why. Read it before troubleshooting.
+
+2. **Verify parameter support for your model**
+   - Review the [W&B Inference supported models list](/inference/supported-models) for the specific parameters and ranges each model accepts.
+
+3. **Validate your messages array**
+   - Each message must have a `role` (`system`, `user`, or `assistant`) and a `content` string. Tool call messages require additional fields—consult the API reference for the correct schema.
+
+4. **Remove unsupported parameters**
+   - If you are adapting code from another provider, remove any parameters that are not in the W&B Inference API reference. Extra parameters that the model does not support will trigger a 422.
+
+For more information, see [Serverless Inference](/inference) and the [chat completions API reference](/inference/api-reference/chat-completions).
+
+---
+
+{/* AUTO-GENERATED: tab badges */}
+<Badge stroke shape="pill" color="orange" size="md">[Server Errors](/support/inference/tags/server-errors)</Badge>
+{/* END AUTO-GENERATED: tab badges */}
@@ -1,10 +1,16 @@
 ---
 title: "Server Errors"
-tag: "2"
+tag: "4"
 generator: "knowledgebase-nav"
 template: "scripts/knowledgebase-nav/templates/support_tag.mdx.j2"
 ---
 
+<Card title="API error code 404 - Model not found" href="/support/inference/articles/api-error-code-404-model-not-found" arrow="true" horizontal>
+  A 404 error with the message "Model not found" means the model identifier in your request does not match any model avail ...
+</Card>
+<Card title="API error code 422 - Invalid request parameters" href="/support/inference/articles/api-error-code-422-invalid-request-parameters" arrow="true" horizontal>
+  A 422 error means the request was understood by the server but could not be processed because one or more parameters are ...
+</Card>
 <Card title="API error code 500 - The server had an error while processing your request" href="/support/inference/articles/api-error-code-500-the-server-had-an-err" arrow="true" horizontal>
   A 500 error with the message The server had an error while processing your request indicates an internal server error in ...
 </Card>