base_url | None | Override the Perceptron endpoint (e.g., edge gateway). |
api_key | None | API key to send with requests. |
provider | None → auto-detected ("fal" when relevant) | Transport/provider name. |
model | None | Preferred model ID if your provider exposes multiple versions. |
timeout | 60.0 | Per-request timeout in seconds. |
retries | 3 | Automatic retry attempts on transient errors. |
strict | False | Enforce strict parsing (raises on schema mismatches). |
allow_multiple | False | Allow multiple responses when the task permits it. |
warn_on_implicit_anchor | True | Emit warnings if prompts rely on implicit spatial anchors. |
temperature | 0.0 | Generation temperature for text answers. |
max_tokens | 1024 | Token cap for generations. |
top_p | 1.0 | Nucleus sampling parameter. |
top_k | None | Limits sampling to the top-k candidates when set. |
max_buffer_bytes | None | Caps streamed buffer size (useful for long outputs). |
resize_max_side | None | Downscale images so the longest side matches this value before upload. |
auto_coerce_paths | False | Automatically open local file paths passed into SDK helpers. |