֍
All additions and updates to the Substrate API, in reverse-chronological order.
Your API and SDK version controls the API behavior you see. Upgrade to the latest SDK version for new nodes and the latest functionality.
Aug 4, 2024
PyPI 220240617.1.7 (opens in a new tab) | npm 120240617.1.8 (opens in a new tab)
- You can now use OpenAI and Anthropic (GPT-4o and Claude 3.5 Sonnet) using Substrate. First, add your OpenAI or Anthropic API key when initializing Substrate (opens in a new tab), and then select a model using the model (opens in a new tab) parameter.
- All Llama 3 models have been upgraded to Llama 3.1. We've also added Meta's most advanced language model, Llama 3.1 405B. To use these models, select them using the model (opens in a new tab) parameter.
- Added a new node:
StableVideoDiffusion(opens in a new tab). This node takes an image as input, and generates a short (4 second) animated GIF or video.
July 18, 2024
PyPI 220240617.1.3 (opens in a new tab) | npm 120240617.1.5 (opens in a new tab)
- Added a new node:
SplitDocument(opens in a new tab). This node parses and splits up text and code documents into related chunks of text. We've found this especially useful for processing large documents before embedding them. - We've also introduced a handful of new options that can be specified on nodes:
cache_agecaches the result of this node for the number of seconds specified. This can speed up subsequent runs of the graph.cache_keysare a set of node input arguments to use when computing the cache key forcache_agemax_retriesspecifies the number of times a node should be retried on error. This is useful for handling transient errors.
June 20, 2024
- Added
IfandBox - Renamed
GenerateText->ComputeText - Renamed
MultiGenerateText->MultiComputeText - Renamed
BatchGenerateText->BatchComputeText - Renamed
GenerateJSON->ComputeJSON - Renamed
MultiGenerateJSON->MultiComputeJSON - Renamed
BatchGenerateJSON->BatchComputeJSON
June 12, 2024
PyPI 220240612.0.0 (opens in a new tab) | npm 120240612.0.0 (opens in a new tab)
- The TypeScript SDK now supports streaming (opens in a new tab). Streaming support in the Python SDK is coming soon.
RunPython(opens in a new tab) is generally available in the Python SDK. Use it to run your own locally-defined functions in a remote sandbox. There is no equivalent in the TypeScript SDK, but it is on our roadmap.BatchGenerateJSON(opens in a new tab) has a newmodelparameter for switching betweenMistral7BInstructandLlama3Instruct8B.- Removed
GenerateTextVision– callGenerateText(opens in a new tab) with the optionalimage_urisparameter instead. - Added
system_prompt,frequency_penalty,repetition_penalty,presence_penalty, andtop_pto:Mistral7BInstructMixtral8x7BInstructLlama3Instruct8BLlama3Instruct70B
- Renamed
CreateVectorStoretoFindOrCreateVectorStore - Renamed
GenerativeEditImagetoInpaintImage - Renamed
MultiGenerativeEditImagetoMultiInpaintImage - Renamed
FillMasktoEraseImage - Renamed
TranscribeMediatoTranscribeSpeech - Renamed the
nodeparameter tomodelin:GenerateText,MultiGenerateText,BatchGenerateTextGenerateJSON,MultiGenerateJSON,BatchGenerateJSON