Release 2.6.0
Summary
This release introduces experimental batch search and file output capabilities, enabling high-throughput MSA generation workflows. It also raises the maximum input sequence length from 4,096 to 10,240 characters.
New Features
Batch Search API (Experimental)
All three search endpoints now accept batch inputs when the NIM is started with NIM_EXPERIMENTAL=1:
- Monomer and template search: Pass
sequenceas a list of strings instead of a single string - Paired search: Pass
sequencesas a list of lists (each inner list is one complex) - Per-item status: The response wraps individual results in a
resultsarray; failed items do not affect other items in the batch - Configurable limit: Maximum batch size is controlled by
NIM_MSA_API_MAX_BATCH_SIZE(default: 300)
Refer to Batch Search (Experimental) for endpoint details and examples.
File Output Mode (Experimental)
When NIM_EXPERIMENTAL=1 and NIM_MSA_OUTPUT_DIRECTORY are set, the NIM writes alignment files to disk for batch requests and returns file:// URI references instead of inline content. Non-batch responses remain inline. Files are written to a temporary location and then renamed into place using POSIX rename(). On filesystems that provide atomic rename semantics (local disk, NFS), readers do not observe partial content; verify that your storage backend supports POSIX rename semantics if atomicity is required.
Refer to File Output Mode for configuration details.
Increased Maximum Sequence Length
The maximum input protein sequence length is raised from 4,096 to 10,240 characters. This applies to all search endpoints and is controlled by NIM_MSA_API_MAX_SEQUENCE_LENGTH (default: 10240).
Migration Notes
No breaking changes. All experimental features are opt-in and require NIM_EXPERIMENTAL=1. The increased sequence-length limit is backward compatible.