| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| README.md | 2026-09-15 | 3.2 kB | |
| v1.40.1 source code.tar.gz | 2026-09-15 | 6.9 MB | |
| v1.40.1 source code.zip | 2026-09-15 | 9.1 MB | |
| Totals: 3 Items | 16.0 MB | 0 | |
Release Notes
Features
- Add JAEC native inference pipeline — Introduced a native inference pipeline for Joint Acoustic Echo Cancellation (JAEC) under the audio/AEC module.
- Add FLA-TF-Locoformer and FLA-T-SepReformer separators — Added two new speech separation models (FLA-TF-Locoformer and FLA-T-SepReformer) for 2-speaker 8 kHz audio, including shared attention layers, pipeline registration, test cases, and ONNX export support.
- Add Agent IDP and MCP/Studio OpenAPI — Added SDK and CLI compatibility shims for agent identity management (key rotation, status, signed token issuance) and new
deploy_mcp_server/undeploy_mcp_servermethods toMCPApi.
Fixes
- Resolve OpenEuler build and x86 dependency issues for vLLM Ascend — Added the
patchpackage to build dependencies for both Debian-based and OpenEuler-based Ascend Docker images, and added the PyTorch CPU wheel index toPIP_EXTRA_INDEX_URLto fix dependency resolution on x86 hosts. - Unified upload environments — Unified the upload environment variables and added custom README content for uploading AIGC model weights.
- Fix constants issue — Resolved a constants-related issue in the MCP/Studio delegation logic.
Enhancements
- Support configurable source installation of mcore-bridge in Ascend Docker image — The Ascend Docker image now supports building mcore-bridge from source with configurable branch/version, improving flexibility for Megatron-based training environments.
- Update modelscope-hub version — Updated the dependency on modelscope-hub to a newer version, bringing in CLI unification, auth fixes, and endpoint normalization improvements.
- MCP API consistency — Routed MCP discovery, operational-server listing, and server-detail requests through
modelscope-hubfor consistent authentication, retry, and error handling; consolidated Studio CLI as a re-export of the hub implementation.
What's Changed
- fix: resolve OpenEuler build and x86 dependency issues for vLLM Ascend by @addsubmuldiv in https://github.com/modelscope/modelscope/pull/1783
- [Fix] Unified upload envs by @wangxingjun778 in https://github.com/modelscope/modelscope/pull/1787
- Support configurable source installation of mcore-bridge in Ascend Docker image by @addsubmuldiv in https://github.com/modelscope/modelscope/pull/1790
- feat(audio/aec): add JAEC native inference pipeline by @echocatzh in https://github.com/modelscope/modelscope/pull/1792
- feat(audio/separation): Add FLA-TF-Locoformer and FLA-T-SepReformer separators by @Mashiro009 in https://github.com/modelscope/modelscope/pull/1791
- [Feature] Add Agent IDP and mcp/studios openapi by @wangxingjun778 in https://github.com/modelscope/modelscope/pull/1788
- [Feature] Update modelscope-hub version by @wangxingjun778 in https://github.com/modelscope/modelscope/pull/1794
- Fix constants issue by @wangxingjun778 in https://github.com/modelscope/modelscope/pull/1796
New Contributors
- @echocatzh made their first contribution in https://github.com/modelscope/modelscope/pull/1792
Full Changelog: https://github.com/modelscope/modelscope/compare/v1.40.0...v1.40.1