Is this page helpful?

Release Notes#

Release 2.0.0#

Summary#

This release upgrades GenMol to the NV-GenMol-89M-v2 checkpoint and updates the NIM documentation, validation matrix, and API contract for the v2 service.

Model and Platform Updates#

  • Updated the GenMol NIM to version 2.0.0

  • Integrated the NV-GenMol-89M-v2 checkpoint

  • Validated GPU matrix expanded to 13 SKUs:

    • Ampere: A10G, A100

    • Ada Lovelace: L40S, RTX6000 Ada

    • Hopper: H100, H200

    • Blackwell: B200, B300, RTX6000 Blackwell

    • Grace Hopper: GH200

    • Grace Blackwell: GB200, GB300

    • DGX Spark: GB10

Generation Quality#

  • Motif-extension wall-time on H100 improved from 1.82s in v1 to 1.051s in v2 — a reduction of approximately 42%, at comparable scaffold-decoration latency

  • Scaffold-decoration validity on H100 improved from 0.967 in v1 to 0.995 in v2

  • Higher output diversity on fragment-completion tasks

  • Substantially improved success rate on one-step linker design, a v2 capability where v1’s success rate was prohibitively low

For the full per-metric H100 v1 vs v2 comparison (wall-time and accuracy across both benchmark tasks), refer to the Benchmarks page.

API and Runtime Updates#

  • temperature default remains 1.0

  • noise default is 1.0 with supported range 0.0-2.0

  • step_size is deprecated in v2 and is accepted only for backward compatibility

  • Startup guidance now includes persistent cache mounting for model artifacts

Bug Fixes#

  • Fixed incorrect version metadata responses for the v2 service

  • Fixed plain SMILES request handling

  • Fixed under-delivery when invalid generations were filtered out

  • Improved uniqueness behavior for generated molecules

  • Improved fragment preservation for guided generation

  • Mitigated recursion-depth failures under concurrent sustained load by serializing access to the shared in-process model instance

Telemetry Control#

NIM Telemetry helps NVIDIA deliver a faster, more reliable experience with greater compatibility across a wide range of environments, while maintaining strict privacy protections and giving users full control.

Benefits:

  • Enhances performance and reliability with anonymous system and NIM-level insights that help identify bottlenecks, tune performance across hardware configurations, and improve runtime stability

  • Improves compatibility across deployments by helping detect and resolve version, driver, and environment compatibility issues earlier

  • Accelerates troubleshooting and bug resolution, enabling faster diagnosis of errors and regressions

  • Informs future optimizations and releases using aggregated real-world deployment insights

  • Protects user privacy by collecting minimal, anonymous metadata (for example, hardware type and NIM version); user prompts and generated molecule outputs are not collected

  • Is fully optional and configurable; telemetry collection is disabled by default

Configuration:

  • Set NIM_TELEMETRY_MODE=0 to disable telemetry (default)

  • Set NIM_TELEMETRY_MODE=1 to enable telemetry

For more information about data privacy, what is collected, and telemetry configuration, refer to:

Release 1.0.1#

Summary#

This release includes security updates and CVE fixes to improve the overall security posture of the NIM.

Security Updates#

  • Fixed multiple Common Vulnerabilities and Exposures (CVE) issues in dependencies

  • Updated container base images to address known security vulnerabilities

  • Enhanced security scanning and validation processes

Bug Fixes#

  • Resolved potential security vulnerabilities in third-party dependencies

  • Improved error handling for edge cases

Release 1.0.0#

Summary#

This is the first general release of NIM.

Limitation#

  • GenMol NIM uses only one GPU device. If multiple devices are visible to the NIM container, only the first one will be used.