NeuReality Software Development Kit V1.3 Release Notes
Release Date
June 30, 2025
Target Audience
This is a NeuReality SDK v1.3 release.
The use of the SDK is subject to the Early Access Program Agreement (or similar written instrument) executed by and between your organization and NeuReality (“License Agreement”).
What’s new in this release?
AI Models
Meta/Llama-3.1-70B
Meta/Llama-3.1-8B
Meta/Meta Llama-3.3-70B
Meta/Llama-Guard-8B
QwenLM/Qwen2.5-Coder-32B
Mistralai/Mixtral-8x7B
DeepSeek-AI/DeepSeek-R1-Distill-Llama 8B
AI Libraries
Added support for popular agentic and RAG libraries:
LlamaIndex - NR1 inference solution can be used as a backend for LlamaIndex library.
LangChain - NR1 inference solution can be used as a backend for LangChain library.
OpenAI-compatible API
Moderation API - Moderation endpoint support is added.
Deployment SDK (Toolchain)
Improved error logging - Extended logging capabilities are added. Now logging regime can be configured through YAML file or YAML override file (incremental). Additionally, the log overrides can be applied via NR_LOG_OVERRIDES environment variable. Information on the design, structure, components and usage can be found in the Deployment SDK documentation - Logging section.
Management and BMC
BMC UI
Topology view - A part of BMC UI, shows all the components comprising NR1 inference appliance including detailed info and status of each component. For more information, please refer to the Developer Portal -> Documents -> BMC User’s Guide document.
Certificate deployment - The screen enables to manage certificates used for HTTPs secure connection over the management port. For more details please refer to the corresponding BMC User guide.
BMC toolbox - Added a user-friendly toolbox for basic BMC delivery package with useful wrapper scripts for IPMI tool. These wrappers enable running complex frequently used operations with a single command line, e.g., commands in batch mode applied to multiple cards and remote firmware upgrade.
Persistent provisioning - The purpose of the persistent provisioning feature is to allow NR1 to recover the last provisioning configuration after a reboot without requiring manual intervention. NR1 stores the provisioned metadata persistently, ensuring that on startup, the system re-provisions the models from the most optimal source (e.g. from cache if available or from the original remote storage) automatically.
Provisioning models from NFS - The NR1 provisioning process supports model downloading by using NFS URI in addition to local provisioning and provisioning from S3 supported earlier.
Networking
Network segmentation support with VLAN ID - Added VLAN tagging capabilities to support traffic segregation, QoS management, and secure network partitioning. This feature enables the system to process, maintain, and propagate IEEE 802.1Q VLAN tags across network interfaces while preserving traffic priorities and enforcing security boundaries. VLAN tagging allows multiple logical networks to coexist on the same physical infrastructure by adding a tag to each Ethernet frame that identifies to which VLAN it belongs. The NR1 system supports standard VLAN operations.
REST serving endpoints - Added support for REST serving endpoints operating with OpenAI-compatible API.
Delivery Packages and Upgrades
NeuReality SDK is delivered as follows:
Software Package |
Delivery |
|---|---|
NR1 Appliance Platform Firmware |
|
NR1 Module AI Platform Firmware |
|
NR1 Appliance BMC Firmware |
|
Toolchain library (Deployment SDK) |
|
Client library (Serving API supporting AI over Fabric protocol, not required while using OpenAI-compatible API) |
|
Kubernetes plugins |
Documentation
Please visit the NeuReality Developer Portal for more information about NeuReality SDK, tutorials, API reference, videos, and events.
Known Limitations
# |
Description |
Workaround |
Temporary Limitation |
|---|---|---|---|
AV-4373 |
Auto-negotiation is frequently used in Ethernet networks to find the most efficient mode of operation for two linked devices. NR1 module auto negotiation feature is not functional in the present release. |
The settings controlled by auto-negotiation (such as port speed and flow control configurations) should be manually configured to the desired values using the remote management interface (IPMI). |
Yes |
AV-4373 |
Dual port is not supported in NR1 card. Single port mode is supported. |
NA |
NA |
AV-4183 |
NR1 card supports an Ethernet port (port0) with speeds of 25Gbps and 50Gbps. |
NA |
NA |
AV-4181 |
Once the NR1 module network interface is up, it cannot be brought down without an NR1 module reboot. |
Changing the network MTU which requires bringing down the network interface requires rebooting NR1 module to finalize the MTU change. |
NA |
ASAI-2304 |
PFC support |
Priority flow control (PFC) is not supported on NR1 module network port. Avoid enabling this feature. |
Yes |
AV-3829 |
Large receive offload (LRO) is enabled for connections where the total header length, including the TCP constant header, exceeds 63 bytes. |
NA |
Yes |
ASAI-7553 |
Non-contiguous arrays are not supported in NeuReality programming model and Deployment SDK. Using non-contiguous arrays will not be flagged as an error by the compiler, leading to run time errors during simulation or device execution. |
User code must avoid creating non-contiguous arrays. Input and output arrays submitted to the Deployment SDK should be contiguous in memory. For details on generating a contiguous array, please refer to the following: numpy.ascontiguousarray — NumPy v2.0 Manual |
Yes |
ASAI-10473 |
TopK runtime parameter is limited to 8K. |
NA |
NA |
How to Get Help
Need a Hand? We’ve Got You Covered. Create a ticket with us, and let’s tackle those challenges together. Use one of the following channels to raise the issue: