v26.08.01

Added

• Kubernetes LLM inference recipes for B200 (inference/k8s_recipes/):
  • GLM-5 FP8
  • Nemotron Ultra NVFP4
  • Qwen3 0.6B KVBM

Changed

• Reorganized inference recipes into a unified inference/ structure: Slurm serving recipes under inference/slurm_recipes/ and Helm charts under inference/tools/single_node_helm/.

Fixed

• DeepSeek V3 nsys profiling on B200 now uses the 26.06.01 container default instead of the pinned version.

v26.08.00

Added

• FP8 support for Qwen3 235B and 30B pretrain on GB300, GB200, B300, and B200.
• Qwen3 235B FP8 in the Exemplar suite on GB300, GB200, B300, and B200.
• MLPerf pretrain recipes for GPT-OSS 20B and Llama 3.1 8B, and an MLPerf Llama 2 70B LoRA finetune recipe.
• llmb-collector system report microbenchmark (opt-in).
• llmb-run submit --mbridge-arg passthrough for Megatron-Bridge workload arguments.
• AWS EFA container guidance for verifying and updating EFA support.

Changed

• Renamed the project to Exemplar Performance; the clone URL is now https://github.com/NVIDIA/exemplar-performance.git.
• Updated recipes to NeMo 26.06.01 where applicable, including DeepSeek V3, GPT-OSS 120B, Llama 3.1, Llama 3 LoRA finetune, Nemotron 3, Qwen3, and the NCCL and system info microbenchmarks. Nemotron-H remains on NeMo 26.04.01.
• Llama 3 LoRA finetune now runs in offline mode to match other recipes.

Removed

• FP8 for Nemotron 3 Super 120B on B200.
• H100 support from the Kimi-K2 recipe.
• NVFP4 for DeepSeek V3 on B200.

Known Issues

• EFA limitations remain for DeepSeek V3 (Megatron-Bridge H100, TorchTitan, RL), Nemotron-H, and Qwen3 (30B H100, 235B H100); see Known Issues section of README for details.
• Optional PCT fixed-core CPU binding may improve select workloads on B300 Granite Rapids systems where PCT is enabled. Validate the cluster's PCT core layout before enabling ENABLE_PCT_BINDING=true or applying the patch described in the README.

End of Support

• LLMB v26.05.x and earlier are no longer supported as of v26.08.00. These release lines will not receive further updates, fixes, or support.

v26.06.01

Fixed

• Removed non-existent Nemotron 3 120B NVFP4 GB300 configuration from Exemplar.
• Enabled MLA up-projection recomputation for DeepSeek V3 GB200 FP8 proxy runs to resolve OOM issue.
• Nemotron 3 launch configuration now preserves externally supplied overrides.

v26.06.00

Added

• Agentic inference recipes for GLM-5 and Kimi-K2.6 short-context TRT-LLM Dynamo, plus Qwen3 long-context SGLang. These recipes are not part of the Exemplar suite at this time.
• DeepSeek V3 NeMo-RL GRPO workload for GB200.
• DeepSeek V3 NVL72 GB300 pretrain configuration and 72-GPU weak scaling support for GPT-OSS pretrain.
• llmb-install support for HuggingFace repository downloads into LLMB-managed workload or shared directories, per-architecture uv binaries, and SLURM-based compute-node architecture detection.
• llmb-run NeMo-RL performance metrics in job history, including s/iter and Tokens/s/GPU.
• Workload display names and aliases in metadata, with uniqueness validation and Beacon publishing support.

Changed

• Updated DeepSeek V3, Llama 3.1, and Nemotron-H Megatron-Bridge recipes to NeMo 26.04.01 where applicable.
• Refreshed Qwen3, Nemotron 3, Kimi-K2, B200, and NCCL recipe configurations.
• Optional B300 Priority Core Turbo binding is now controlled with ENABLE_PCT_BINDING=true.

Fixed

• Improved system info HCA checks for NVSHMEM workloads.

Known Issues

• EFA limitations remain for DeepSeek V3 (Megatron-Bridge H100, TorchTitan, RL), Nemotron-H, and Qwen3 (30B H100, 235B H100); see Known Issues section of README for details.
• Optional PCT fixed-core CPU binding may improve select workloads on B300 Granite Rapids systems where PCT is enabled. Validate the cluster's PCT core layout before enabling ENABLE_PCT_BINDING=true or applying the patch described in the README.

Removed

• DeepSeek R1 inference recipes (Dynamo, SGLang, TRT-LLM), GPT-OSS inference recipes (Slurm, Kubernetes), and Llama 3.3 inference recipe.

End of Support

• LLMB v26.02.x and earlier are no longer supported as of v26.06.00. These release lines will not receive further updates, fixes, or support.

v26.05

Added

• Kimi K2 MXFP8 pretrain support.
• Nemotron 3 Nano (30B) and Super (120B) pretrain recipes.
• Slurm topology checks and CPU governor reporting in the system info microbenchmark.
• llmb-run job history and log handling.
• llmb-run flags: --env for container env overrides, additional Slurm pass-through flags, and dump-env Megatron-Bridge mode.

Changed

• Updated recipes to NeMo 26.04.00 where applicable.
• Refreshed DeepSeek V3, Nemotron 3, and Qwen3 configurations.

Fixed

• Legacy-parser grad-norm NaN handling.
• Archive exclusion for nsys_profile and PyTorch profiling output directories.
• Torchtitan container compatibility.

Removed

• Deprecated Grok1 and Nemotron4 recipes.
• Legacy setup_script installer path and Conda support.
• Deprecated llmb-run commands.

Known Issues

• DeepSeek V3 Megatron-Bridge on H100 requires uv <=0.9.28 during setup.
• EFA limitations remain for DeepSeek V3 (Megatron-Bridge H100, TorchTitan) and Qwen3 (30B H100, 235B H100); see Known Issues section of README for details.
• Optional PCT fixed-core CPU binding may improve select workloads on Granite Rapids systems where PCT is enabled. See the README Known Issues section before applying the patch.

End of Support

• LLMB v25.12.x and earlier are no longer supported as of v26.05.00. These release lines will not receive further updates, fixes, or support.

v26.02.01

Added

• Llama3 LoRa finetuning support for B300 and B200.
• PyTorch Profiler support for selected Megatron-Bridge recipes, including DeepSeek V3, GPT-OSS 120B, Llama3.1, Nemotron-H, Qwen3, and Llama3 LoRa finetuning.

Changed

• Updated recipes to NeMo 26.02.01 where applicable.
• Refreshed Blackwell recipe configurations, including GPT-OSS 120B, Qwen3, and Llama3.1.

Fixed

• Improved llmb-install reliability when resuming failed installs, creating virtual environments, and auto-detecting SLURM GRES on heterogeneous partitions.
• Improved llmb-run submit validation and error messages for explicit workload selections.

Known Issues

• Qwen3 on select B300 Granite Rapids systems may benefit from the optional qwen3/pretrain/b300_numa_cpu_pinning.patch workaround when PCT is available and enabled.
• EFA incompatibility for certain recipes, see Known Issues section of README
  for more details.

v26.02

Added

• B300 support
  • Pretrain recipes: Llama 3.1, DeepSeek V3, Nemotron-H, Qwen3
  • NCCL benchmark
  • CPU overhead microbenchmark
• GPT-OSS pretrain recipe.
• DeepSeek V3 Torchtitan FP8 support for GB300 and GB200.
• DeepSeek V3 proxy models for 64 GB300/GB200 GPUs.
• System info script for IB, container, and enroot diagnostics.
• llmb-run archive command to package experiment logs into tarball.
• Exemplar program documentation and tooling.

Changed

• Updated recipes to NeMo 26.02.00 where applicable.
• Llama3 LoRa finetuning ported to Megatron Bridge.
• Torchtitan optimizations for DeepSeek V3.
• Centralized peak throughput (TFLOP/GPU) as primary performance metric in READMEs.
• Qwen3 235B GB200 removed FP8 support.

Removed

• Run:ai support.

Known Issues

• Recipes using NeMo 26.02.00 container will not work with EFA, see Known Issues section of README for workaround.
• DeepSeek V3 on EFA clusters may encounter connectivity issues.

v25.12.02

Fixed

• Pin uv to <=0.9.28 in install.sh to avoid strict parsing failures when installing pinned nemo_run commits with uv 0.9.29+.

v25.10.02

Fixed

• Pin uv to <=0.9.28 in install.sh to avoid strict parsing failures when installing pinned nemo_run commits with uv 0.9.29+.

v25.08.02

Fixed

• Pin uv to <=0.9.28 in install.sh to avoid strict parsing failures when installing pinned nemo_run commits with uv 0.9.29+.