Miles runs on NVIDIA H/B-series and AMD MI300X with the same launch scripts. Each platform page covers driver versions, build flags, and the FP8 / ROCm quirks you need to know before kicking off a job.Documentation Index
Fetch the complete documentation index at: https://www.radixark.com/llms.txt
Use this file to discover all available pages before exploring further.
NVIDIA GPUs
The default GB300 / GB200 / B200 / B100 / H200 / H100 with FP8, NVLink, and InfiniBand.
AMD GPUs
ROCm 6.3+ with patches for virtual memory management. Same launch scripts.
Supported features by GPU
| Feature | H / B-series | A100 | MI300X |
|---|---|---|---|
| BF16 training | ✅ | ✅ | ✅ |
| FP8 GEMM | ✅ | - | ✅ |
| INT4 W4A16 QAT | ✅ | ✅ | ⚠️ |
| Speculative decoding | ✅ | ✅ | ✅ |
| Miles Router (R3) | ✅ | ✅ | ✅ |
| P2P weight transfer (RDMA) | ✅ IB / RoCEv2 | ✅ | ✅ Infinity Fabric |
| Megatron CP | ✅ | ✅ | ✅ |
| Deterministic inference | ✅ | ✅ | ⚠️ |
Storage and network
Independent of GPU vendor, you’ll want:- Shared filesystem for multi-node — NFS, GPFS, Lustre. Reads dominate writes during training; provision read bandwidth.
- High-bandwidth interconnect — IB (NVIDIA), RoCEv2 (NVIDIA), Slingshot, or Infinity Fabric (AMD). 200+ GB/s per node is typical for trillion-param training.
- NVMe local scratch for SGLang radix cache and Ray spill — at least 1 TB.