vLLM Ecosystem RL Framework Vime Natively Supports AMD Instinct GPU
vllm_project · x · 2026-07-17
Vime, a reinforcement learning post-training framework within the vLLM ecosystem, now natively supports the AMD platform. Through collaboration with the AMD team, Vime successfully achieved end-to-end RL post-training on AMD Instinct MI355X GPUs.
The framework uses vLLM directly as the rollout backend, eliminating the need to write dedicated ROCm code. It currently supports GRPO training, Megatron-LM training, and models like Qwen3. Benchmarks on the MI355X show that the training throughput for the Qwen3-8B model can stably reach around 4100 tokens/GPU/s.
More from Infra
- A100 rental prices still rising in 2026 per a16z data, Jevons paradox hits compute — VraserX · 2026-10-27
- Anthropic S-1 reveals $800B+ in off-books future compute commitments — GaryMarcus · 2026-10-27
- ~42% of Anthropic's ~11GW contracted compute is NVIDIA GPUs — ns123abc · 2026-10-27
- EdgeRock-4B holds KV cache at 72 MiB across 256K context, 99.1% smaller than Qwen3.5-4B — alifcoder · 2026-10-27
- RTX 5080 price jumps from $1658 to $2318 in ten days on Amazon — Frizzy-MacDrizzle · 2026-10-27
- ComfyUI on 12GB VRAM: measured peaks show offloading turns 24s into 11 minutes — Impossible-Mail-4619 · 2026-10-27