TRL v1.13 ships long-context training: 1M+ token sequences on a single 8-GPU H100 node
SergioPaniego · x · 2026-09-13
Hugging Face's TRL v1.13 is out with support for training on sequences beyond 1M tokens, plus a full guide. The motivation: agent sessions accumulate hundreds of thousands of tokens, so models need training on equally long sequences — which don't fit even on 8 GPUs without help. The example trains exactly one million-token sequence per step on a single 8x H100 node, using gradient checkpointing offload from transformers main.
Related event: Hugging Face TRL v1.13 Enables Training on 1M-Token Sequences(2 posts)→
More from Infra
- xAI's Memphis datacenter now has its own power sources en route to $100b ARR — PaulYacoubian · 2026-09-13
- DeepSeek's V4.1-Flash KV cache compression could undercut OpenAI and Anthropic's compute moat — justlikemedics · 2026-09-13
- Google to spend $15.1B on Finnish AI infrastructure, including three data centers and 22-year nuclear deal — Beth_Kindig · 2026-09-13
- Dev picks LFM2.5-2.6B and MiniCPM5-2B as favorite edge-device models — reach_vb · 2026-09-13
- Prompt caching can inflate your LLM bill: check write premiums, TTL and actual reuse — gethackteam · 2026-09-13
- Speculative decoding: the trick behind Google's 2-3x faster LLM inference in production — hongyangzh · 2026-09-13