Introducing Olmo-core 3: Open, scalable training infrastructure for large MoEs

Hugging Face Blog · Publicado el 1 oct 2026

Kyle Wiggers’ post on Hugging Face reports Ai2’s own Olmo-core 3 benchmarks. It describes scaling the expert pool with roughly fixed active parameters, a preliminary eight-GPU throughput comparison with the earlier stack, and a controlled MXFP8 comparison with BF16. These are reported benchmark results from the post, not independent measurements. Lee 3 puntos de vista con sus evidencias y enlaces a las fuentes.

Kyle Wiggers

Leer los puntos clave Abrir Editor

De un vistazo

Maintaining throughput while scaling expert count
The post reports that increasing the expert pool from 8 to 128 while selecting only four experts per token kept active parameters per token roughly fixed at ~3.2B; total parameter capacity grew from 4.6B to 47B with less than 5% drop in training throughput.
Ver el momento de apoyo · Párrafo 5
2.7× throughput gain with DDP-based MoE stack
The post reports that on eight NVIDIA B300 GPUs, Olmo-core 3’s new distributed data parallelism (DDP)-based MoE training stack achieved 52,000 tokens/sec/GPU versus 19,400 tokens/sec/GPU with the prior FSDP-based implementation — a ~2.7× throughput improvement.
Ver el momento de apoyo · Párrafo 11
MXFP8 improves throughput and peak active memory in a controlled B300 benchmark
The post reports that in a controlled benchmark on four NVIDIA B300 GPUs with uniform expert load, enabling MXFP8 in performance-critical parts increased end-to-end training throughput by ~21% over BF16 baseline and reduced peak active memory from 103 GiB to 95 GiB; most gains came from feed-forward computation and inter-expert data movement, not attention alone.
Ver el momento de apoyo · Párrafo 21

Pasajes clave3

Pasajes atribuidos con contexto para verificarlos. Abra el texto original para comprobar la fuente.

Párrafo 11

training infrastructure performance

2.7× throughput gain with DDP-based MoE stack

Extracto original

In a preliminary test on eight NVIDIA B300 GPUs, a 47-billion-parameter MoE processed 52,000 tokens per second per GPU with the new stack, compared with 19,400 using our earlier implementation—about 2.7× the throughput.

Kyle Wiggers·1 oct 2026

Contexto

NVIDIA’s Megatron-Core is an established option for training large MoEs. Olmo-core 3 brings an integrated MoE training stack to the framework behind Olmo, with a redesign that improves throughput over our earlier FSDP-based implementation.

Párrafo 5

MoE scaling efficiency

Maintaining throughput while scaling expert count

Extracto original

In one benchmark, we increased the expert pool from 8 to 128 while still selecting only four experts per token – the small units of text a language model processes – keeping the number of active parameters per token roughly fixed at about 3.2B. Total parameter capacity grew from 4.6B to 47B, while training throughput fell by less than 5%.

Kyle Wiggers·1 oct 2026

Contexto

Olmo-core 3 is built to close that gap.

Párrafo 21

numerical precision optimization

MXFP8 improves throughput and peak active memory in a controlled B300 benchmark

Extracto original

With MXFP8 enabled across the parts of the system where it helped most, training throughput was about 21% higher than with BF16, the higher-precision format we used as our baseline, while peak active memory fell from 103 GiB to 95 GiB. Most of the gain came from feed-forward computation and moving data between experts rather than attention alone.

Kyle Wiggers·1 oct 2026

Contexto

We measured MXFP8’s effect on end-to-end training throughput in a controlled benchmark on four NVIDIA B300 GPUs, with work distributed uniformly across experts.

Fuente y metodología

Estas perspectivas enlazan a sus fuentes originales. Las paráfrasis están identificadas y no son citas textuales.

Abrir transcripción o material de origen Reportar un problema