Introducing Olmo-core 3: Open, scalable training infrastructure for large MoEs | Ai2

Ai2 Research · Publicado el 1 oct 2026

Olmo-core 3 is presented as an open, scalable training infrastructure for large mixture-of-experts (MoE) models. The material reports three technical findings: maintaining throughput while scaling expert count from 8 to 128 with fixed active parameters per token; achieving ~2.7× higher training throughput than a prior FSDP-based MoE stack on eight NVIDIA B300 GPUs; and observing ~21% higher throughput and reduced peak GPU memory when using MXFP8 precision versus BF16 in a controlled benchmark. Lee 3 puntos de vista con sus evidencias y enlaces a las fuentes.

Ai2

Leer los puntos clave Abrir Editor

De un vistazo

Maintaining throughput while scaling expert count
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 4
2.7× throughput gain over prior FSDP-based MoE stack
On eight NVIDIA B300 GPUs, Olmo-core 3 achieved 52,000 tokens/sec/GPU for a 47B-parameter MoE, compared to 19,400 tokens/sec/GPU with the earlier FSDP-based implementation — a ~2.7× improvement in throughput.
Ver el momento de apoyo · Párrafo 10
MXFP8 yields 21% higher throughput and lower memory vs BF16
In a controlled benchmark on four NVIDIA B300 GPUs with uniform expert load, enabling MXFP8 across optimal system components increased end-to-end training throughput by ~21% versus BF16 baseline, while peak active GPU memory decreased from 103 GiB to 95 GiB.
Ver el momento de apoyo · Párrafo 20

Pasajes clave3

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

Párrafo 10

training infrastructure performance

2.7× throughput gain over prior FSDP-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.

Ai2·1 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 4

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%.

Ai2·1 oct 2026

Contexto

Olmo-core 3 is built to close that gap.

Párrafo 20

numerical precision optimization

MXFP8 yields 21% higher throughput and lower memory vs BF16

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.

Ai2·1 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. Most of the gain came from feed-forward computation and moving data between experts rather than attention alone.

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