Beirit Harvey

Telecommunications doesn’t need another generic chatbot. It needs systems that can reason over alarms, traces, standards, configuration drift, and operational runbooks accurately, repeatably, and under real-world constraints. That’s the intent behind OTel: a new open model suite engineered around network-domain work, not broad internet tasks.

OTel: Proven at Scale on TensorWave

Telecommunications doesn’t need another generic chatbot. It needs systems that can reason over alarms, traces, standards, configuration drift, and operational runbooks accurately, repeatably, and under real-world constraints. That’s the intent behind OTel: a new open model suite engineered around network-domain work, not broad internet tasks.

The release spans thirty purpose-trained components, covering three core functions:

 * Generative models for telecom-specific understanding and assistanc

Telecommunications doesn’t need another generic chatbot. It needs systems that can reason over alarms, traces, standards, configuration drift, and operational runbooks accurately, repeatably, and under real-world constraints. That’s the intent behind OTel: a new open model suite engineered around network-domain work, not broad internet tasks.

The release spans thirty purpose-trained components, covering three core functions:

 * Generative models for telecom-specific understanding and assistance
 * Vector models for search, retrieval, and semantic matching
 * Ranking models to improve relevance and ordering in retrieval pipelines

Across the set, sizes range from hundreds of millions to tens of billions of parameters, enabling teams to choose the right footprint for latency, cost, and accuracy requirements.


Full-visibility release, not a black box

Instead of shipping a sealed product that can only run inside a single vendor’s ecosystem, this initiative publishes the building blocks needed for auditability and repeatability. Now operators and researchers can validate behavior, reproduce results, and adapt the work for their own environments.


Hardware flexibility by design

The suite is designed to run across modern accelerator fleets without forcing a single GPU path. That matters in telecom, where infrastructure is heterogeneous and long-lived.


How TensorWave Made the Proof Real

Open releases only earn trust when they perform under real load. It’s one thing to publish weights and code; it’s another to prove the training stack stays efficient across a large, distributed run with sustained utilization.

That’s where TensorWave materially enabled the effort.

TensorWave supplied the compute environment used to demonstrate large-scale training efficiency on a 256-GPU AMD Instinct MI325X cluster, using ScalarLM for distributed execution. 

The outcome was a nearly 3 hour demo! It showcased that this approach can maintain strong efficiency at a scale that resembles real deployments.

What that support unlocked:

 * A practical validation of “open at scale.” Not just published weights, but demonstrated training behavior under real pressure.
 * A reference path for AMD-based clusters. High utilization on a large MI325X run signals readiness for serious workloads, not just research prototypes.
 * A reproducible baseline for the community. When performance is measurable and repeatable, teams can iterate without guessing.

In short,  TensorWave didn’t just support the effort but powered it.  We provided the platform that verifies telecom-focused models can be trained efficiently, using an open stack, on modern AMD hardware on TensorWave cloud infrastructure.


A Scaled Build

This program drew participation from over one hundred contributors across operators, infrastructure providers, and universities, spanning names such as AT&T, RelationalAI, AMD, GSMA, and multiple academic research groups. That breadth is a key part of why the release is positioned for industry credibility: it reflects many viewpoints, not a single lab’s assumptions.


A Hands-On Deep Dive in Barcelona

For teams who want to understand the mechanics, not just the headline, project leads ran a practical session in Barcelona that walks through training one of the instruction-tuned variants on the same 256-GPU MI325X setup in the TensorWave environment, covering distributed configuration, efficiency tuning, and domain-targeted alignment.

Watch the full recording of the masterclass:


TensorWave Cloud: Engineered for AI Velocity

TensorWave is the industry’s premier AMD-exclusive cloud provider, delivering memory optimized AI infrastructure for training, fine-tuning and inference. 

Our platform combines the raw computational power of AMD's latest GPU architectures with innovative liquid cooling technology and ultra-high-speed networking, creating an optimized environment for the most demanding AI workloads. From large language model training to real-time inference, TensorWave provides the performance, scalability, and cost-efficiency that modern AI initiatives demand.

Ready to get started? Connect with a sales engineer.

AMD Instinct™ MI355X GPUs Now Available

OTel: Proven at Scale on TensorWave

Subscribe to our Blog

Stay ahead of the curve with the latest in AI, AMD accelerators, and all things TensorWave.

Product

Resources

Company

© 2026 TensorWave Inc. - All rights reserved.