Beirit Harvey

See how TensorWave and ScalarLM bring open, portable enterprise AI to AMD GPU clusters, Kubernetes/HPC scale, SOTA SQL reasoning, and ~90% cost savings via fine-tuning.

TensorWave + ScalarLM: Open, Portable Enterprise AI on AMD

A live panel during Open Source AI Week brought together TensorWave, ScalarLM, RelationalAI, and AT&T to show how open standards plus AMD GPU clusters can turn enterprise data into real production wins, from SQL reasoning that tops leaderboards to massive cost cuts through targeted fine-tuning.









Summary

ScalarLM is an open-source framework that unifies training and inference, running on TensorWave’s large AMD Instinct clusters with Kubernetes/HPC orchestration. Built from Megatron-Core,

A live panel during Open Source AI Week brought together TensorWave, ScalarLM, RelationalAI, and AT&T to show how open standards plus AMD GPU clusters can turn enterprise data into real production wins, from SQL reasoning that tops leaderboards to massive cost cuts through targeted fine-tuning.









Summary

ScalarLM is an open-source framework that unifies training and inference, running on TensorWave’s large AMD Instinct clusters with Kubernetes/HPC orchestration. Built from Megatron-Core, Hugging Face, and vLLM, it lets teams spin up experiments at scale and deploy quickly. RelationalAI layers on “super-alignment,” converting Snowflake/relational data into model-learnable semantics so LLMs actually understand enterprise schemas and business logic.

AT&T described their internal “Ask AT&T” platform (100k+ employees), where the pattern is consistent: start with the biggest model to prove value, then fine-tune smaller ones to slash costs, reporting ~90% savings on a production case. Using TensorWave clusters, the team ran 150+ training trials to beat open benchmarks on log understanding and demonstrated state-of-the-art SQL generation (Spider). Looking ahead, a multi-telco consortium is open-sourcing domain-tuned telecom models (English first, then multimodal/multilingual) with a public milestone targeted for GSMA Barcelona in March 2026.


Key Takeaways

 * Open stack: ScalarLM (Megatron-Core + HF + vLLM) on AMD clusters, designed to be GPU-agnostic.
 * Cloud-native HPC: Kubernetes deployment with HPC constructs to keep scale and performance.
 * Super-alignment: RelationalAI turns relational/Snowflake estates into tokens and semantics LLMs can learn from.
 * Private data > public data: biggest gains now come from safely using enterprise data.
 * Production at scale: AT&T’s “Ask AT&T” processes billions of tokens/day across many internal use cases.
 * Fine-tune to save: after proving quality with large models, small fine-tuned models delivered ~90% cost reduction.
 * Benchmark wins: state-of-the-art SQL reasoning (Spider) and strong BIRD dev results from the joint stack.
 * Industry momentum: multi-telco open model effort (LLM + embeddings), first release aiming for GSMA Barcelona 2026.


About TensorWave

TensorWave is the AI neocloud purpose-built for performance. Powered exclusively by AMD Instinct™ Series GPUs, we deliver high-bandwidth, memory optimized infrastructure that scales with your most demanding models–training or inference. 

Ready to get started? Connect with a sales engineer. 

AMD Instinct™ MI355X GPUs Now Available

TensorWave + ScalarLM: Open, Portable Enterprise AI on AMD

Subscribe to our Blog

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

Product

Resources

Company

© 2025 TensorWave Inc. - All rights reserved.