Compliance monitoring · Production

Dynamo AI trains multilingual 8B LLM in 10 days using Databricks Mosaic AI Training

The problem

Dynamo AI needed a multilingual, enterprise-compliant foundation model for AI guardrailing but found no open-source model that met their requirements, and early experimentation on other platforms failed to deliver expected efficiency gains.

First attempt

Architecture experiments on other platforms failed to deliver expected efficiency gains. During the training run, unexpected memory leakage forced a reduction in batch size that slowed training.

Workflow diagram · grounded in source
1
Decision to build own model
trigger
“Dynamo AI decided to train its own foundation model, Dynamo 8B, and turned to the team at Databricks Mosaic AI for help”
2
Multilingual data aggregation
integration
“We were prioritizing multilingual data because we wanted to make sure our model could serve many different countries, so we did a lot of public dataset aggregation”
3
Architecture efficiency optimization
human_review
“with help from the Databricks Databricks AI Research team to address the efficiency issues, we were able to train the model around 20% faster than what we found on competitors”
4
Memory leakage diagnosis
human_review
“Your team was able to look at our model weights and help us figure out where memory leakage was happening. Once we solved that issue, our training went much faster”
5
Foundation model pretraining
ai_action
“with Mosaic AI Training's built-in speedups and GPU availability on demand, it took Dynamo AI just 10 days to pretrain an 8-billion parameter multilingual LLM”
6
Compliance product deployment
output
“These product modules enable compliance, privacy and security throughout the generative AI stack, in support of sensitive workloads such as multilingual customer support, customer onboarding, claims processing and fraud detection”
Reported outcome

Using Databricks Mosaic AI Training, Dynamo AI pretrained an 8-billion parameter multilingual LLM in 10 days, achieved training around 20% faster than competitors, and saved weeks of development time — moving from experimentation to revenue generation in just a few months.

Reported metrics
Training speed vs competitors20% faster
LLM pretraining duration10 days
Model parameter count8-billion parameter
Development time saved on training setupweeks' worth of development time
Show all 5 reported metrics
training speed vs competitors20% faster
LLM pretraining duration10 days
model parameter count8-billion parameter
development time saved on training setupweeks' worth of development time
time from experimentation to revenue generationjust a few months
Reported stack
Databricks Mosaic AI Training
Source
https://www.databricks.com/customers/dynamo-ai
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Using Databricks Mosaic AI Training, Dynamo AI pretrained an 8-billion parameter multilingual LLM in 10 days, achieved training around 20% faster than competitors, and saved weeks of development time — moving from exp…

What tools did this team use?

Databricks Mosaic AI Training.

What results were reported?

Training speed vs competitors: 20% faster; LLM pretraining duration: 10 days; Model parameter count: 8-billion parameter; Development time saved on training setup: weeks' worth of development time (source-reported, not independently verified).

What failed first in this deployment?

Architecture experiments on other platforms failed to deliver expected efficiency gains.

How is this compliance monitoring AI workflow structured?

Decision to build own model → Multilingual data aggregation → Architecture efficiency optimization → Memory leakage diagnosis → Foundation model pretraining → Compliance product deployment.

Related compliance monitoring cases