Dahlia 1.0
Version Tag: NS24.07.02
The Dahlia is built from the ground up for hypermodality —
reasoning seamlessly across text, images, number and video, thus powering a growing family of
use cases at Udyam AI.
Responsible by design
We've built our Dahlia model responsibly from the start,
incorporating safeguards and working together with partners to
ensure responsible and trustworthy AI solutions through curated
datasets and rigorous tuning.
Unmatched on performance
From text, image, time series and video understanding to
mathematical reasoning, Dahlia's performance exceeds or matches the current state-of-the-art
results on various widely-accepted benchmarks.
600M+
Data Points
100B
Model Parameters
∞
Infinite Applciations
Benchmark
Our research team is continually exploring new ideas at the frontier of AI,
building innovative products that show consistent progress on a range of benchmarks.
| Benchmark | Dahlia 1.0 | Llama 2 70B | Gemini 1.5 | Mistral 7B | Mistral 8x7B | GPT 3.5 |
|---|---|---|---|---|---|---|
MMLU (General) |
81.9% | 69.9% | 78.9% | 62.47% | 70.63% | 70.0% |
WMT23 (Multilingual) |
71.7% | - | 74.1% | - | - | - |
MMMU (Image) |
58.4% | - | 56.1% | - | - | - |
MATH (Math) |
58.5% | - | 54.9% | - | - | 40.1% |
GPQA (Reasoning) |
51.2% | - | 39.5% | - | - | - |
M4 (Forecasting Error) |
13.26% | - | - | - | - | - |
Note: The above benchmark is a subset of the actual benchmark used to establish performance of Dahlia 1.0. Most of the models mentioned above do not posses multimodal properties. Gemini 1.5 is the only model here after Dahlia 1.0 that has been built from the ground up for modality.
Advance with Udyam AI
Adopt our cutting-edge Decision Intelligence to
transform your business.