INDIAAI FOUNDATION MODEL INITIATIVE
Dhīr 2B + Dhīr 5B MoE
Two models.
One original foundation-model stack.
PRETRAINED FROM SCRATCH ON GOOGLE CLOUD TPUs
Supported under the IndiaAI Mission (MeitY) Foundation Model initiative | ARTPARK IISc Bangalore
ZENTEIQ / DHIR / INDIAAI LAUNCH
INDIAAI FOUNDATION MODEL INITIATIVE
Dhīr 2B + Dhīr 5B MoE
Two models.
One original foundation-model stack.
Supported under the IndiaAI Mission (MeitY) Foundation Model initiative
ZENTEIQ / DHIR / INDIAAI LAUNCH
PRETRAINED FROM SCRATCH ON GOOGLE CLOUD TPUs
DHIR IN NOKKX
See Dhīr 5B MoE work.
Indian-language response with English subtitles
ZENTEIQ / DHIR / INDIAAI LAUNCH
ORIGINAL TOKENIZER
We started with the tokens.
Same scientific text. Named comparator. Actual counts.
ZENTEIQ / DHIR / INDIAAI LAUNCH
DATA GENERATION / QUALITY
Generate. Verify. Curate.
Original data-engineering workflow.
DEFINE
>
GENERATE
>
VERIFY
>
ACCEPT
ZENTEIQ / DHIR / INDIAAI LAUNCH
PRETRAINING / HARDWARE CHOICE
Built from the first training step.
An alternative route to foundation-model development.
01
STEP-ZERO EVIDENCE
02
GOOGLE CLOUD TPUs
03
CHECKPOINT LINEAGE
ZENTEIQ / DHIR / INDIAAI LAUNCH
RL + POST-TRAINING
Then we refined the model.
Verified post-training and evaluation evidence.
01
GENUINE TRAINING RUN
02
RELEASE LINEAGE
03
VERIFIED RESULT
ZENTEIQ / DHIR / INDIAAI LAUNCH
WHY COMPACT?
Compact by design.
Efficiency. Control. Closer to the user.
01
EFFICIENCY
Designed to limit compute needs
02
CONTROL
More deployment choices
03
PROXIMITY
AI closer to where it is used
Design direction; actual deployment depends on tested hardware and software.
ZENTEIQ / DHIR / INDIAAI LAUNCH
THE MOBILE REVEAL / DHIR 2B
Dhīr 2B. On your phone.
On-device. No network connection.
UNCUT / REAL TIME
TESTED DEVICE
MOTOROLA Edge 70 Fusion | 12 GB
MODEL ARTIFACT
4 bit quantization
ZENTEIQ / DHIR / INDIAAI LAUNCH
THE MOBILE REVEAL / DHIR 5B MoE
Dhīr 5B MoE. In your hand.
Trained on TPUs. Running in your hand.
UNCUT / REAL TIME / FLIGHT MODE
MOTOROLA Edge 70 Fusion | 12 GB
MODEL ARTIFACT
4-bit quantization
ZENTEIQ / DHIR / INDIAAI LAUNCH
BRAHMAI EVAL
Measured. Reproducible.
Open measurement with actual provenance.
01
PINNED CONFIGURATION
02
REPRODUCIBLE RUN
03
APPROVED METRIC
ZENTEIQ / DHIR / INDIAAI LAUNCH
PUBLIC RELEASE
Search Dhīr on AIKosh.
A real, public-browser discovery path.
1
SEARCH: Dhīr
2
FIND BOTH
3
OPEN MODEL
4
ACCESS ARTIFACTS
ZENTEIQ / DHIR / INDIAAI LAUNCH
TOWARDS SCIENTIFIC INTELLIGENCE
Search Dhīr on AIKosh.
Explore. Evaluate. Build.
From original foundation models to scientific and physical-world AI.
Our longer-term research direction
We acknowledge support under the IndiaAI Mission (MeitY) Foundation Model initiative and ARTPARK IISc Bangalore.
ZENTEIQ / DHIR / INDIAAI LAUNCH