DBRX Instruct
Release · version Instruct
Maintained by Databricks1
DBRX Instruct is the finetuned version of DBRX Base, a mixture-of-experts language model with 132B total and 36B active parameters (16 experts, 4 active per input) pretrained on 12T tokens of text and code with a 32K-token context.15
- Release notes: Announcement: Introducing DBRX (external site: databricks.com)
- License: Databricks Open Model License (external site: databricks.com)
- License: Databricks Open Model Acceptable Use Policy (external site: databricks.com)
- Documentation: DBRX in Hugging Face Transformers (external site: huggingface.co)
Availability and license
Overall availability
Weights were published on Hugging Face at release in March 2024. As of this review Databricks' Hugging Face organization lists no public models, Hugging Face's Transformers documentation states the original DBRX Instruct checkpoint was closed, and Databricks retired DBRX from its hosted Foundation Model APIs and fine-tuning service during 2025.16578
Availability is separate from permission: read the license before using or redistributing.
Custom license dated March 27, 2024 covering the DBRX models, weights, and code identified at the databricks/dbrx GitHub repository. It grants a royalty-free limited license to use, modify, and distribute DBRX, but prohibits using DBRX, its derivatives, or its outputs to improve other large language models; requires a separate license from Databricks for licensees whose products had more than 700 million monthly active users on the release date; and requires compliance with the Databricks Open Model Acceptable Use Policy.34
Model-disclosure tier
The weights for this release are documented as not publicly available.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| WeightsCan the general public download the model parameters for this release? | Not public | Published on Hugging Face at release (March 2024). Hugging Face's Transformers documentation states the original databricks/dbrx-instruct checkpoint was closed, and Databricks' Hugging Face organization lists no public models.165 |
| Inference codeIs code for running the model published? | Public | Hugging Face Transformers includes a DBRX model implementation (DbrxForCausalLM). The databricks/dbrx GitHub repository referenced at release returned HTTP 404 when checked on 2026-09-29.5 |
| Training codeIs the code used to train the model published? | Unknown | The announcement says DBRX was trained with optimized versions of Databricks' open-source libraries MegaBlocks, LLM Foundry, Composer, and Streaming; the exact training code used for DBRX was not located.1 |
| Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access. | Unknown | The announcement gives only a high-level description (12T tokens of curated text and code, processed with Databricks tools) and does not list data sources.1 |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Partial | The announcement describes the architecture, curriculum learning over the data mix, and the training cluster (3,072 NVIDIA H100 GPUs) at a high level, without a full training configuration.1 |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Partial | Benchmark results were published in the announcement; evaluation code specific to this release was not assessed.1 |
Organization context
Provenance and derivatives
Databricks' release announcement presents DBRX Instruct as the finetuned counterpart of its DBRX Base model; the finetuning data is not described.1
- Derived from: DBRX Base — Databricks base model
U.S. eligibility
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.