Rnj-1.5 Instruct
Release in the Rnj family · version rnj-1.5-instruct
Maintained by Essential AI1
A long-context follow-up to Rnj-1 Instruct that extends the context window from 32K to 160K tokens. Essential AI built it from the Rnj-1 base model, switching most attention layers to block-local attention with a group of global layers in the middle, and added long-context mid-training data and more software-engineering training data.1
- Model hub: Model card (Hugging Face) (external site: huggingface.co)
- Model hub: Rnj-1 base model card (external site: huggingface.co)
Availability and license
Overall availability
Downloadable from Hugging Face without gating.1
Availability is separate from permission: read the license before using or redistributing.
The card states that the repository and model weights are licensed under Apache 2.0 and links to the license file in the rnj-1-instruct repository.1
Model-disclosure tier
The model parameters for this release can be downloaded by the public. License terms may still restrict use, redistribution, or commercial use.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| WeightsCan the general public download the model parameters for this release? | Public | 1 |
| Inference codeIs code for running the model published? | Public | The card points to the Rnj-1 usage instructions and says support was added to vLLM in v0.20.0, with the block-local attention implemented in Triton.12 |
| Training codeIs the code used to train the model published? | Unknown | Not assessed. |
| Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access. | Partial | The card describes the added mid-training data (science PDFs converted to text with olmOCR 2 and repository-level code with fill-in-the-middle examples), synthetic long-context tasks, and about 600k synthetic software-engineering trajectories from three teacher models it does not name. No dataset links are given.1 |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Partial | The card describes the attention-layer pattern and the kinds of mid-training and synthetic task data at a high level, without hyperparameters.1 |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Partial | The card reports benchmark results, including a "lookback" long-context evaluation Essential AI built from GitHub repositories; no evaluation data or code links are given for this release.1 |
What it is useful for
Coding, including agentic software-engineering tasks run in harnesses such as SWE-Agent and mini-swe-agent, and tasks that need retrieval over long inputs. The card says the model was optimized for long-context comprehension rather than long-context generation.1
Organization context
Provenance and derivatives
Built by Essential AI from its Rnj-1 base model, which it trained from scratch. The card says the synthetic software-engineering trajectories used in training were generated by three teacher models, which it does not identify.12
- Derived from: Rnj-1 (base) (external site: huggingface.co) — Essential AI base model; no separate catalog record.
Other releases in the Rnj family
- Rnj-1 InstructModel-disclosure tier (USASI rubric v0.1): Open-weight
U.S. eligibility
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.