Snowflake Arctic Base
Release in the Snowflake Arctic family · version Base
Maintained by Snowflake AI Research1
Arctic Base is the pretrained checkpoint of Snowflake Arctic, a dense-MoE hybrid transformer (a 10B dense model plus a residual 128x3.66B MoE MLP) with 480B total and 17B active parameters, pretrained from scratch by the Snowflake AI Research Team. It takes text input and generates text and code.1
- Model hub: Model card (external site: huggingface.co)
- Repository: Snowflake-Labs/snowflake-arctic repository (external site: github.com)
- License: Apache License 2.0 (repository LICENSE) (external site: github.com)
- Release notes: Arctic announcement (external site: snowflake.com)
Availability and license
Overall availability
Downloadable from Hugging Face without gating; loading with Transformers requires trust_remote_code for the custom Arctic model code.1
Availability is separate from permission: read the license before using or redistributing.
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 | Checkpoint published on Hugging Face under Apache-2.0 without gating.1 |
| Inference codeIs code for running the model published? | Public | Custom modeling code for Transformers is included with the checkpoint, and the GitHub repository provides Hugging Face and vLLM inference examples.14 |
| Training codeIs the code used to train the model published? | Unknown | The repository's training folder contains a LoRA fine-tuning script and a checkpoint converter; pretraining code for Arctic was not located.4 |
| Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access. | Unknown | The model card refers readers to a cookbook series for data details; this review did not verify those pages.1 |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Unknown | The announcement describes a cookbook series on pretraining, fine-tuning, data, and systems; this review did not verify whether it documents the full Arctic training configuration.5 |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Partial | Benchmark comparisons are published as figures in the repository README; evaluation code was not located.4 |
Run and use notes
- The model card documents use with Transformers 4.39 or later (with trust_remote_code) and DeepSpeed 0.14.2 or later, and the GitHub repository includes a vLLM deployment tutorial.14
- Because of the model size, the model card recommends a single 8xH100 instance; its example loads the model in bfloat16 with DeepSpeed FP8 quantization (FP6 is also supported) and a max_memory setting of 150GiB per GPU.1
Organization context
Provenance and derivatives
Pretrained from scratch by the Snowflake AI Research Team, per the model card.1
Catalog records that name this entry in their provenance:
Other releases in the Snowflake Arctic family
- Snowflake Arctic 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.