DeepHermes Ascension Maze
DeepHermes Ascension Maze is an 8B research release from the Atropos reinforcement-learning framework. Nous does not recommend it for broad general-purpose use.
Research release
Training and model behavior
The introductory note describes an experiment over WildChat conversations. The rest of the card is model-generated and explicitly warns that its numerical and capability claims may be false. Those claims are not treated as specifications here.DeepHermes Ascension Maze model card
Working with this model
Practical guidance. Use the weights for controlled behavior research. Preserve the exact prompt, sampling settings and model revision when comparing experiments. Do not use the generated card as validation evidence.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Launch date | April 2025 |
| Access | Downloadable |
| Model identifiers | NousResearch/DeepHermes-AscensionMaze-RLAIF-8b-Atropos |
| Base model | meta-llama/Meta-Llama-3.1-8BDeepHermes Ascension Maze model card |
| Repository license label | llama3DeepHermes Ascension Maze model card |
Related models
- DeepHermes Egregore v1
Nous Research · Atropos experiments
An experimental DeepHermes reinforcement-learning artifact.
Launch date April 2025Research releaseDownloadable - DeepHermes Egregore v2
Nous Research · Atropos experiments
An experimental DeepHermes reinforcement-learning artifact.
Launch date April 2025Research releaseDownloadable - DeepHermes Tool Calling Specialist
Nous Research · Atropos experiments
An experimental model trained to improve tool calls during reasoning.
Launch date April 2025Research releaseDownloadable