MythoMax L2 13B — Model Documentation
Published by Nextbit256 S.L. | Last updated: May 2026 Compliance reference: AI Act Regulation (EU) 2024/1689, Art. 53.1.c and 53.1.d
Model Identification
| Field | Detail |
|---|---|
| Full model name | MythoMax-L2-13b |
| Developed by | Gryphe (merge/fine-tune author); base models: Meta AI (Llama 2 13B), The-Face-Of-Goonery (Huginn-13b), and the MythoLogic-L2-13B intermediate merge |
| Release date | 2023 |
| Model card | huggingface.co/Gryphe/MythoMax-L2-13b |
| Technical report | Not published by Gryphe |
| Developer blog | github.com/Gryphe/BlockMerge_Gradient |
Architecture and Parameters
| Field | Detail |
|---|---|
| Architecture | Dense (decoder-only transformer) — Llama 2 architecture |
| Total parameters | 13B |
| Active parameters per token | N/A — dense model (all parameters active) |
| Non-embedding parameters | Not published by Gryphe |
| Layers | 40 (standard Llama 2 13B architecture) |
| Attention | Multi-head attention (standard Llama 2 13B configuration) |
| Native context length | 4,096 tokens (Llama 2 base) |
| Extended context | Not published by Gryphe |
| Precision | Not published by Gryphe (FP16 typical for this model class) |
License and Commercial Use
License: Meta Llama 2 Community License
MythoMax-L2-13b is built on Llama 2 13B. The Llama 2 Community License is a proprietary Meta license — it is not an open-source license such as Apache 2.0 or MIT, and it contains important restrictions that differ from standard permissive licenses.
Key terms of the Meta Llama 2 Community License:
- Commercial use: Permitted for organizations with fewer than 700 million monthly active users (MAU). Organizations whose products or services exceeded 700 million MAU in the preceding calendar month must request a separate license from Meta before exercising any rights.
- Restriction on improving competing models: Llama 2 outputs may not be used to train, improve, or fine-tune competing large language models (other than Llama 2 itself or its derivatives).
- Attribution: Licensees must retain the copyright notice "Llama 2 is licensed under the LLAMA 2 Community License, Copyright © Meta Platforms, Inc."
- Distribution: When sharing Llama materials with third parties, a copy of the Llama 2 Community License must be provided to those parties.
- Indemnification: Licensees must indemnify Meta against third-party claims arising from their use or distribution.
Nextbit's verification: Nextbit has reviewed the Llama 2 Community License and confirmed that serving MythoMax-L2-13b via API for commercial use is permitted, provided Nextbit's customer base does not collectively constitute products or services exceeding 700 million monthly active users. Nextbit currently operates well below this threshold.
Restrictions relevant to users: Users of Nextbit's API who access this model are themselves bound by the Llama 2 Community License terms. In particular:
- Users whose own products exceed 700 million MAU must obtain separate Meta approval.
- Outputs from this model may not be used to train competing large language models.
Note on the merge author's license: The model card lists the license as "Other" (non-standard). The effective license governing use is the Meta Llama 2 Community License, as the base model (Llama 2 13B) is the foundational work on which the merge is built.
Training Data Summary
This model has two distinct layers: the base model pretraining data and the merge-specific data.
(a) Base model pretraining data — Llama 2 13B
MythoMax-L2-13b is built on the Llama 2 13B base model. The pretraining data for Llama 2 is documented by Meta AI. Nextbit does not reproduce that documentation here; for details see the Meta Llama 2 model card and the Llama 2 Technical Report.
Key known facts about Llama 2 13B pretraining:
- Approximately 2 trillion tokens
- Publicly available web text, primarily in English
- Knowledge cutoff: September 2022
(b) Merge data — Gryphe's BlockMerge_Gradient method
MythoMax-L2-13b is a model merge, not a conventional fine-tune on new data. It was created by merging tensor weights from two intermediate models:
- MythoLogic-L2-13B (for robust instruction understanding)
- Huginn-13b-FP16 by The-Face-Of-Goonery (for creative writing capability)
The merge used Gryphe's experimental BlockMerge_Gradient technique, which applies individual ratios to each of the 363 unique tensor layers rather than a uniform blending ratio. This is a weight interpolation process; no new training data was introduced during the merge.
Merge data: Not applicable — the merge process does not use training data; it interpolates existing model weights.
Not published by Gryphe:
- The exact per-tensor merge ratios used (available as YAML files in the linked GitHub repository)
- Any fine-tuning data applied after the merge
Languages Supported
Primarily English. MythoMax-L2-13b inherits Llama 2's language distribution, which is predominantly English-language. Multilingual capability is limited. Full language list: see the Llama 2 Technical Report.
Intended Uses
This model is designed for:
- Roleplaying and character-driven narratives
- Creative writing and story generation
- Interactive fiction
The model was specifically engineered to combine strong instruction understanding (from MythoLogic) with expressive creative writing capability (from Huginn). It is optimized for roleplay and storywriting use cases rather than factual question answering, coding, or reasoning tasks.
Gryphe does not state explicitly excluded uses in the model card, but the model's training lineage (Llama 2) is subject to Meta's Acceptable Use Policy. Use for content involving minors in sexual contexts or other prohibited uses is forbidden under the Llama 2 Community License and Meta's Acceptable Use Policy.
Systemic Risk Assessment
| Field | Detail |
|---|---|
| Training FLOPs | Not published by Meta AI for Llama 2 13B; not applicable for the merge step |
| Relevant base for threshold assessment | Llama 2 13B (the pretrained base model); the merge step involves no significant additional compute |
| Estimated FLOPs (Llama 2 13B pretraining) | ~2.4 × 10²³ (estimate based on 13B parameters × ~2T tokens; formula: 6 × N × D) |
| Exceeds 10²⁵ FLOPs threshold? | No — estimated to be approximately 40× below the systemic risk threshold |
| Note on dense architecture | Llama 2 13B is a dense model; all 13B parameters are active per token. The full parameter count is used for FLOPs estimation. |
| AI Office designation | Not designated as a systemic risk model as of May 2026 |
| Art. 55 obligations apply? | No |
The FLOPs relevant for systemic risk threshold assessment are those of the base model (Llama 2 13B pretraining). The merge step performed by Gryphe involves weight interpolation only and does not constitute model training in the regulatory sense. This estimate is based on publicly available information from the Llama 2 Technical Report. If Meta publishes official compute figures, this section will be updated.
This documentation is published by Nextbit256 S.L. in accordance with Article 53(1)(c) and 53(1)(d) of Regulation (EU) 2024/1689 (AI Act). Nextbit256 S.L. serves this model via its inference API but did not develop or train it. The merge was performed by Gryphe; the base model was developed by Meta AI. All information is sourced from public documentation published by Gryphe and Meta AI. This model is served under the Meta Llama 2 Community License.
For questions: [email protected]