Alpha
Contact
Support · MORENA 1.5B

Will it last?

Ownership levelPartialnone·limited·partial·substantial·fullAnalytical input D ยท 46.8/100

This page is a projection of the one entry record, the Transparency factor that Support covers. The full verdict is set by all four factors together, floor-weighted so the weakest caps the whole.

Which domain expands which factor
  • AssessUse & modify + Transparency
  • ImplementData control + Reliability
  • UseReliability
  • SupportTransparency

Common problems & fixes

Aggregated from the cards, the paper and the repo files (ev-card-instruct, ev-trainer, ev-paper). The list is not exhaustive.

Garbage or looping output usually means <|user|> strings or the wrong markers. Use <reserved_0> and <reserved_1> as in the chat template.

Refusals on harmless requests are a known issue: 41% of ordinary benign requests are not answered well. Rephrase, or fine-tune for your domain.

About one generation in six does not emit an end-of-sequence token within 200 tokens. Set a hard token limit and stop on <eos>.

In 59 of 1,000 into-English prompts the model repeated the instruction instead of translating, and Yoruba was worst. Check the prompt and the language.

Loading with transformers fails by design. Use the publisher PyTorch file or the GGUF.

load_example.py still says "preview weights", but the card and paper describe the same files as the release.

Versions, changelog & cadence

All eight repos were created on 2026-09-15 and made public on 2026-09-18, the release date in the paper. The only later change to the instruct repo is tokenizer_config.json with a chat template on 2026-10-14. There is no changelog file or release notes, so pin the commit hash.

The card compares against a "previous version" (benign answering 47.9%, grounded faithfulness 31%) that is not published in the org. No release cadence or roadmap is stated beyond the paper's promise of a corpus and datasheet "once the legal review is complete" (ev-hf-platform, ev-paper).

Community & support channels

The Hugging Face discussions tab on vamboai/morena-1.5b-instruct is open but empty (2026-11-19). The Vambo AI site (vambo.ai) offers contact and enterprise forms. Its blog has no posts, and the MORENA paper is listed only on the research page (ev-vambo-site).

The Fikira dataset card lists a general questions address. It is not a support or security channel for the model.

Tracked known issues

There is no public issue tracker. The publisher lists its own known issues in the card and paper (ev-card-instruct, ev-paper).

Over-refusal of benign requests (41%) is only partly fixed. Grounded generation faithfulness fell from 31% to 23% against the previous version. The chosen checkpoint misses the publisher's own selection rule on privacy (by 0.1 points), drugs (1.0), violence (2.6) and degeneracy (1.3% against a 1% limit). Retrieval QA and African multiple choice are near chance.

isiNdebele is evaluated on transcripts only. The 0.5B model loses about a third of the teacher's English reasoning, and neither smaller model has been safety-tested.

How this scores

The ownership factor this domain covers, drawn from the one entry record.

2

TransparencyDo you know what it is: weights, training, behaviour, and legible terms?

Moderate

Open weights with a detailed paper, but the corpus, SFT corpus, configs and evaluation harness are not released, and the machine-translation and synthetic-data generators are unnamed (ev-paper, ev-trainer).

How this scores (AOI sub-dimensions)
Provenance2/5how well we can trace and verify what went into the modelSafetensors only, a published SHA256SUMS that matches the host's LFS hash for the weights, and a clean Hugging Face malware scan give about 4 of 8 checklist boxes.
Governance2/5how accountable and well-documented the publisher isA named company (Vambo Technologies (Pty) Ltd) with named founders, a prior public dataset (Fikira, December 2025) and a post-release fix (tokenizer_config.json, 2026-10-14).
What this means for adoptionYou own the use of the weights: Apache-2.0 is declared, ungated and commercial use is stated, and your data stays on your hardware. You do not own a finished assistant. The paper says the instruct model is not one, it cannot be rebuilt without the unreleased corpus and configs, and the evaluations are the publisher's own. The level rises if Vambo ships a LICENSE file, releases its corpus and harness, and clears the Gemma Terms question on the SFT data.

Sources

The same evidence records as the entry sheet. Read means the text was verified; unverified means it is known to exist but not yet read.

Model cardread2026-11-19
The instruct card declares Apache-2.0, names twelve languages, and gives 1.485B parameters and a 4,096-token context.
Model cardread2026-11-19
The base card describes a base checkpoint beside the instruct one (vamboai/morena-1.5b-base) with the same Apache-2.0 declaration.
Licenceread2026-11-19
Every MORENA repo declares Apache-2.0 in the front matter and the card 'Licence' section ('Commercial use is permitted, including fine-tuning and redistributing derivatives').
Licenceread2026-11-19
Canonical Apache License 2.0 text, the licence the MORENA declaration points to: section 2 grants copyright, section 3 grants patent rights, and section 4 governs redistribution and requires passing on a copy of the licence.
Technical_reportread2026-11-19
Paper 'MORENA: An African Foundation Model' (23 pages, 18 September 2026, single author).
Documentationread2026-11-19
modeling_morena.py (52,103 bytes, identical in base and instruct) calls itself the 'Morena production trainer' (FSDP2, Muon, mixture sampler, checkpointing) and also defines the model.
Documentationread2026-11-19
SHA256SUMS in the repo lists five files.
Documentationread2026-11-19
Hugging Face platform metadata.
Model cardread2026-11-19
Sibling cards: morena-0.5b-mini and morena-0.2b-nano (pruned and distilled from the 1.5B base, not safety-tested, Apache-2.0) and morena-1.5b-instruct-gguf (Q4_K_M 0.92 GB, Q8_0 1.58 GB, f16 2.97 GB, 'verified for logit equivalence').
Documentationread2026-11-19
vamboai/fikira is a public, ungated MIT dataset of 50,000 synthetic reasoning examples in ten languages, copyright 'Vambo Technologies (trading as Vambo AI)', one named SFT component; vamboai/morena-sft-corpus, listed in the card metadata, returns HTTP 401 and is not public.
Vendor announcementread2026-11-19
The vambo.ai footer names Vambo Technologies (Pty) Ltd, and the site names co-founders.
Third-party analysisread2026-11-19
Three press articles (Tech In Africa 13 Oct 2026, CIO Africa 16 Oct 2026, Daba Finance 7 Oct 2026) describe a South African start-up releasing an open-weight model and say the results are the developer's own and unverified.
Terms of serviceread2026-11-19
Gemma Terms of Use (last modified April 1, 2026).