---

license: apache-2.0

base_model: convaiinnovations/laya-multilingual

tags:

- gguf

- laya

- multilingual

- system-one

- mesh-llm

---

Laya Multilingual F16 GGUF for MeshLLM CI

This repository contains a llama.cpp-compatible F16 conversion of

convaiinnovations/laya-multilingual

for MeshLLM's portable System One smoke tests.

It is a test fixture, not a new checkpoint. The weights remain the Apache-2.0

licensed upstream Laya Multilingual model.

Provenance

• Source repository: convaiinnovations/laya-multilingual
• Source revision: 4bb4d65403a3a7b8abd9e6876ccb5e75cf923b5c
• MeshLLM llama.cpp upstream pin: 8212c7802455255460ab8e18fc34754560031b34
• MeshLLM patched llama.cpp tree: 34abd2f5c4cb393a92921d881b1691c87276c9d0
• Conversion: convert_hf_to_gguf.py --outtype f16
• Output SHA-256: 03881e45cdc5d0cf5e3a841cafe2ebdb8280e53e113c489e6847f7f45ca9c139
• Output bytes: 658805088

The upstream tokenizer_config.json declares extra_special_tokens as a JSON

array, while the pinned Transformers release expects that optional field to be

a map. Conversion removes only that redundant field; the tokens remain in

tokenizer.json and the model weights are unchanged.

Compatibility

This file uses the Laya GGUF layout implemented by MeshLLM's pinned llama.cpp

patch queue. GGUFs produced by the separate ggmlc compiler use a different

layout and are not interchangeable.