russian-oracle/laya-multilingual-GGUF

laya-multilingual GGUF (Q8_0)

GGUF conversion of convaiinnovations/laya-multilingual (Apache-2.0) for the POST /v1/systemone endpoint of llama-server (llama.cpp PR #29818). Not affiliated with the model's authors.

llama-server -m laya-multilingual-Q8_0.gguf

How it was made

• Source: convaiinnovations/laya-multilingual at revision e4e9ddf21a7b1903b7acffd8814ad4307bf63a67.
• Converter: convert_hf_to_gguf.py of llama.cpp at commit eec18f5d32099fb15d4ba15003a231bcc72757d5, --outtype q8_0 --model-name Laya-Multilingual, with its pinned requirements (transformers 4.57.6).
• One change to the source before converting: tokenizer/tokenizer_config.json lists extra_special_tokens (["<start_of_turn>", "<end_of_turn>"]), which transformers 4.57.6 cannot load; it was rewritten as the mapping {"extra_0": "<start_of_turn>", "extra_1": "<end_of_turn>"}. Weights are unchanged.
• laya-multilingual-Q8_0.gguf sha256: 8fcb3a3e9c5e119fa2676208ce4eea7f47e32fea434d8e63d8635fb74c12417b.

Loaded and answered /v1/systemone with llama-server build 11371 (99b95488c): decision model type: laya.

See the source model card for capabilities and limits (it ships uncalibrated).