Mtrini-27B-Tellus โ technical release ๐งช
27B-class model
4096 context
QLoRA / NF4
BF16 compute
700 training steps
11,200 training examples
Final training loss: 0.4302
And yes โ thereโs an 8.47 GiB IQ2_XS GGUF for local inference.
#LLM#GGUF#LlamaCpp#AI
๐ Mtrini-27B-Tellus is LIVE.
Our most capable Model
Available as:
โข LoRA Adapter
โข Merged Transformers
โข F16 GGUF
โข IQ2_XS GGUF
โข IQ2_XS imatrix
Built by CompiwerAI in Morocco ๐ฒ๐ฆ
Building AI For Everyone.
#AI#LLM#OpenSource#Morocco
๐จ MTRINI-SVL-1.1 IS HERE ๐ช๐ฒ๐ฆ
Our newest model is LIVE.
๐ง Reasoning
๐ป Coding
๐งฎ Math
๐ฒ๐ฆ Moroccan Darija
๐ Self-Verifying Loop (SVL)
8B parameters. Built by Compiwer AI.
๐ค Hugging Face: https//huggingface.co/CompiwerAI/Mtrini-SVL-1.1
#AI#LLM#OpenSource#Mtrini
Mtrini Tellus 12B got cooked ๐ 63.1% vs GPT-5.6 Sol 98.4% & Opus 5 98.1% in our ML-Pro1 and last with 50% overall in our RMCMMK benchmark. Check it at https://t.co/KpJkSm4Pxr. Weโre fixing it in public. Moroccan AI, open source. ๐ฒ๐ฆ #AI#LLM