--- base_model: FINAL-Bench/Darwin-36B-Opus base_model_relation: quantized tags: - gguf - llama.cpp - imatrix - moe - reasoning - qwen3 - opal pipeline_tag: text-generation ---

πŸ’Ž OPAL

Consumer-GPU Quantization Harness

This repo contains OPAL Quants of FINAL-Bench/Darwin-36B-Opus

Tier Name Target Size Middle Layer Strategy
πŸ’Ž quality~23 GBIQ4_XS
βš–οΈ balanced~25 GBQ5_K
πŸ“¦ compact~18 GBQ3_K
πŸš€ mini~14 GBIQ2_S
πŸ”¬ micro~11 GBIQ1_M
🀏 nano~12 GBIQ2_XXS
πŸ›‘οΈ exp-minimalist~16 GBIQ2_XXS (Protected Edges)

πŸ“š Credits

πŸ‘‰ APEX Quantization Method
Ettore Di Giacinto & Richard Palethorpe (LocalAI Team). OPAL adapts the layer-wise precision gradients and MoE-aware tensor classification outlined in the APEX technical paper.

πŸ‘‰ Bartowski and Lamim
For the excellent semantic imatrix calibration dataset that powers OPAL's activation scaling.

πŸ‘‰ llama.cpp
Georgi Gerganov and contributors for the foundational inference and quantization engine.

πŸ‘‰ HuggingFace Accelerate
For the init_empty_weights() context manager that makes the 0-RAM "Ghost Model" possible.

πŸ‘‰ Jackrong
For the sexy markdown readme inspo.

Support the Project

A coffee in Ethereum would be cool! Although I don't drink coffeeβ€”I think it tastes like burnt waterβ€”but a pink lemonade would be fire! πŸ”₯

0xDEE7fa8C421BD038D32e4441ea1aDe72fE973706