license: apache-2.0
tags:
- gguf
- qwen3-coder
- moe
- abliterated
- tool-calling
- cypher-studio
base_model: mradermacher/Huihui-Qwen3-Coder-30B-A3B-Instruct-abliterated-i1-GGUF
Cypher Qwen3-Coder 30B-A3B Abliterated (i1-Q4_K_M GGUF)
Miroir souverain Cypher Studio (cypher-opencode) optimise pour Kaggle Dual-T4 (30 Go VRAM) sous llama-server --jinja.
- Architecture : MoE 30.5B total / 3.3B actifs, mode direct sans latence
<think> - Quantification :
i1-Q4_K_Mimatrix (17.28 GiB) - Capacites : Agent de code haute vitesse (~25-27 tok/s sur 2x T4), Tool Calling natif, contexte 64K-256K
- Source amont :
mradermacher/Huihui-Qwen3-Coder-30B-A3B-Instruct-abliterated-i1-GGUF