🏛️ Zen Alta 1-3B (Initial Baseline / Legacy)
This repository contains the initial unpruned 28-layer baseline (3.21B parameters) from the early stages of the Zen Alta project, including raw FP16 safetensors shards and an unpruned Q4_K_M GGUF.
[!IMPORTANT] Active Production Releases: For the faster, 24-layer pruned & DPO-aligned production model and its speculative draft counterpart, please see:
- Production Target Model (24-Layer GGUF): ZenithLLM/ZenAlta-1-3B-Phase2-GGUF
- Speculative Decoding Draft Model: ZenithLLM/ZenAlta-Draft
- Phase 2 DPO LoRA Adapter: ZenithLLM/ZenAlta-1-3B-Phase2
📦 Repository Files
model-00001-of-00002.safetensors/model-00002-of-00002.safetensors: Full 28-layer base weightsllama-3.2-3b-instruct.Q4_K_M.gguf: Unpruned baseline GGUF (1.92 GB)Modelfile: Ollama configuration