Granite-4.0-H-Tiny — APEX GGUF
===============================

Quantized GGUF derivative assembled from these components:

1. Base model — Granite-4.0-H-Tiny
   Copyright (c) IBM · License: Apache-2.0
   https://huggingface.co/ibm-granite/granite-4.0-h-tiny
   Baseline: IBM's own bf16 GGUF (ibm-granite/granite-4.0-h-tiny-GGUF).

2. Quantization engine — llama.cpp
   Copyright (c) 2023-2026 The ggml authors · License: MIT
   https://github.com/ggml-org/llama.cpp

3. APEX quantization recipe & toolkit — apex-quant
   Copyright (c) 2026 Ettore Di Giacinto (LocalAI) · License: MIT
   https://github.com/localai-org/apex-quant

4. Calibration data — Bartowski calibration_datav3 (used to build the imatrix)
   https://gist.github.com/bartowski1182/eb213dccb3571f863da82e99418f81e8

This derivative is released under Apache-2.0 (matching the base). Modifications:
MoE-aware mixed-precision (APEX) requantization to GGUF. Unofficial community
quantization; not affiliated with or endorsed by IBM.