---

license: apache-2.0

pipeline_tag: text-generation

tags:

• gguf
• llama.cpp
• pebble
• mamba2
• hybrid
• small-language-model

base_model: basically-ai/Pebble-25M

---

Pebble-25M-GGUF

GGUF conversions of basically-ai/Pebble-25M (Apache 2.0).

IMPORTANT: patched llama.cpp required

Pebble uses a custom hybrid Mamba2 + attention architecture. These GGUFs carry

general.architecture = "pebble", which upstream llama.cpp refuses to load.

Everything needed to run them lives in the support repo:

rootendpoint/basicallyai_llama.cpp_support

• llama.cpp-pebble.patch - adds the pebble architecture to llama.cpp

(applies cleanly against upstream commit 0eadefe)

• basicallyai_to_gguf.py - standalone converter (numpy + safetensors only)
• numpy_reference.py - independent reference implementation used to verify correctness

Apply the patch, rebuild llama.cpp, then:

llama-cli -m pebble-25m-f16.gguf -p "The capital of France" -n 64

Files

| Quant | Size | Type |

|-------|------|------|

| f16 | 49.1 MB | F16 |

| q8_0 | 26.3 MB | mostly Q8_0 |

| q4_k_m | 18.5 MB | mostly Q4_K_M |

Verification

Outputs were cross-checked token-by-token against an independent pure-numpy

reference implementation over an 8-prompt battery (CPU and CUDA backends).

Identical greedy sequences on all prompts. Quantized builds verified

greedy-identical to f16 (q8_0: 7/8 prompts, one genuine argmax tie-flip;

q4_k_m: 2/8, expected at this scale).

Model

• 25M parameters, hidden 608, 8 layers (6 Mamba2 + 2 attention), ctx 2048, vocab 2048
• A research-scale model: expect toy-level output quality.

CPU support (no GPU required)

Pure-CPU support for these models (no mamba-ssm, no CUDA) lives in the

basicallyai_cpu_support repository:

https://github.com/rootendpoint/basicallyai_cpu_support

It runs the original HF checkpoints in pure PyTorch on plain CPU

(~118 tok/s for 10M, ~65 tok/s for 25M on a Ryzen 5 2600X), verified

token-identical against this GGUF pipeline and an independent numpy oracle.