ngxson's picture
ngxson HF Staff
upload tiny test model
f5e0a20 verified
|
Raw History Blame Contribute Delete
1.25 kB
metadata
license: other
license_name: lfm1.0
license_link: LICENSE
base_model: LiquidAI/d1-3B
tags:
  - testing

tiny-d1-3B-for-testing

This is a modified, damaged copy of LiquidAI/d1-3B, made only to test software. Its output is gibberish. Do not use it for anything else.

It keeps the architecture, tokenizer and chat template of d1-3B, so it runs the same code paths as the real model in llama.cpp (lfm2-d1 decision type: LFM2-VL, option codes read at the answer position, image input), but it is small enough for CI.

GGUF version: ggml-org-dev/tiny-d1-3B-for-testing-GGUF

Changes from the original model

Sliced from the safetensors with slice_d1_3b.py, which follows the slice-model skill of llama.cpp:

  • Text model: 4 of 30 layers (0, 2, 3, 5), hidden size 2048 -> 64, heads 32 -> 1, KV heads 8 -> 1, FFN 10752 -> 256
  • Vision encoder (SigLIP2): 3 of 27 layers, hidden size 1152 -> 144, heads 16 -> 2, FFN 4304 -> 288, projector 2048 -> 256
  • The Python code of the original repo is not included

License

Derived from d1-3B by Liquid AI, under the LFM Open License v1.0.