tiny-clef-for-testing
This is a modified, damaged copy of Cloudflare/clef, made only to test software. Its output is gibberish. Do not use it for anything else.
It keeps the architecture, tokenizer, chat template and joint decision head layout of clef, so it runs the same code paths as the real model in llama.cpp (clef decision type: Qwen3.5 trunk with linear and full attention, joint schema head with routing and joint layers, packed questions, image input), but it is small enough for CI.
GGUF version: ggml-org-dev/tiny-clef-for-testing-GGUF
Changes from the original model
Sliced from the safetensors with slice_clef.py, which follows the slice-model skill of llama.cpp:
- Text model: 4 of 64 layers (0-3: 3 linear attention + 1 full attention), hidden size 5120 -> 32, FFN 17408 -> 128, heads 24 -> 2, KV heads 4 -> 1, linear attention key heads 16 -> 2, value heads 48 -> 6 (head dims kept)
- Vision encoder: 2 of 27 blocks (0, 1), hidden size 1152 -> 144, heads 16 -> 2, FFN 4304 -> 288, output size 5120 -> 32
- Joint head (
joint_head.safetensors,joint_head_config.json): all 2 routing and 4 joint layers kept, width 1024 -> 32, heads 16 -> 2, FFN 4096 -> 64, input hidden size 5120 -> 32 - The Python code of the original repo (
joint_schema_model.py) is not included
License
Derived from clef by Cloudflare, under the Apache License 2.0.
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support