tiny-pplx-decider-v1-27b-for-testing
This is a modified, damaged copy of perplexity-ai/pplx-decider-v1-27b, made only to test software. Its output is gibberish. Do not use it for anything else.
It keeps the architecture, tokenizer, chat template, decision_config.json (codes, token ids, temperature) and readout of pplx-decider-v1-27b, so it runs the same code paths as the real model in llama.cpp (pplx-decider decision type: non-causal readout of 255 option codes stored as an LM head, calibrated temperature, image input), but it is small enough for CI.
GGUF version: ggml-org-dev/tiny-pplx-decider-v1-27b-for-testing-GGUF
Changes from the original model
Sliced from the safetensors with slice_qwen35_decision.py (python3 slice_qwen35_decision.py perplexity-ai/pplx-decider-v1-27b <out dir>), 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 -> 64, heads 24 -> 2, KV heads 4 -> 1 (head_dim 256 kept), linear attention key heads 16 -> 2, value heads 48 -> 6 (head dims 128 kept)
- Vision encoder: 2 of 27 blocks, hidden size 1152 -> 144, heads 16 -> 2 (head_dim 72 kept), FFN 4304 -> 288, output 5120 -> 32
readout.safetensors: 255 x 5120 -> 255 x 32 (first columns)- The 11 safetensors shards are merged into one
model.safetensors - The Python code of the original repo (
source/,inference.py) andrelease-manifest.jsonare not included
License
Derived from pplx-decider-v1-27b by Perplexity (itself derived from Qwen/Qwen3.8-27B), under the Apache License 2.0. See NOTICE.
- Downloads last month
- -