c43g-s600m2 — GGUF (Q5_K_M)
Quantized build of
sjin59/c43g-s600m2.
Given 4 shuffled frames from a video plus a caption describing what happens, it recovers
the original chronological order of the frames. Output is four letters in chronological
order, e.g. "C,B,A,D".
Files
| File | Size | |
|---|
c43-Q5_K_M.gguf | 19.23 GB | quantized weights |
c43-mmproj-F16.gguf | 0.93 GB | multimodal projector — required, the model cannot see images without it |
Usage
Run with llama.cpp. Pass --jinja so the chat template is applied.
Fits on one RTX 3090 24GB — measured peak 21,437 MiB at --parallel 2 --ctx-size 26624,
including model, projector, KV cache and buffers.
Licensed under Apache-2.0, following the base model Qwen3.5-27B.
