Fabio
c36510c21e
Some checks failed
Linux Build / Linux cpu (push) Has been cancelled
Linux Build / Linux cpu-clang (push) Has been cancelled
Linux Build / Linux vulkan (push) Has been cancelled
mac-build / macOS CPU (push) Has been cancelled
Nix Build / Nix (cpu) (push) Has been cancelled
Nix Build / Nix (vulkan) (push) Has been cancelled
windows-build / Windows CPU (push) Has been cancelled
1.2 KiB
1.2 KiB
SAMSONE
SAMSONE is a compact English audio-language model for captioning, question answering, and sound, music, and speech understanding. The native port supports the upstream 99M, 134M, and 356M variants with audio clips up to 30 seconds.
BF16 and Q8_0 packages are available for each variant.
Usage
build/debug/bin/audiocpp_cli \
--task asr \
--family samsone \
--model /path/to/samsone-99m-bf16.gguf \
--backend cuda \
--audio input.wav \
--request-option "instruct=what can be heard in this audio?" \
--log \
--text-out response.txt
Request Options
Use these with --request-option key=value.
| Option | Values | Default | Description |
|---|---|---|---|
instruct |
English text | describe the audio |
Question or instruction about the audio. |
max_tokens |
positive integer within checkpoint context budget | 100 |
Maximum response tokens. Must be positive; audio/prompt tokens plus this budget must fit checkpoint max_position_embeddings. |
The upstream repository does not currently publish a license file. Review the upstream terms before redistributing converted weights.
Upstream
- SamsungLabs/samsone, release
v1.0.0