OmniASR LLM-300M โ€” GGUF

GGUF conversion of facebook/omniASR-LLM-300M for use with CrispASR.

OmniASR is Meta's multilingual ASR model family supporting 1600+ languages. Apache-2.0 license.

Autoregressive LLM decoder with language conditioning. Near-perfect English output.

Files

File Size
omniasr-llm-300m-v2-f16.gguf 3.0 GB
omniasr-llm-300m-v2-q4_k.gguf 1018 MB
omniasr-llm-300m-v2-q8_0.gguf 1.7 GB

Quick Start

git clone https://github.com/CrispStrobe/CrispASR && cd CrispASR
cmake -B build -DCMAKE_BUILD_TYPE=Release && cmake --build build -j$(nproc)

./build/bin/crispasr --backend omniasr-llm -m auto --auto-download -f audio.wav

Conversion

Converted using CrispASR's converter scripts with fixed positional conv weight normalization (per-kernel-position norm, not per-output-channel).

Downloads last month
448
GGUF
Model size
2B params
Architecture
omniasr-ctc
Hardware compatibility
Log In to add your hardware

8-bit

16-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for cstr/omniasr-llm-300m-v2-GGUF

Quantized
(1)
this model