---

license: apache-2.0

language:

• zh
• en

pipeline_tag: automatic-speech-recognition

tags:

• audio
• speech-recognition
• transcription
• gguf
• firered
• conformer
• ctc
• mandarin
• chinese

library_name: ggml

base_model: FireRedTeam/FireRedASR2-AED

---

FireRedASR2-AED -- GGUF

GGUF conversions and quantisations of FireRedTeam/FireRedASR2-AED for use with CrispStrobe/CrispASR.

Available variants

| File | Quant | Size | Notes |

|---|---|---|---|

| firered-asr2-aed.gguf | F16 | 2.3 GB | Full precision |

| firered-asr2-aed-q8_0.gguf | Q8_0 | 1.4 GB | High quality |

| firered-asr2-aed-q4_k.gguf | Q4_K | 919 MB | Best size/quality tradeoff |

All variants produce identical transcription on test audio.

Model details

• Architecture: Conformer encoder (16L, d=1280, 20 heads, relative positional encoding with pos_bias_u/v, macaron FFN, depthwise separable conv k=33) + CTC head
• Parameters: 1.1B
• Languages: Mandarin Chinese, English, 20+ Chinese dialects
• License: Apache 2.0
• CER: 3.05% (Mandarin average, per paper)
• Encoder: Hybrid ggml/CPU — ggml for matmuls, CPU for relative position attention scoring

Usage with CrispASR

git clone https://github.com/CrispStrobe/CrispASR && cd CrispASR
cmake -S . -B build && cmake --build build -j8

# Auto-detect backend from GGUF
./build/bin/crispasr -m firered-asr2-aed-q4_k.gguf -f audio.wav

# Explicit backend
./build/bin/crispasr --backend firered-asr -m firered-asr2-aed-q4_k.gguf -f audio.wav -osrt

Note: Output is in UPPERCASE (the model was trained with uppercase English text). CTC decoding is used; beam search decoder not yet implemented.

Conversion

python models/convert-firered-asr-to-gguf.py --input FireRedTeam/FireRedASR2-AED --output firered-asr2-aed.gguf
crispasr-quantize firered-asr2-aed.gguf firered-asr2-aed-q4_k.gguf q4_k

Provenance and EU AI Act Art. 53 note

• Upstream model: FireRedTeam/FireRedASR2-AED — published by FireRedTeam.
• Upstream licence: apache-2.0. This repository redistributes under the same terms; it grants no rights the upstream licence does not.
• What was done here: format conversion and/or quantisation only (GGUF/GGML). No training, no fine-tuning, no merging, no distillation, no change to architecture, vocabulary or capability. Only the numeric representation of the upstream weights differs.
• Training data: documented — where it is documented at all — by the upstream provider; see the upstream model card. No training data was used, added or selected by this repository. No training-content summary was found on the upstream model card at the time of writing; that documentation gap is upstream's and is not filled here.
• Provider status: under Regulation (EU) 2024/1689 the upstream authors remain the provider of this model. Converting the serialisation format does not make this repository the provider of a new general-purpose AI model, and no such claim is made. Questions about training content, copyright policy or model capability belong upstream.