OpenASR commited on
Commit
50df0b3
Β·
verified Β·
1 Parent(s): aa791e7

Update firered2-llm model card with measured RTF/RAM

Browse files
Files changed (1) hide show
  1. README.md +119 -0
README.md ADDED
@@ -0,0 +1,119 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ base_model: FireRedTeam/FireRedASR2-LLM
4
+ pipeline_tag: automatic-speech-recognition
5
+ library_name: openasr
6
+ tags:
7
+ - automatic-speech-recognition
8
+ - speech-to-text
9
+ - openasr
10
+ - oasr
11
+ - firered2-llm
12
+ ---
13
+
14
+ <div align="center">
15
+
16
+ # FireRedASR2 LLM Β· OpenASR
17
+
18
+ **FireRedTeam's LLM-backbone Mandarin-first bilingual ASR β€” 8B+ parameters engineered for state-of-the-art Chinese and dialect accuracy**
19
+
20
+ [![License](https://img.shields.io/badge/license-Apache--2.0-2563eb.svg)](https://huggingface.co/FireRedTeam/FireRedASR2-LLM)
21
+ [![Format](https://img.shields.io/badge/format-.oasr-7c3aed.svg)](https://github.com/QuintinShaw/openasr)
22
+ [![Runtime](https://img.shields.io/badge/runtime-OpenASR-111827.svg)](https://openasr.org)
23
+ [![Base model](https://img.shields.io/badge/base-FireRedASR2--LLM-f59e0b.svg)](https://huggingface.co/FireRedTeam/FireRedASR2-LLM)
24
+
25
+ Native speech-to-text in the **[OpenASR](https://github.com/QuintinShaw/openasr)** runtime β€”
26
+ engineered for peak performance on CPU & GPU, **no Python at inference time**.
27
+
28
+ </div>
29
+
30
+ ---
31
+
32
+ ## ✨ Highlights
33
+
34
+ - πŸ₯‡ **Best-in-class Mandarin accuracy** β€” 2.89% average CER across four public Mandarin benchmarks, outperforming Doubao-ASR (3.69%), Qwen3-ASR (3.76%), and Fun-ASR (4.16%) on the same comparison table (arXiv:2603.10420; FireRedTeam's FireRedASR2-LLM model card)
35
+ - πŸ—£οΈ **Leading dialect and accent coverage** β€” 11.55% average CER across 19 public Chinese dialect/accent benchmarks, ahead of Doubao-ASR (15.39%) and Qwen3-ASR (11.85%) (arXiv:2603.10420; FireRedTeam's FireRedASR2-LLM model card)
36
+ - πŸ‡¨πŸ‡³πŸ‡¬πŸ‡§ **Bilingual Mandarin + English** β€” one 8B+ parameter checkpoint handles both languages, as demonstrated in the upstream model card's bilingual examples (FireRedTeam's FireRedASR2-LLM model card)
37
+ - 🧠 **LLM-scale decoder backbone** β€” Encoder-Adapter-LLM architecture at 8B+ parameters, the bigger and more accurate sibling of the already-available firered-aed-l-v2 (1.1B AED, 3.05%/11.67% avg CER) (FireRedTeam's FireRedASR2-LLM model card)
38
+ - πŸ”’ **Fully local, q4_k build** β€” runs 100% on-device via OpenASR's .oasr runtime with no cloud upload, under the Apache-2.0 license inherited from upstream (FireRedTeam's FireRedASR2-LLM model card)
39
+ - πŸ¦€ **Native in OpenASR** β€” `.oasr` packs run with no Python at inference, engineered for peak performance on CPU & GPU
40
+
41
+ ## πŸš€ Quickstart
42
+
43
+ ```bash
44
+ # 1. Install the OpenASR CLI Β· https://openasr.org
45
+ # 2. Pull a build (pick a quant β€” see the table below)
46
+ openasr pull firered2-llm:q4
47
+
48
+ # 3. Transcribe
49
+ openasr transcribe audio.wav --model firered2-llm
50
+ ```
51
+
52
+ All builds for this model:
53
+
54
+ ```bash
55
+ openasr pull firered2-llm:q4
56
+ ```
57
+
58
+ ## πŸ“¦ Available builds
59
+
60
+ | Quant | File (`.oasr`) | Size | RAM peak | RTF Β· M1 CPU | RTF Β· M1 GPU | JFK Ξ”WER vs fp16 |
61
+ |:------|:---------------|-----:|---------:|-------------:|-------------:|-----------------:|
62
+ | q4_k | `firered2-llm-q4_k.oasr` | 5.10 GB | 9.39 GB | 0.77Γ— | 0.40Γ— | n/a |
63
+
64
+ <sub>RTF = real-time factor on the fixed 11s JFK clip (**lower is faster**); RAM peak measured per pack
65
+ in an isolated subprocess. JFK Ξ”WER compares each quantized build's JFK transcript to this model's
66
+ fp16 JFK transcript, so it measures quantization drift rather than absolute recognition accuracy.
67
+ **q4_k** is the recommended default β€” near-reference quality at a fraction of the
68
+ footprint.</sub>
69
+
70
+ ## 🧠 About FireRedASR2 LLM
71
+
72
+ FireRedASR2-LLM is the Encoder-Adapter-LLM member of **FireRedASR2**, the successor to
73
+ FireRedTeam's open-source industrial-grade **FireRedASR** speech-recognition family, released as
74
+ part of the **FireRedASR2S** all-in-one ASR system. At 8B+ parameters with an LLM-scale decoder
75
+ backbone, it is the bigger and more accurate sibling of the already-available
76
+ **firered-aed-l-v2** (FireRedASR2-AED, a 1.1B-parameter attention encoder-decoder reporting
77
+ 3.05%/11.67% avg CER). The FireRedASR2S technical report (arXiv:2603.10420) and the upstream model
78
+ card both report **2.89% average Character Error Rate** across four public Mandarin benchmarks
79
+ and **11.55% average CER** across 19 public Chinese dialect/accent benchmarks -- outperforming
80
+ Doubao-ASR (3.69%/15.39%), Qwen3-ASR (3.76%/11.85%), and Fun-ASR (4.16%/12.76%) on the paper's own
81
+ comparison table. These numbers belong to FireRedASR2-LLM and should not be confused with the
82
+ smaller firered-aed-l-v2's 3.05%/11.67% figures. The model is bilingual (Mandarin Chinese and
83
+ English). This OpenASR pack distributes the q4_k quantization build, repackaged as an `.oasr`
84
+ pack that runs natively in the OpenASR runtime -- fully local, no Python at inference time, no
85
+ cloud upload. Licensed under Apache-2.0, inherited from the upstream release.
86
+
87
+ ## βš™οΈ How these packs were made
88
+
89
+ Converted from [FireRedTeam/FireRedASR2-LLM](https://huggingface.co/FireRedTeam/FireRedASR2-LLM) with the OpenASR importer:
90
+
91
+ ```bash
92
+ openasr model-pack import firered-llm <src> <out>.oasr \
93
+ --package-id firered2-llm --quantization {fp16,q8-0,q4-k}
94
+ ```
95
+
96
+ The `.oasr` container is GGUF-backed; packs use zero-copy mmap weight binding and graph
97
+ buffer reuse to keep peak memory low.
98
+
99
+ ## βš–οΈ License
100
+
101
+ These packs **inherit the upstream model's license: Apache-2.0**
102
+ ([source](https://huggingface.co/FireRedTeam/FireRedASR2-LLM)). OpenASR packaging retains the upstream copyright and
103
+ NOTICE; the only modifications are format conversion and quantization.
104
+
105
+ ## πŸ™ Acknowledgements
106
+
107
+ This pack is a redistribution of **FireRedASR2-LLM**, created and released by **FireRedTeam**
108
+ ([FireRedTeam/FireRedASR2-LLM](https://huggingface.co/FireRedTeam/FireRedASR2-LLM),
109
+ [FireRedTeam/FireRedASR2S](https://github.com/FireRedTeam/FireRedASR2S)). All credit for the
110
+ architecture, training, and weights belongs to FireRedTeam; the license is inherited from and
111
+ identical to the upstream model (**Apache-2.0**, as declared on the upstream model card). Thank
112
+ you to FireRedTeam for releasing their work openly. OpenASR only performs format conversion,
113
+ quantization, runtime verification, and local-inference adaptation.
114
+
115
+ ## πŸ”— Links
116
+
117
+ - πŸ¦€ **OpenASR** β€” <https://github.com/QuintinShaw/openasr>
118
+ - 🌐 **Website** β€” <https://openasr.org>
119
+ - πŸ€— **Upstream model** β€” [FireRedTeam/FireRedASR2-LLM](https://huggingface.co/FireRedTeam/FireRedASR2-LLM)