OpenASR commited on
Commit
61306bc
Β·
verified Β·
0 Parent(s):

publish firered-punc OpenASR packs

Browse files
Files changed (3) hide show
  1. .gitattributes +1 -0
  2. README.md +111 -0
  3. firered-punc-fp16.oasr +3 -0
.gitattributes ADDED
@@ -0,0 +1 @@
 
 
1
+ *.oasr filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ base_model: FireRedTeam/FireRedPunc
4
+ pipeline_tag: automatic-speech-recognition
5
+ library_name: openasr
6
+ tags:
7
+ - openasr
8
+ - oasr
9
+ - firered-punc
10
+ ---
11
+
12
+ <div align="center">
13
+
14
+ # FireRedPunc Β· OpenASR
15
+
16
+ **Chinese punctuation restoration for OpenASR -- a BERT token classifier that adds full-width marks to unpunctuated transcripts, fully on-device**
17
+
18
+ [![License](https://img.shields.io/badge/license-Apache--2.0-2563eb.svg)](https://huggingface.co/FireRedTeam/FireRedPunc)
19
+ [![Format](https://img.shields.io/badge/format-.oasr-7c3aed.svg)](https://github.com/QuintinShaw/openasr)
20
+ [![Runtime](https://img.shields.io/badge/runtime-OpenASR-111827.svg)](https://openasr.org)
21
+ [![Base model](https://img.shields.io/badge/base-FireRedPunc-f59e0b.svg)](https://huggingface.co/FireRedTeam/FireRedPunc)
22
+
23
+ A capability-pack support model for the **[OpenASR](https://github.com/QuintinShaw/openasr)**
24
+ runtime β€” pure-Rust inference, **no Python at inference time**. Not a standalone
25
+ transcription model: it augments another OpenASR ASR model's own decode path.
26
+
27
+ </div>
28
+
29
+ ---
30
+
31
+ ## ✨ Highlights
32
+
33
+ - 🈢 **Chinese punctuation restoration** -- restores full-width `οΌŒγ€‚οΌŸοΌ` on an unpunctuated ASR transcript (e.g. Dolphin, FireRedASR2-AED) in a finalize-only post-processing pass
34
+ - 🧠 **BERT token classifier** -- a 12-layer `chinese-lert-base` encoder with a 5-class head predicting which of none/οΌŒγ€‚οΌŸοΌ follows each subword
35
+ - πŸ‡¨πŸ‡³ **Chinese-only by construction** -- the released label set has no English half-width marks, so only Chinese text is punctuated
36
+ - πŸ”Œ **Opt-in and auto-gated** -- runs only when the ASR model is unpunctuated and this pack is installed, so punctuating families are never double-punctuated
37
+ - πŸ¦€ **Native in OpenASR** β€” `.oasr` packs run with no Python at inference, engineered for peak performance on CPU & GPU
38
+
39
+ ## πŸš€ Quickstart
40
+
41
+ ```bash
42
+ # 1. Install the OpenASR CLI Β· https://openasr.org
43
+ # 2. Pull the pack
44
+ openasr pull firered-punc:fp16
45
+
46
+ # 3. Use it as an opt-in refinement for another model's transcribe call
47
+ openasr transcribe meeting.wav --model <asr-model> --word-timestamps=aligned
48
+ ```
49
+
50
+ ## πŸ“¦ Pack
51
+
52
+ | Quant | File (`.oasr`) | Size |
53
+ |:------|:---------------|-----:|
54
+ | fp16 | `firered-punc-fp16.oasr` | 204 MB |
55
+
56
+ ## 🧠 About FireRedPunc
57
+
58
+ FireRedPunc is a **punctuation-prediction** model from **FireRedTeam**, part of the
59
+ FireRedASR2S all-in-one ASR system. It adopts a BERT-style encoder initialized from
60
+ `chinese-lert-base` (a 12-layer, 768-hidden, 12-head BERT) with a token-classification head
61
+ that predicts, for each subword, which of five classes follows it: none, or one of the four
62
+ Chinese full-width marks `,`, `。`, `?`, `!`. It is a text-in / labels-out post-processor,
63
+ not an ASR model -- no audio frontend, no autoregressive decode.
64
+
65
+ In OpenASR it is packaged as an optional **punctuation capability pack**: an unpunctuated
66
+ family's finished transcript (e.g. Dolphin or FireRedASR2-AED) is passed through the classifier
67
+ once, and the predicted marks are re-inserted into the original characters. The stage is
68
+ auto-gated on the ASR model's catalog `emits_punctuation == false` and only runs when this pack
69
+ is installed, so punctuating families are never double-punctuated. Because the released label
70
+ set is Chinese-only, the integration is Chinese-only by construction (the architecture cannot
71
+ emit English half-width marks).
72
+
73
+ This OpenASR repo repackages the upstream weights as `.oasr` packs that run natively in the
74
+ OpenASR runtime -- no Python at inference, all decoding local. It ships in **fp16**.
75
+
76
+ **Not a standalone transcription model.** This pack cannot transcribe audio by itself; it is an
77
+ opt-in post-process applied to an unpunctuated ASR model's output.
78
+
79
+ **Verification status:** this is a brand-new catalog entry (no prior public listing). Local
80
+ verification covers exact per-token label parity against the upstream PyTorch forward across a
81
+ set of Chinese golden sentences (all four punctuation classes plus the no-mark class). This
82
+ pack is staged in a private repo, not yet publicly listed.
83
+
84
+ ## βš™οΈ How this pack was made
85
+
86
+ Converted from [FireRedTeam/FireRedPunc](https://huggingface.co/FireRedTeam/FireRedPunc) with the OpenASR local-source
87
+ converter (`convert_local_qwen_forced_aligner_source_to_runtime_pack`, not yet wired
88
+ to a public `openasr model-pack import` subcommand -- the pack format and tensor
89
+ mapping are stable, but CLI/family-registry wiring is a separate stage).
90
+
91
+ The `.oasr` container is GGUF-backed; packs use zero-copy mmap weight binding.
92
+
93
+ ## βš–οΈ License
94
+
95
+ This pack **inherits the upstream model's license: Apache-2.0**
96
+ ([source](https://huggingface.co/FireRedTeam/FireRedPunc)). OpenASR packaging retains the upstream copyright;
97
+ the only modifications are format conversion and quantization.
98
+
99
+ ## πŸ™ Acknowledgements
100
+
101
+ This pack is a redistribution of **FireRedPunc**, created and open-sourced by **FireRedTeam**
102
+ ([FireRedTeam/FireRedPunc](https://huggingface.co/FireRedTeam/FireRedPunc)). All credit for the
103
+ original architecture, training, and weights belongs to the authors; the license is inherited
104
+ from and identical to the upstream model (Apache-2.0). OpenASR only performs format conversion,
105
+ runtime verification, and local-inference adaptation.
106
+
107
+ ## πŸ”— Links
108
+
109
+ - πŸ¦€ **OpenASR** β€” <https://github.com/QuintinShaw/openasr>
110
+ - 🌐 **Website** β€” <https://openasr.org>
111
+ - πŸ€— **Upstream model** β€” [FireRedTeam/FireRedPunc](https://huggingface.co/FireRedTeam/FireRedPunc)
firered-punc-fp16.oasr ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:95033ed3842e5399f6eec430c28950d34ecba0b41d0c0f7069ccbd8280f48234
3
+ size 203631008