Automatic Speech Recognition
LiteRT
LiteRT
speech
audio
parakeet
tdt
int8
android
on-device
buffered-streaming
Instructions to use spybyscript/parakeet-tdt-litert with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LiteRT
How to use spybyscript/parakeet-tdt-litert with LiteRT:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
Revert manifest rename and restore original Parakeet bundle
Browse filesRestore the original manifest.json, ZIP, documentation, publication metadata and checksums from 39d57af6. Remove parakeet-manifest.json as requested. Preserve model weights and legacy archive.
- INTEGRATION.md +0 -16
- README.md +3 -26
- SHA256SUMS +5 -5
- parakeet-manifest.json → manifest.json +0 -0
- parakeet-tdt-0.6b-v3-litert.zip +2 -2
- publication.json +1 -5
INTEGRATION.md
CHANGED
|
@@ -1,21 +1,5 @@
|
|
| 1 |
# Integrating the INT8 Parakeet bundle
|
| 2 |
|
| 3 |
-
## Manifest filename and internal repackaging
|
| 4 |
-
|
| 5 |
-
Read Parakeet's model metadata from **`parakeet-manifest.json`**. The current ZIP contains no generic `manifest.json`. An internal packaging system may create that generic filename for its own file catalog; these are two independent documents with different schemas.
|
| 6 |
-
|
| 7 |
-
Keep `parakeet-manifest.json` and all four model assets in the same extracted bundle directory, and include them in your internal catalog. Change only the metadata file lookup in any consumer that previously expected the generic name; the Parakeet JSON schema, release ID, tensor contracts and weight hashes are unchanged. Do not use the distribution catalog as the Parakeet model configuration.
|
| 8 |
-
|
| 9 |
-
For example, the relevant Kotlin lookup becomes:
|
| 10 |
-
|
| 11 |
-
```kotlin
|
| 12 |
-
val parakeetMetadata = JSONObject(
|
| 13 |
-
File(bundleDirectory, "parakeet-manifest.json").readText()
|
| 14 |
-
)
|
| 15 |
-
```
|
| 16 |
-
|
| 17 |
-
A consumer that embeds all model settings may not read this metadata at inference time, but consumers that load tensor/frontend/decoding settings from it must preserve it under the new name. LiteRT itself does not automatically load either JSON file. Historical bundles retain `legacy/manifest.json`; pin the older repository revision if using a consumer that still requires that legacy layout.
|
| 18 |
-
|
| 19 |
## Migrating from the FP16 release
|
| 20 |
|
| 21 |
Use the root INT8 encoder with root decoder, joint and tokenizer. Their filenames remain stable. Decoder/joint/tokenizer bytes match the previous release. Check `release_id` and SHA256 in the manifest when invalidating downloaded-file caches. The previous complete release is in `legacy/`; its documentation describes that historical release, not the current INT8 memory measurements.
|
|
|
|
| 1 |
# Integrating the INT8 Parakeet bundle
|
| 2 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
## Migrating from the FP16 release
|
| 4 |
|
| 5 |
Use the root INT8 encoder with root decoder, joint and tokenizer. Their filenames remain stable. Decoder/joint/tokenizer bytes match the previous release. Check `release_id` and SHA256 in the manifest when invalidating downloaded-file caches. The previous complete release is in `legacy/`; its documentation describes that historical release, not the current INT8 memory measurements.
|
README.md
CHANGED
|
@@ -35,34 +35,11 @@ Community conversion of [NVIDIA Parakeet-TDT-0.6B-v3](https://huggingface.co/nvi
|
|
| 35 |
| [decoder_step.tflite](decoder_step.tflite) | 23.7 | Original FP16 LSTM prediction network |
|
| 36 |
| [joint_step.tflite](joint_step.tflite) | 12.7 | Original FP16 token/duration joint network |
|
| 37 |
| [tokenizer.model](tokenizer.model) | 0.4 | Unchanged SentencePiece tokenizer |
|
| 38 |
-
| [
|
| 39 |
| [validation.json](validation.json) | — | Measured quality and memory results, with limitations |
|
| 40 |
| [mapped_model_file.h](mapped_model_file.h) | — | Optional native source-page memory hint; Linux-tested example |
|
| 41 |
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
**Everything at the repository root belongs to the current INT8 package, including `parakeet-tdt-0.6b-v3-litert.zip`.** Only `legacy/` contains the previous FP16 package. The two ZIPs have the same basename but different paths and contents. The decoder, joint and tokenizer are intentionally unchanged between releases; the current encoder is INT8.
|
| 45 |
-
|
| 46 |
-
For an app integration, download these **five files** together:
|
| 47 |
-
|
| 48 |
-
- `encoder_T1500.tflite`
|
| 49 |
-
- `decoder_step.tflite`
|
| 50 |
-
- `joint_step.tflite`
|
| 51 |
-
- `tokenizer.model`
|
| 52 |
-
- `parakeet-manifest.json` — Parakeet tensor, frontend and decoding metadata
|
| 53 |
-
|
| 54 |
-
The four binary assets total approximately **665 MB on disk**. Alternatively, download and extract the current [ZIP bundle](https://huggingface.co/spybyscript/parakeet-tdt-litert/resolve/main/parakeet-tdt-0.6b-v3-litert.zip). It includes all five files plus the model card, integration guide, attribution, validation/provenance metadata and optional native memory helper. You do not need both the ZIP and separate copies of its files, or anything from `legacy/`.
|
| 55 |
-
|
| 56 |
-
The ZIP is a complete **model-asset bundle**, not a runnable app: the consuming app supplies LiteRT, SentencePiece handling, the audio/log-mel frontend and TDT decoding loop described in [INTEGRATION.md](INTEGRATION.md). `validation.json` and `publication.json` are informational; `mapped_model_file.h` is optional. Preserve attribution notices when redistributing. Repository `.gitattributes` is not a runtime asset.
|
| 57 |
-
|
| 58 |
-
Root [SHA256SUMS](SHA256SUMS) includes the ZIP checksum. Inside the ZIP, `parakeet-manifest.json` includes checksums for the four binary assets.
|
| 59 |
-
|
| 60 |
-
### Repackaging and the two manifests
|
| 61 |
-
|
| 62 |
-
The current package deliberately has **no root `manifest.json`**, either as a loose download or inside the ZIP. Your distribution system can add its own `manifest.json` describing the archive contents without overwriting Parakeet's model metadata. Register/preserve `parakeet-manifest.json` alongside the four assets in your internal catalog.
|
| 63 |
-
|
| 64 |
-
Update any Parakeet metadata reader from `manifest.json` to `parakeet-manifest.json`. Its JSON contents and the model weights are unchanged; LiteRT graphs do not read this filename themselves. The old `legacy/manifest.json` remains part of the archived release and should not be copied into the current package.
|
| 65 |
-
|
| 66 |
|
| 67 |
To avoid downloading the legacy release and duplicate ZIPs, select root files explicitly:
|
| 68 |
|
|
@@ -74,7 +51,7 @@ snapshot_download(
|
|
| 74 |
local_dir="parakeet-int8",
|
| 75 |
allow_patterns=[
|
| 76 |
"encoder_T1500.tflite", "decoder_step.tflite", "joint_step.tflite",
|
| 77 |
-
"tokenizer.model", "
|
| 78 |
"README.md", "INTEGRATION.md", "NOTICE.md", "mapped_model_file.h",
|
| 79 |
],
|
| 80 |
)
|
|
|
|
| 35 |
| [decoder_step.tflite](decoder_step.tflite) | 23.7 | Original FP16 LSTM prediction network |
|
| 36 |
| [joint_step.tflite](joint_step.tflite) | 12.7 | Original FP16 token/duration joint network |
|
| 37 |
| [tokenizer.model](tokenizer.model) | 0.4 | Unchanged SentencePiece tokenizer |
|
| 38 |
+
| [manifest.json](manifest.json) | — | Tensor contracts, profile, component hashes and provenance |
|
| 39 |
| [validation.json](validation.json) | — | Measured quality and memory results, with limitations |
|
| 40 |
| [mapped_model_file.h](mapped_model_file.h) | — | Optional native source-page memory hint; Linux-tested example |
|
| 41 |
|
| 42 |
+
The four runtime assets total approximately **665 MB on disk**. Download them individually, or use the updated [ZIP bundle](https://huggingface.co/spybyscript/parakeet-tdt-litert/resolve/main/parakeet-tdt-0.6b-v3-litert.zip). Root [SHA256SUMS](SHA256SUMS) includes the ZIP checksum; the ZIP contains component checksums in its manifest.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
|
| 44 |
To avoid downloading the legacy release and duplicate ZIPs, select root files explicitly:
|
| 45 |
|
|
|
|
| 51 |
local_dir="parakeet-int8",
|
| 52 |
allow_patterns=[
|
| 53 |
"encoder_T1500.tflite", "decoder_step.tflite", "joint_step.tflite",
|
| 54 |
+
"tokenizer.model", "manifest.json", "validation.json",
|
| 55 |
"README.md", "INTEGRATION.md", "NOTICE.md", "mapped_model_file.h",
|
| 56 |
],
|
| 57 |
)
|
SHA256SUMS
CHANGED
|
@@ -1,14 +1,14 @@
|
|
| 1 |
11ad7efa24975ee4b0c3c3a38ed18737f0658a5f75a0a96787b576a78a023361 .gitattributes
|
| 2 |
-
|
| 3 |
5bf6b396ccf5b900a009f166a33d2464ba0fdce0ec117ad7f002cf5dc66ee14e NOTICE.md
|
| 4 |
-
|
| 5 |
eb0bf3559a0b4cbdc3ca05b7e8ff948ee5ef158ce424667b62a85f6c769a9ce1 decoder_step.tflite
|
| 6 |
b26804686f98d68f7394baf3b13bdcea8e846ebf4ecf2903e3fc30fe3a3882a8 encoder_T1500.tflite
|
| 7 |
0e28c22fc426df9900ef4a1bd15760ec757e44f0fd1818e0afb51c4fe79031be joint_step.tflite
|
| 8 |
2c6b2684937eaa67b2b6e68a0ad4aff96c0a114e8a135435de554822b258f3d0 legacy/archive.json
|
|
|
|
| 9 |
4226d273861baa0328a5dcf8ef5d9eba604074137111ba53116f709ba63399c0 mapped_model_file.h
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
a682d396c4af3d91c1244cd515bcc1e772639e09c2b4bc566b2e31d63eede726 publication.json
|
| 13 |
eacec2b0a77f336d4a2ca4a25a7047575d3c2b74de47e997f4c205126ed3135e tokenizer.model
|
| 14 |
f869d60f54a5d1cfbe2f3c73e2a1f3845fccf64109decc6cdd12345a89862b16 validation.json
|
|
|
|
| 1 |
11ad7efa24975ee4b0c3c3a38ed18737f0658a5f75a0a96787b576a78a023361 .gitattributes
|
| 2 |
+
510b2919183633eb53418a6c0ef65cf6b72ac6e04845893e3a67f70d840a2ccb INTEGRATION.md
|
| 3 |
5bf6b396ccf5b900a009f166a33d2464ba0fdce0ec117ad7f002cf5dc66ee14e NOTICE.md
|
| 4 |
+
c0dac492e179384912d4a623522d349d7f212a0012d5bc1db54c52adfea3cef4 README.md
|
| 5 |
eb0bf3559a0b4cbdc3ca05b7e8ff948ee5ef158ce424667b62a85f6c769a9ce1 decoder_step.tflite
|
| 6 |
b26804686f98d68f7394baf3b13bdcea8e846ebf4ecf2903e3fc30fe3a3882a8 encoder_T1500.tflite
|
| 7 |
0e28c22fc426df9900ef4a1bd15760ec757e44f0fd1818e0afb51c4fe79031be joint_step.tflite
|
| 8 |
2c6b2684937eaa67b2b6e68a0ad4aff96c0a114e8a135435de554822b258f3d0 legacy/archive.json
|
| 9 |
+
12bc7c5c126db8f7f97b617eb151d8e5b14d0ee3d57244f952693faa2a8fde98 manifest.json
|
| 10 |
4226d273861baa0328a5dcf8ef5d9eba604074137111ba53116f709ba63399c0 mapped_model_file.h
|
| 11 |
+
cd49232e6756b62ba14049bb31bab799838e30170b235a6993888e57662fdc95 parakeet-tdt-0.6b-v3-litert.zip
|
| 12 |
+
8efdaeebf2ead197dc5a265e07be3b184cce31056d53a2315a6a3e535f4dbcd3 publication.json
|
|
|
|
| 13 |
eacec2b0a77f336d4a2ca4a25a7047575d3c2b74de47e997f4c205126ed3135e tokenizer.model
|
| 14 |
f869d60f54a5d1cfbe2f3c73e2a1f3845fccf64109decc6cdd12345a89862b16 validation.json
|
parakeet-manifest.json → manifest.json
RENAMED
|
File without changes
|
parakeet-tdt-0.6b-v3-litert.zip
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:cd49232e6756b62ba14049bb31bab799838e30170b235a6993888e57662fdc95
|
| 3 |
+
size 665277590
|
publication.json
CHANGED
|
@@ -7,9 +7,5 @@
|
|
| 7 |
"encoder_precision": "dynamic_INT8",
|
| 8 |
"decoder_joint_precision": "FP16 weights, FP32 interfaces",
|
| 9 |
"android_validated": false,
|
| 10 |
-
"private_evaluation_material_included": false
|
| 11 |
-
"package_revision": "20260916-namespaced-manifest-v1",
|
| 12 |
-
"manifest_file": "parakeet-manifest.json",
|
| 13 |
-
"model_weights_unchanged": true,
|
| 14 |
-
"change": "Rename bundle metadata to avoid collisions with internal distribution manifests"
|
| 15 |
}
|
|
|
|
| 7 |
"encoder_precision": "dynamic_INT8",
|
| 8 |
"decoder_joint_precision": "FP16 weights, FP32 interfaces",
|
| 9 |
"android_validated": false,
|
| 10 |
+
"private_evaluation_material_included": false
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
}
|