Text Generation
LiteRT-LM
LiteRT
English
litertlm
on-device
edge
asr
automatic-speech-recognition
text-normalization
inverse-text-normalization
punctuation
truecasing
speech-to-text
dictation
post-processing
qwen3
Instructions to use mlboydaisuke/S1-mini-LiteRT with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LiteRT-LM
How to use mlboydaisuke/S1-mini-LiteRT with LiteRT-LM:
# LiteRT-LM runs on various platforms (Android, iOS, Windows, Linux, macOS, IoT, Web/WASM) # and supports many APIs (C++, Python, Kotlin, Swift, JavaScript, Flutter). # For platform-specific integration guides, please refer to the official developer website: # https://ai.google.dev/edge/litert-lm # To try LiteRT-LM, the easiest way is to use our CLI tool. # 1. Install the LiteRT-LM CLI tool: pip install -U litert-lm # 2. Download and run this model locally: # See: https://ai.google.dev/edge/litert-lm/cli # A single .litertlm file in the repo is picked automatically; otherwise the CLI asks which one to run # (or pass its name right after the repo id). litert-lm run \ --from-huggingface-repo=mlboydaisuke/S1-mini-LiteRT \ --prompt="Write me a poem"
- LiteRT
How to use mlboydaisuke/S1-mini-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
Plain wording in the litertlm_manifest.json notes
Browse files- litertlm_manifest.json +8 -7
litertlm_manifest.json
CHANGED
|
@@ -1,7 +1,7 @@
|
|
| 1 |
{
|
| 2 |
"manifest_schema": "0.1.3",
|
| 3 |
"repo": "mlboydaisuke/S1-mini-LiteRT",
|
| 4 |
-
"generated": "2026-
|
| 5 |
"generator": "make_manifest.py",
|
| 6 |
"model": {
|
| 7 |
"display_name": "S1-mini (LiteRT-LM)",
|
|
@@ -70,7 +70,7 @@
|
|
| 70 |
"platform": "android",
|
| 71 |
"device_class": "flagship",
|
| 72 |
"backend": "gpu",
|
| 73 |
-
"reason": "measured on one Galaxy S26 with one binary (
|
| 74 |
}
|
| 75 |
],
|
| 76 |
"requirements": {
|
|
@@ -145,7 +145,7 @@
|
|
| 145 |
"device": "iPhone 17 Pro",
|
| 146 |
"os": "iOS",
|
| 147 |
"backend": "gpu",
|
| 148 |
-
"runtime": "litert-lm (Swift
|
| 149 |
"prompt_tokens": 180,
|
| 150 |
"prefill_tps": 801,
|
| 151 |
"decode_tps": 32.0,
|
|
@@ -159,7 +159,7 @@
|
|
| 159 |
"device": "iPhone 17 Pro",
|
| 160 |
"os": "iOS",
|
| 161 |
"backend": "cpu",
|
| 162 |
-
"runtime": "litert-lm (Swift
|
| 163 |
"prompt_tokens": 180,
|
| 164 |
"prefill_tps": 281,
|
| 165 |
"decode_tps": 15.5,
|
|
@@ -184,7 +184,7 @@
|
|
| 184 |
"cache": "no",
|
| 185 |
"runs": 2,
|
| 186 |
"date": "2026-09-05",
|
| 187 |
-
"source": "
|
| 188 |
},
|
| 189 |
{
|
| 190 |
"device": "Galaxy S26 (SM-S942Q, Qualcomm SM8850, Adreno)",
|
|
@@ -200,14 +200,15 @@
|
|
| 200 |
"cache": "no",
|
| 201 |
"runs": 2,
|
| 202 |
"date": "2026-09-05",
|
| 203 |
-
"source": "
|
| 204 |
}
|
| 205 |
],
|
| 206 |
"known_issues": [
|
| 207 |
"Pixel 8a GPU: on number-dense input a date separator can be lost ('March 3rd, 2026.' -> 'March 32026.'), deterministic on repeat runs; the CPU backend on the same phone is byte-identical to macOS across all 12 test cases. Prefer CPU on this phone when exactness matters, GPU when time-to-first-token matters.",
|
| 208 |
"The released checkpoint itself drifts from its own upstream card's example table (it keeps 'Hmm,' / 'So' discourse markers the card drops); parity here is stated against HF fp32, not against those published examples.",
|
| 209 |
"All iPhone 17 Pro rows were measured with the device reporting a 'serious' thermal state; they are floors, not peaks."
|
| 210 |
-
]
|
|
|
|
| 211 |
}
|
| 212 |
]
|
| 213 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"manifest_schema": "0.1.3",
|
| 3 |
"repo": "mlboydaisuke/S1-mini-LiteRT",
|
| 4 |
+
"generated": "2026-11-14",
|
| 5 |
"generator": "make_manifest.py",
|
| 6 |
"model": {
|
| 7 |
"display_name": "S1-mini (LiteRT-LM)",
|
|
|
|
| 70 |
"platform": "android",
|
| 71 |
"device_class": "flagship",
|
| 72 |
"backend": "gpu",
|
| 73 |
+
"reason": "measured on one Galaxy S26 with one binary (2026-09-05, 205-token prompt, 2 cold runs per backend, cooled before each run): gpu decodes 25.13-25.8 tok/s against cpu's 12.29 (2.1x), prefill 940.7-1027.5 vs 166.7-175.3 tok/s, peak 1323 vs 2129 MB, engine init 6.3 vs 2.2 s. Same-device CPU control taken - a measured win for gpu on decode. On the Pixel 8a's Mali GPU this model deterministically dropped a date separator on number-dense input ('March 3rd, 2026.' -> 'March 32026.'); the same date vector run on this Adreno GPU came back intact ('The invoice came to $23,450 and it's due on March 3, 2026.', probe 2026-09-05, same binary), so that Mali defect does not reproduce here. The full 12-case exactness check was not repeated on Adreno; where byte-exactness is the job, cpu remains the conservative choice"
|
| 74 |
}
|
| 75 |
],
|
| 76 |
"requirements": {
|
|
|
|
| 145 |
"device": "iPhone 17 Pro",
|
| 146 |
"os": "iOS",
|
| 147 |
"backend": "gpu",
|
| 148 |
+
"runtime": "litert-lm (Swift)",
|
| 149 |
"prompt_tokens": 180,
|
| 150 |
"prefill_tps": 801,
|
| 151 |
"decode_tps": 32.0,
|
|
|
|
| 159 |
"device": "iPhone 17 Pro",
|
| 160 |
"os": "iOS",
|
| 161 |
"backend": "cpu",
|
| 162 |
+
"runtime": "litert-lm (Swift)",
|
| 163 |
"prompt_tokens": 180,
|
| 164 |
"prefill_tps": 281,
|
| 165 |
"decode_tps": 15.5,
|
|
|
|
| 184 |
"cache": "no",
|
| 185 |
"runs": 2,
|
| 186 |
"date": "2026-09-05",
|
| 187 |
+
"source": "205-token prompt with --benchmark, 2 cold runs (compile/weight caches deleted between runs, device cooled below 42 C before each run); GPU gate full LITERT_CL delegation (15474 ops across 12 subgraphs), generation verified on the gate prompt; the cpu row of the same date is the same-device control"
|
| 188 |
},
|
| 189 |
{
|
| 190 |
"device": "Galaxy S26 (SM-S942Q, Qualcomm SM8850, Adreno)",
|
|
|
|
| 200 |
"cache": "no",
|
| 201 |
"runs": 2,
|
| 202 |
"date": "2026-09-05",
|
| 203 |
+
"source": "same-device CPU control - identical binary, prompt file and flags, only --backend differs; 205-token prompt with --benchmark, 2 cold runs (compile/weight caches deleted between runs, device cooled below 42 C before each run); same-device control for the gpu row of the same date"
|
| 204 |
}
|
| 205 |
],
|
| 206 |
"known_issues": [
|
| 207 |
"Pixel 8a GPU: on number-dense input a date separator can be lost ('March 3rd, 2026.' -> 'March 32026.'), deterministic on repeat runs; the CPU backend on the same phone is byte-identical to macOS across all 12 test cases. Prefer CPU on this phone when exactness matters, GPU when time-to-first-token matters.",
|
| 208 |
"The released checkpoint itself drifts from its own upstream card's example table (it keeps 'Hmm,' / 'So' discourse markers the card drops); parity here is stated against HF fp32, not against those published examples.",
|
| 209 |
"All iPhone 17 Pro rows were measured with the device reporting a 'serious' thermal state; they are floors, not peaks."
|
| 210 |
+
],
|
| 211 |
+
"context_length": 4096
|
| 212 |
}
|
| 213 |
]
|
| 214 |
}
|