kind-hill commited on
Commit
c2cdf05
·
verified ·
1 Parent(s): ffa008c

Publish v1 card

Browse files
Files changed (2) hide show
  1. README.md +4 -4
  2. SHA256SUMS +1 -1
README.md CHANGED
@@ -25,7 +25,7 @@ datasets:
25
 
26
  A 350M-class English text model for cleaning speech-to-text transcripts on a device. It removes fillers and repetitions, applies spoken corrections, and adjusts punctuation, casing and layout using a control line. It takes **text**, not audio. It is a dedicated cleanup model, not a general assistant.
27
 
28
- **Private release candidate: `v1-rc1-attribution`.** This package stages experiment `v7-replay-tgt`. The Q8_0 has offline evaluation; a direct-helper S25 recheck has completed (short p50/p95 155/178 ms; long 1,702/2,718 ms with Parakeet resident), while midrange and product end-to-end qualification remain open. Private staging does not make it ScribeKey's production default. Two public-input BF16 CPU smoke cases also ran twice with identical outputs (`bf16-smoke.json`); this is an inference check, not a BF16 quality benchmark.
29
 
30
  ## Model overview
31
 
@@ -51,7 +51,7 @@ import torch
51
  from transformers import AutoModelForCausalLM, AutoTokenizer
52
 
53
  MODEL = "scribekey/scribekey-cleanup-fast"
54
- REVISION = "v1-rc1-attribution" # Pin the reviewed commit SHA for a deployment.
55
 
56
  tokenizer = AutoTokenizer.from_pretrained(MODEL, revision=REVISION)
57
  model = AutoModelForCausalLM.from_pretrained(
@@ -145,7 +145,7 @@ The following are **Q8_0 results at the ScribeKey app boundary**, not a BF16 ben
145
 
146
  The intervals do not establish superiority over Sotto or equivalence to S1-mini. The candidate met the experiment's provisional non-regression criteria; the broader release bar remains unresolved. These values are specific benchmark metrics, not universal accuracy rates. The new-value rate is a real limitation: check names, dates, amounts, negations and important details before using the output. Keep a recoverable raw transcript.
147
 
148
- Phone results often quoted for this family are for **the preceding `v7-replay` artifact**, not this package: S25 Ultra long-input warm p50/p95 1,696/2,101 ms, short-input 156/178 ms, peak combined PSS 1,567 MiB on the measured app/ASR build. They are context, not final-artifact performance claims. Midrange performance and a minimum-device guarantee are unverified. The draft catalogue retains a conservative 6 GB device-memory floor.
149
 
150
  ## Training and provenance
151
 
@@ -180,4 +180,4 @@ Some training sources carry CC BY or CC BY-SA licences. They are credited separa
180
  }
181
  ```
182
 
183
- The `v1-rc1-attribution` update changes documentation and terms only; weight bytes and prompt/runtime controls are unchanged from `v1-rc1`. The original tag is preserved.
 
25
 
26
  A 350M-class English text model for cleaning speech-to-text transcripts on a device. It removes fillers and repetitions, applies spoken corrections, and adjusts punctuation, casing and layout using a control line. It takes **text**, not audio. It is a dedicated cleanup model, not a general assistant.
27
 
28
+ **Release `v1`** (experiment `v7-replay-tgt`). ScribeKey Fast is ScribeKey's on-device cleanup model, replacing Sotto. The Q8_0 has offline evaluation and a direct-helper S25 Ultra check of this exact artifact (short p50/p95 155/178 ms; long 1,702/2,718 ms with Parakeet resident); midrange and product end-to-end performance have not been measured. Two public-input BF16 CPU smoke cases also ran twice with identical outputs (`bf16-smoke.json`); this is an inference check, not a BF16 quality benchmark.
29
 
30
  ## Model overview
31
 
 
51
  from transformers import AutoModelForCausalLM, AutoTokenizer
52
 
53
  MODEL = "scribekey/scribekey-cleanup-fast"
54
+ REVISION = "v1" # Pin the commit SHA for a deployment.
55
 
56
  tokenizer = AutoTokenizer.from_pretrained(MODEL, revision=REVISION)
57
  model = AutoModelForCausalLM.from_pretrained(
 
145
 
146
  The intervals do not establish superiority over Sotto or equivalence to S1-mini. The candidate met the experiment's provisional non-regression criteria; the broader release bar remains unresolved. These values are specific benchmark metrics, not universal accuracy rates. The new-value rate is a real limitation: check names, dates, amounts, negations and important details before using the output. Keep a recoverable raw transcript.
147
 
148
+ Phone results often quoted for this family are for **the preceding `v7-replay` artifact**, not this package: S25 Ultra long-input warm p50/p95 1,696/2,101 ms, short-input 156/178 ms, peak combined PSS 1,567 MiB on the measured app/ASR build. They are context, not final-artifact performance claims. Midrange performance and a minimum-device guarantee are unverified. ScribeKey's catalogue keeps a conservative 6 GB device-memory floor.
149
 
150
  ## Training and provenance
151
 
 
180
  }
181
  ```
182
 
183
+ The `v1` release is the public `v1-rc1-attribution` package with updated card wording; weight bytes, terms and prompt/runtime controls are unchanged from `v1-rc1`. The original RC tags are preserved.
SHA256SUMS CHANGED
@@ -2,7 +2,7 @@
2
  4d28ca14dedc0b3d0fcc2b3339f0e79931faa33874f3d24f522183a8fc70068c LICENSE
3
  09dfb7155ec9f316cd19e2cf68b818b04344be000e44f29d7779cd200743337c MODIFICATIONS.md
4
  4d0d40d15ca51c17c0d9b61ed5a0f9c06ed3640a106ae20cfea4a8f36cd02b73 NOTICE
5
- dfcf8e9cd451fb1d5ed867e59075e6c547bcf4c09f34ae9d4bc2cf990ae762b3 README.md
6
  75ba5b790d851b4d8084fbd0eb8624cfd3eb154bc5182fbb2c81b91ef0aea948 bf16-smoke.json
7
  d0ca71c262852fd7020e4f77e7f424daddbe052637a05fbe63e50a2b16980c25 chat_template.jinja
8
  13ab9863955d2016c32417ff97b1a9840951d2481b6d3095f7ae98d10611a7d2 config.json
 
2
  4d28ca14dedc0b3d0fcc2b3339f0e79931faa33874f3d24f522183a8fc70068c LICENSE
3
  09dfb7155ec9f316cd19e2cf68b818b04344be000e44f29d7779cd200743337c MODIFICATIONS.md
4
  4d0d40d15ca51c17c0d9b61ed5a0f9c06ed3640a106ae20cfea4a8f36cd02b73 NOTICE
5
+ e3fffa6fa8c673148e5d7db9ca5fea2ed90428bb950a11d42aaf436e22011d9b README.md
6
  75ba5b790d851b4d8084fbd0eb8624cfd3eb154bc5182fbb2c81b91ef0aea948 bf16-smoke.json
7
  d0ca71c262852fd7020e4f77e7f424daddbe052637a05fbe63e50a2b16980c25 chat_template.jinja
8
  13ab9863955d2016c32417ff97b1a9840951d2481b6d3095f7ae98d10611a7d2 config.json