Instructions to use scribekey/scribekey-cleanup-fast with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use scribekey/scribekey-cleanup-fast with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="scribekey/scribekey-cleanup-fast")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("scribekey/scribekey-cleanup-fast")
model = AutoModelForCausalLM.from_pretrained("scribekey/scribekey-cleanup-fast", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use scribekey/scribekey-cleanup-fast with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "scribekey/scribekey-cleanup-fast"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "scribekey/scribekey-cleanup-fast",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/scribekey/scribekey-cleanup-fast

SGLang

How to use scribekey/scribekey-cleanup-fast with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "scribekey/scribekey-cleanup-fast" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "scribekey/scribekey-cleanup-fast",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "scribekey/scribekey-cleanup-fast" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "scribekey/scribekey-cleanup-fast",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use scribekey/scribekey-cleanup-fast with Docker Model Runner:
```
docker model run hf.co/scribekey/scribekey-cleanup-fast
```

scribekey-cleanup-fast / NOTICE

kind-hill

Add ScribeKey Fast naming and attribution terms; retain inherited LFM obligations

ffa008c verified 15 days ago

Raw History Blame Contribute Delete

5.71 kB

ScribeKey Fast by ScribeKey — v1 release candidate, 25 October 2026

Copyright 2026 ScribeKey contributors (ScribeKey modifications).
Based on LFM2.5-350M-Base by Liquid AI, Inc., and Sotto by juanquivilla.
ScribeKey Fast Model Terms v1.0 (ADDITIONAL_TERMS.md) add naming and
attribution conditions. The weights also inherit LFM Open License v1.0 (LICENSE). Sotto's MIT notice is
retained verbatim below. No upstream endorsement is implied.

Changes made by ScribeKey: continued LoRA training of reconstructed Sotto v6
on 4,350 rows, BF16 merge, F16 conversion and Q8_0 quantization. The tokenizer
task template and greedy generation configuration in the BF16 package were
updated. See MODIFICATIONS.md and provenance.json for files and hashes.

Training-source and pipeline attributions, aggregated from the actual final
training mix (no training text or source audio is distributed here):

- DeepSeek V4.1 Flash (deepseek-ai/DeepSeek-V4.1-Flash@314113f8a7dadb388215396ba5fbee2c610a04d6), MIT
- External: datasets/MagicNoThief/handy-dictation-editing@7ad0b0758559cf70d319a3cb1680b08f9ea204d8, CC BY 4.0 (dataset card at the pinned revision)
- External: datasets/ductai199x/dictation-speech@692f01ad28a5d6f26bde093b6d717e89cfae0067, CC BY-SA 4.0 (dataset card at the pinned revision)
- GLM-5.3 Flash (zai-org/GLM-5.3-Flash@46a050bec13002ad29ee795c16e61cf4fa3c4a9c), MIT
- Kokoro v1.0 by hexgrad (sherpa-onnx export by csukuangfj), Apache-2.0
- MLCommons: datasets/MLCommons/peoples_speech@f10597c5d3d3a63f8b6827701297c3afdf178272, CC BY 4.0 (the clean and dirty subsets; the _sa subsets are never used)
- MiMo-V2.6-Flash (XiaomiMiMo/MiMo-V2.6-Flash-RL@dcd3fd83fb181622fbc606fbb1709901982a62c2), MIT
- Moonshine Base v2 by Useful Sensors (sherpa-onnx export by csukuangfj), MIT
- Mozilla Foundation / Common Voice: cmu5nqn1h00vwmi07b4dbk085@2026-09-17T15:01:25.158Z, CC0-1.0
- NeMo text-processing 1.2.0 by NVIDIA, Apache-2.0
- Parakeet TDT 0.6B v2 by NVIDIA (sherpa-onnx export by csukuangfj), CC-BY-4.0

- SottoASR Transcript Cleanup Dataset by juanquivilla, declared MIT:
  https://huggingface.co/datasets/juanquivilla/sotto-transcript-cleanup
  revision 183cc8fd58532f13fa192980185214de1bcd5acc (2,000 replay rows).
- llama.cpp by its contributors (MIT), conversion and quantization:
  https://github.com/ggml-org/llama.cpp/tree/4c1a0af40d88c7fbb3b15c85bf2e8016d1d5b64c

Source links and terms:
- Handy, MagicNoThief:
  https://huggingface.co/datasets/MagicNoThief/handy-dictation-editing/tree/7ad0b0758559cf70d319a3cb1680b08f9ea204d8
- dictation-speech, ductai199x:
  https://huggingface.co/datasets/ductai199x/dictation-speech/tree/692f01ad28a5d6f26bde093b6d717e89cfae0067
- People's Speech, MLCommons (clean and dirty only):
  https://huggingface.co/datasets/MLCommons/peoples_speech/tree/f10597c5d3d3a63f8b6827701297c3afdf178272
- Common Voice Spontaneous Speech 5.0 English, Mozilla Foundation and contributors:
  https://datacollective.mozillafoundation.org/datasets/cmu5nqn1h00vwmi07b4dbk085
  archive SHA-256 390b5a54c9afe0cc01da039ad206248f85682f247dd2b27d4cc0ab9a68e860b6
- CC BY 4.0: https://creativecommons.org/licenses/by/4.0/
- CC BY-SA 4.0: https://creativecommons.org/licenses/by-sa/4.0/
- CC0 1.0: https://creativecommons.org/publicdomain/zero/1.0/

Source recordings/text were transcribed, filtered, assigned controls and used
for training. Source licences apply to their respective material; this NOTICE
does not relicense that material or resolve whether share-alike obligations
extend to trained weights. Source files are not redistributed in this package.
Attributions to teacher, TTS and ASR models describe training tools, not weights
bundled here. Hidden CleanBench was used for evaluation, never as training data.
Inherited third-party training declarations have not been independently rebuilt.

The following predecessor NOTICE is preserved verbatim:
------------------------------------------------------
Sotto 350M, ScribeKey tune, GGUF

This repository contains a fine-tune, in GGUF (Q8_0), of:

  Sotto transcript cleanup, LFM2.5-350M
  https://huggingface.co/juanquivilla/sotto-cleanup-lfm25-350m
  revision 6df6f019170b8b55333c047b901886a51750a965
  by juanquivilla, released under the MIT License.

which is a fine-tune of:

  LFM2.5-350M-Base
  https://huggingface.co/LiquidAI/LFM2.5-350M-Base
  Copyright Liquid AI, Inc., licensed under the LFM Open License v1.0 (see LICENSE).

Changes made by ScribeKey: a LoRA adapter was trained on ScribeKey's own synthetic dictation
data and merged into the weights, which were then converted to GGUF with llama.cpp's
convert_hf_to_gguf.py and quantised to Q8_0.

MIT License (applies to the Sotto fine-tune)

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and
associated documentation files (the "Software"), to deal in the Software without restriction,
including without limitation the rights to use, copy, modify, merge, publish, distribute,
sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or
substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT
NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES
OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

	ScribeKey Fast by ScribeKey — v1 release candidate, 25 October 2026

	Copyright 2026 ScribeKey contributors (ScribeKey modifications).
	Based on LFM2.5-350M-Base by Liquid AI, Inc., and Sotto by juanquivilla.
	ScribeKey Fast Model Terms v1.0 (ADDITIONAL_TERMS.md) add naming and
	attribution conditions. The weights also inherit LFM Open License v1.0 (LICENSE). Sotto's MIT notice is
	retained verbatim below. No upstream endorsement is implied.

	Changes made by ScribeKey: continued LoRA training of reconstructed Sotto v6
	on 4,350 rows, BF16 merge, F16 conversion and Q8_0 quantization. The tokenizer
	task template and greedy generation configuration in the BF16 package were
	updated. See MODIFICATIONS.md and provenance.json for files and hashes.

	Training-source and pipeline attributions, aggregated from the actual final
	training mix (no training text or source audio is distributed here):

	- DeepSeek V4.1 Flash (deepseek-ai/DeepSeek-V4.1-Flash@314113f8a7dadb388215396ba5fbee2c610a04d6), MIT
	- External: datasets/MagicNoThief/handy-dictation-editing@7ad0b0758559cf70d319a3cb1680b08f9ea204d8, CC BY 4.0 (dataset card at the pinned revision)
	- External: datasets/ductai199x/dictation-speech@692f01ad28a5d6f26bde093b6d717e89cfae0067, CC BY-SA 4.0 (dataset card at the pinned revision)
	- GLM-5.3 Flash (zai-org/GLM-5.3-Flash@46a050bec13002ad29ee795c16e61cf4fa3c4a9c), MIT
	- Kokoro v1.0 by hexgrad (sherpa-onnx export by csukuangfj), Apache-2.0
	- MLCommons: datasets/MLCommons/peoples_speech@f10597c5d3d3a63f8b6827701297c3afdf178272, CC BY 4.0 (the clean and dirty subsets; the _sa subsets are never used)
	- MiMo-V2.6-Flash (XiaomiMiMo/MiMo-V2.6-Flash-RL@dcd3fd83fb181622fbc606fbb1709901982a62c2), MIT
	- Moonshine Base v2 by Useful Sensors (sherpa-onnx export by csukuangfj), MIT
	- Mozilla Foundation / Common Voice: cmu5nqn1h00vwmi07b4dbk085@2026-09-17T15:01:25.158Z, CC0-1.0
	- NeMo text-processing 1.2.0 by NVIDIA, Apache-2.0
	- Parakeet TDT 0.6B v2 by NVIDIA (sherpa-onnx export by csukuangfj), CC-BY-4.0

	- SottoASR Transcript Cleanup Dataset by juanquivilla, declared MIT:
	https://huggingface.co/datasets/juanquivilla/sotto-transcript-cleanup
	revision 183cc8fd58532f13fa192980185214de1bcd5acc (2,000 replay rows).
	- llama.cpp by its contributors (MIT), conversion and quantization:
	https://github.com/ggml-org/llama.cpp/tree/4c1a0af40d88c7fbb3b15c85bf2e8016d1d5b64c

	Source links and terms:
	- Handy, MagicNoThief:
	https://huggingface.co/datasets/MagicNoThief/handy-dictation-editing/tree/7ad0b0758559cf70d319a3cb1680b08f9ea204d8
	- dictation-speech, ductai199x:
	https://huggingface.co/datasets/ductai199x/dictation-speech/tree/692f01ad28a5d6f26bde093b6d717e89cfae0067
	- People's Speech, MLCommons (clean and dirty only):
	https://huggingface.co/datasets/MLCommons/peoples_speech/tree/f10597c5d3d3a63f8b6827701297c3afdf178272
	- Common Voice Spontaneous Speech 5.0 English, Mozilla Foundation and contributors:
	https://datacollective.mozillafoundation.org/datasets/cmu5nqn1h00vwmi07b4dbk085
	archive SHA-256 390b5a54c9afe0cc01da039ad206248f85682f247dd2b27d4cc0ab9a68e860b6
	- CC BY 4.0: https://creativecommons.org/licenses/by/4.0/
	- CC BY-SA 4.0: https://creativecommons.org/licenses/by-sa/4.0/
	- CC0 1.0: https://creativecommons.org/publicdomain/zero/1.0/

	Source recordings/text were transcribed, filtered, assigned controls and used
	for training. Source licences apply to their respective material; this NOTICE
	does not relicense that material or resolve whether share-alike obligations
	extend to trained weights. Source files are not redistributed in this package.
	Attributions to teacher, TTS and ASR models describe training tools, not weights
	bundled here. Hidden CleanBench was used for evaluation, never as training data.
	Inherited third-party training declarations have not been independently rebuilt.

	The following predecessor NOTICE is preserved verbatim:
	------------------------------------------------------
	Sotto 350M, ScribeKey tune, GGUF

	This repository contains a fine-tune, in GGUF (Q8_0), of:

	Sotto transcript cleanup, LFM2.5-350M
	https://huggingface.co/juanquivilla/sotto-cleanup-lfm25-350m
	revision 6df6f019170b8b55333c047b901886a51750a965
	by juanquivilla, released under the MIT License.

	which is a fine-tune of:

	LFM2.5-350M-Base
	https://huggingface.co/LiquidAI/LFM2.5-350M-Base
	Copyright Liquid AI, Inc., licensed under the LFM Open License v1.0 (see LICENSE).

	Changes made by ScribeKey: a LoRA adapter was trained on ScribeKey's own synthetic dictation
	data and merged into the weights, which were then converted to GGUF with llama.cpp's
	convert_hf_to_gguf.py and quantised to Q8_0.

	MIT License (applies to the Sotto fine-tune)

	Permission is hereby granted, free of charge, to any person obtaining a copy of this software and
	associated documentation files (the "Software"), to deal in the Software without restriction,
	including without limitation the rights to use, copy, modify, merge, publish, distribute,
	sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is
	furnished to do so, subject to the following conditions:

	The above copyright notice and this permission notice shall be included in all copies or
	substantial portions of the Software.

	THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT
	NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
	NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES
	OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
	CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.