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 / MODIFICATIONS.md

kind-hill

ScribeKey Fast v1.1: new weights and cards

c484dad verified 13 days ago

preview code

Raw History Blame Contribute Delete

2.53 kB

# Changes made by ScribeKey

25 October 2026, ScribeKey Fast v1 release candidate (`v7-replay-tgt`).

- **Weights:** continued LoRA training of ScribeKey's Sotto v6 on 4,350 rows, then merged into BF16. The v6 starting point was reconstructed from its production Q8_0, not recovered original BF16 weights. The reconstruction re-exported bit-identically at Q8_0 tensor precision.
- **GGUF:** converted the merged checkpoint to F16 and quantized to Q8_0 with llama.cpp `4c1a0af40d88c7fbb3b15c85bf2e8016d1d5b64c`. The distribution filename is `scribekey-cleanup-fast-350m-v1-Q8_0.gguf`. Its bytes and embedded experiment metadata are unchanged from the evaluated artifact. Its embedded LFM chat template is inherited; use the documented raw prompt or the separate task-specific template.
- **Tokenizer template:** replaced the inherited LFM conversation template in the BF16 package with the exact single-turn Input/Output control-line task wrapper. The tokenizer vocabulary/tokens are unchanged. The same replacement is supplied separately as `prompt-template.jinja` with the GGUF.
- **Generation config:** made greedy decoding and a 256-token output limit explicit; no change to learned weights. Removed the stale local source-revision file from distribution; immutable ancestry and hashes are recorded in `provenance.json` instead.
- **Documentation and manifests:** added ScribeKey Fast naming, observed examples, evaluation limits, immutable provenance, required upstream notices and licence guidance. The original LFM licence and prior Sotto notice are retained verbatim.

The `model.safetensors` and GGUF files carry their modification notices through this accompanying document, `NOTICE` and `provenance.json`; their binary contents have not been rewritten merely to change branding. Distribute these notices with them.

- **Terms-only package update (`v1-rc1-attribution`):** added ADDITIONAL_TERMS.md with ScribeKey model naming/attribution requirements, and updated the cards and notice. Weight bytes, tokenizer, prompts and generation settings are unchanged; the original v1-rc1 tag and upstream LICENSE are preserved.
- **v1.1 (28 October 2026):** new weights. Four LoRA continuations of Sotto v6 (v1's mix plus targeted correction corpora, rounds 2 and 3) were merged and averaged into BF16, then converted to F16 and quantized to Q8_0 with the same llama.cpp revision. The distribution filename is `scribekey-cleanup-fast-350m-v1.1-Q8_0.gguf`. Tokenizer, templates, generation settings and terms are unchanged from v1.

	# Changes made by ScribeKey

	25 October 2026, ScribeKey Fast v1 release candidate (`v7-replay-tgt`).

	- Weights: continued LoRA training of ScribeKey's Sotto v6 on 4,350 rows, then merged into BF16. The v6 starting point was reconstructed from its production Q8_0, not recovered original BF16 weights. The reconstruction re-exported bit-identically at Q8_0 tensor precision.
	- GGUF: converted the merged checkpoint to F16 and quantized to Q8_0 with llama.cpp `4c1a0af40d88c7fbb3b15c85bf2e8016d1d5b64c`. The distribution filename is `scribekey-cleanup-fast-350m-v1-Q8_0.gguf`. Its bytes and embedded experiment metadata are unchanged from the evaluated artifact. Its embedded LFM chat template is inherited; use the documented raw prompt or the separate task-specific template.
	- Tokenizer template: replaced the inherited LFM conversation template in the BF16 package with the exact single-turn Input/Output control-line task wrapper. The tokenizer vocabulary/tokens are unchanged. The same replacement is supplied separately as `prompt-template.jinja` with the GGUF.
	- Generation config: made greedy decoding and a 256-token output limit explicit; no change to learned weights. Removed the stale local source-revision file from distribution; immutable ancestry and hashes are recorded in `provenance.json` instead.
	- Documentation and manifests: added ScribeKey Fast naming, observed examples, evaluation limits, immutable provenance, required upstream notices and licence guidance. The original LFM licence and prior Sotto notice are retained verbatim.

	The `model.safetensors` and GGUF files carry their modification notices through this accompanying document, `NOTICE` and `provenance.json`; their binary contents have not been rewritten merely to change branding. Distribute these notices with them.

	- Terms-only package update (`v1-rc1-attribution`): added ADDITIONAL_TERMS.md with ScribeKey model naming/attribution requirements, and updated the cards and notice. Weight bytes, tokenizer, prompts and generation settings are unchanged; the original v1-rc1 tag and upstream LICENSE are preserved.

	- v1.1 (28 October 2026): new weights. Four LoRA continuations of Sotto v6 (v1's mix plus targeted correction corpora, rounds 2 and 3) were merged and averaged into BF16, then converted to F16 and quantized to Q8_0 with the same llama.cpp revision. The distribution filename is `scribekey-cleanup-fast-350m-v1.1-Q8_0.gguf`. Tokenizer, templates, generation settings and terms are unchanged from v1.