Instructions to use danielrosehill/S1-ASR-Cleanup with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use danielrosehill/S1-ASR-Cleanup with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="danielrosehill/S1-ASR-Cleanup") 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("danielrosehill/S1-ASR-Cleanup") model = AutoModelForCausalLM.from_pretrained("danielrosehill/S1-ASR-Cleanup", 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
- llama.cpp
How to use danielrosehill/S1-ASR-Cleanup with llama.cpp:
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh # Start a local OpenAI-compatible server with a web UI: llama serve -hf danielrosehill/S1-ASR-Cleanup:Q4_K_M # Run inference directly in the terminal: llama cli -hf danielrosehill/S1-ASR-Cleanup:Q4_K_M
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf danielrosehill/S1-ASR-Cleanup:Q4_K_M # Run inference directly in the terminal: llama cli -hf danielrosehill/S1-ASR-Cleanup:Q4_K_M
Build from source code
git clone https://github.com/ggml-org/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-app # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama serve -hf danielrosehill/S1-ASR-Cleanup:Q4_K_M # Run inference directly in the terminal: ./build/bin/llama cli -hf danielrosehill/S1-ASR-Cleanup:Q4_K_M
Use Docker
docker model run hf.co/danielrosehill/S1-ASR-Cleanup:Q4_K_M
- Llama new
- LM Studio
- Jan
- vLLM
How to use danielrosehill/S1-ASR-Cleanup with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "danielrosehill/S1-ASR-Cleanup" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "danielrosehill/S1-ASR-Cleanup", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/danielrosehill/S1-ASR-Cleanup:Q4_K_M
- SGLang
How to use danielrosehill/S1-ASR-Cleanup 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 "danielrosehill/S1-ASR-Cleanup" \ --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": "danielrosehill/S1-ASR-Cleanup", "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 "danielrosehill/S1-ASR-Cleanup" \ --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": "danielrosehill/S1-ASR-Cleanup", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Ollama
How to use danielrosehill/S1-ASR-Cleanup with Ollama:
ollama run hf.co/danielrosehill/S1-ASR-Cleanup:Q4_K_M
- Unsloth Desktop
- Pi
How to use danielrosehill/S1-ASR-Cleanup with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf danielrosehill/S1-ASR-Cleanup:Q4_K_M
Configure the model in Pi
# Install Pi: npm install -g @earendil-works/pi-coding-agent # Add to ~/.pi/agent/models.json: { "providers": { "llama-cpp": { "baseUrl": "http://localhost:8080/v1", "api": "openai-completions", "apiKey": "none", "models": [ { "id": "danielrosehill/S1-ASR-Cleanup:Q4_K_M" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Docker Model Runner
How to use danielrosehill/S1-ASR-Cleanup with Docker Model Runner:
docker model run hf.co/danielrosehill/S1-ASR-Cleanup:Q4_K_M
- Lemonade
How to use danielrosehill/S1-ASR-Cleanup with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull danielrosehill/S1-ASR-Cleanup:Q4_K_M
Run and chat with the model
lemonade run user.S1-ASR-Cleanup-Q4_K_M
List all available models
lemonade list
- Hermes Agent
How to use danielrosehill/S1-ASR-Cleanup with Hermes Agent:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf danielrosehill/S1-ASR-Cleanup:Q4_K_M
Configure Hermes
# Install Hermes: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash hermes setup # Point Hermes at the local server: hermes config set model.provider custom hermes config set model.base_url http://127.0.0.1:8080/v1 hermes config set model.default danielrosehill/S1-ASR-Cleanup:Q4_K_M
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use danielrosehill/S1-ASR-Cleanup with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf danielrosehill/S1-ASR-Cleanup:Q4_K_M
Configure OpenClaw
# Install OpenClaw: npm install -g openclaw@latest # Register the local server and set it as the default model: openclaw onboard --non-interactive --mode local \ --auth-choice custom-api-key \ --custom-base-url http://127.0.0.1:8080/v1 \ --custom-model-id "danielrosehill/S1-ASR-Cleanup:Q4_K_M" \ --custom-provider-id llama-cpp \ --custom-compatibility openai \ --custom-text-input \ --accept-risk \ --skip-health
Run OpenClaw
openclaw agent --local --agent main --message "Hello from Hugging Face"
S1-ASR-Cleanup (v1.0): a fine-tune of S1-mini by Superwhisper
S1-ASR-Cleanup is a fine-tune of S1-mini by Superwhisper (superwhisper/s1-mini,
Qwen3-0.6B) that cleans raw speech-to-text dictation the way one user edits his own. One standalone model: the LoRA
is merged in, nothing else to load.
What it does differently from stock S1-mini
- Paragraphs. Long dictation is split into short paragraphs, one thought each; greetings and sign-offs on their own lines. Stock S1-mini returns one block.
- Dates, deterministic. "the twelfth of March", "March twelfth" ->
March 12; "Thursday the twelfth of March" ->Thursday, March 12; a bare "by the fifteenth" ->by the 15th. - Money. "two thousand two hundred shekels" ->
2,200 NIS; "forty bucks" ->$40;โฌ35,$2 million. - Informal contractions expanded. gonna, wanna, gotta, hafta, lemme, dunno, 'cause, outta, tryna, c'mon, 'em,
shoulda and the like become
going to,want to,have to, ...; ordinary contractions (I'm, don't) stay. - Spelled-out clarifications dropped. "Ronen, that's R O N E N" ->
Ronen. - Mid-sentence capitals. The capital (and stray full stop) a recogniser leaves after a pause is lowercased:
"we could Use those" ->
we could use those; real capitals (names, I, WhatsApp) stay. - Also: disfluencies and false starts, self-corrections resolved to the final intent, sentences broken at a pause rejoined, restated points and asides to people in the room dropped. Romanised Hebrew words and tech names kept.
Examples
Outputs of this model and of stock S1-mini (same prompt, greedy) on held-out pairs, not cherry-picked beyond one per behaviour:
asides
Raw transcript:
Hey Maya the technician just looked at the washing machine in the basement and he thinks the motor is completely dead, but while we were talking the quote for the roof repair comes to $3,400 โ yes, that's fine โ which is slightly higher than the first estimate we received, anyway I need you to message the landlord on WhatsApp check if our renters insurance covers the drum replacement and put a placeholder on Google Calendar for Thursday morning just so I do not double book myself between nine and eleven before the standup.
Stock S1-mini:
Hey Maya, the technician just looked at the washing machine in the basement, and he thinks the motor is completely dead, but while we were talking the quote for the roof repair comes to $3,400, yes, that's fine, which is slightly higher than the first estimate we received.
Anyway, I need you to message the landlord on WhatsApp, check if our renters insurance covers the drum replacement, and put a placeholder on Google Calendar for Thursday morning just so I do not double book myself between 9 and 11 before the standup.
S1-ASR-Cleanup:
Hey Maya,
The technician just looked at the washing machine in the basement and he thinks the motor is completely dead. But while we were talking, the quote for the roof repair comes to $3,400, which is slightly higher than the first estimate we received.
Anyway, I need you to message the landlord on WhatsApp check if our renters insurance covers the drum replacement, and put a placeholder on Google Calendar for Thursday morning just so I do not double-book myself between nine and eleven before the standup.
currency
Raw transcript:
Prompt for Claude compare the Ecobee Premium against the Nest learning thermostat for a Tel Aviv apartment I need to know if paying about seven hundred shekels for the Ecobee is justified over the cheaper version which goes for maybe four hundred fifty shekels or even forty bucks used on eBay wait where was I yeah compare their Zigbee integration with Home Assistant also mention the background of the original founders because didn't they sold the company for 40 million pounds back in 2018 check if either works reliably with the local mazgan setup and export the final recommendation as a clean markdown table.
Stock S1-mini:
Prompt for Claude: compare the Ecobee Premium against the Nest Learning thermostat for a Tel Aviv apartment. I need to know if paying about 700 shekels for the Ecobee is justified over the cheaper version, which goes for maybe 450 shekels or even 40 bucks used on eBay. Wait, where was I?
Yeah, compare their Zigbee integration with Home Assistant. Also, mention the background of the original founders, because didn't they sell the company for ยฃ40 million back in 2018? Check if either works reliably with the local Mazgan setup, and export the final recommendation as a clean markdown table.
S1-ASR-Cleanup:
Prompt for Claude: compare the Ecobee Premium against the Nest learning thermostat for a Tel Aviv apartment.
I need to know if paying about 700 NIS for the Ecobee is justified over the cheaper version, which goes for maybe 450 NIS or even $40 used on eBay. Compare their Zigbee integration with Home Assistant. Also mention the background of the original founders, because didn't they sell the company for ยฃ40 million back in 2018? Check if either works reliably with the local mazgan setup. And export the final recommendation as a clean markdown table.
dates
Raw transcript:
Hi Sharon just checking about the annual tax form deadline because Mas Hachnasah moved it to the twelfth of March instead of March the nineteenth so could you send me the updated PDF ping Guy on WhatsApp about his GitHub contractor invoices check if we still owe eleven percent withholding and confirm if could the glazier come by seven o'clock on September fourth for the office door thanks.
Stock S1-mini:
Hi Sharon, just checking about the annual tax form deadline because Mas Hachnasah moved it to the 12th of March instead of March the 19th, so could you send me the updated PDF ping Guy on WhatsApp about his GitHub contractor invoices? Check if we still owe 11% withholding, and confirm if could the glazier come by 7 o'clock on September 4 for the office door. Thanks.
S1-ASR-Cleanup:
Hi Sharon,
Just checking about the annual tax form deadline because Mas Hachnasah moved it to March 12 instead of March 19.
Could you send me the updated PDF ping Guy on WhatsApp about his GitHub contractor invoices? Check if we still owe 11% withholding and confirm if could the glazier come by 7:00 on September 4 for the office door?
Thanks.
disfluency
Raw transcript:
Prompt for Claude draft an update for WhatsApp about the charity fundraiser I mean the school bitsoah drive for Tuesday write a draft to the parents saying hi Sarah just a quick note to say that that the delivery arrived and it looks like rain for the outdoor bake sale so move the tables into the into the main hall sync the volunteer rota to Google Calendar and another please add a dark mode toggle to the settings page on the donation portal so people can read it easily tonight.
Stock S1-mini:
Prompt for Claude: draft an update for WhatsApp about the school bitsoah drive for Tuesday. Write a draft to the parents saying hi, Sarah, just a quick note to say that the delivery arrived, and it looks like rain for the outdoor bake sale. So move the tables into the main hall, sync the volunteer rota to Google Calendar, and another, please, add a dark mode toggle to the settings page on the donation portal so people can read it easily tonight.
S1-ASR-Cleanup:
Prompt for Claude: draft an update for WhatsApp about the school bitsoah drive for Tuesday.
Write a draft to the parents saying: "Hi Sarah, just a quick note to say that the delivery arrived, and it looks like rain for the outdoor bake sale."
So move the tables into the main hall, sync the volunteer rota to Google Calendar, and another, please add a dark mode toggle to the settings page on the donation portal so people can read it easily tonight.
gonna
Raw transcript:
Hey Google put in my calendar for Thursday night at seven cook dinner for Maya to make up for the disaster. I gotta buy fresh ingredients on the way home outta the office 'cause the shakshuka last week was completely ruined. I shoulda checked the eggs first, plus I sorta forgot to tell her about the change of plan so everything went wrong. Thanks bye.
Stock S1-mini:
Hey Google, put in my calendar for Thursday night at 7:45am for a cook dinner for Maya to make up for the disaster. I got to buy fresh ingredients on the way home out from the office, 'cause the shakshuka last week was completely ruined. I shoulda checked the eggs first, plus I sorta forgot to tell her about the change of plan, so everything went wrong. Thanks, bye.
S1-ASR-Cleanup:
Put in my calendar for Thursday night at 7:00 for Maya to make up for the disaster.
I have to buy fresh ingredients on the way home out of the office because the shakshuka last week was completely ruined. I should have checked the eggs first, and I sort of forgot to tell her about the change of plan, so everything went wrong.
Thanks, bye.
greeting
Raw transcript:
Hi Hannah it's me the washing machine is leaking all over the utility balcony again so I had to shut off the main valve thanks for sending that repair contact yesterday by the way I just sent him a message on WhatsApp to see if he can come by tomorrow morning before my first sprint planning meeting check Google Calendar to see if the timing works for you thanks yalla see you love you Dan
Stock S1-mini:
Hi Hannah, it's me. The washing machine is leaking all over the utility balcony again, so I had to shut off the main valve. Thanks for sending that repair contact yesterday.
By the way, I just sent him a message on WhatsApp to see if he can come by tomorrow morning before my first sprint planning meeting. Check Google Calendar to see if the timing works for you. Thanks, yalla. See you. Love you, Dan.
S1-ASR-Cleanup:
Hi Hannah, it's me.
The washing machine is leaking all over the utility balcony again, so I had to shut off the main valve. Thanks for sending that repair contact yesterday, by the way.
I just sent him a message on WhatsApp to see if he can come by tomorrow morning before my first sprint planning meeting. Check Google Calendar to see if the timing works for you.
Thanks!
Yalla, see you.
Love you,
Dan.
pausebreaks
Raw transcript:
Hi Maya hope you are having a good week. I looked at the staging link you sent on WhatsApp for the photography portfolio. The question is really. How much we want to spend on it. Because the gallery grid needs some custom work on GitHub. Before we push to production. We also need to talk about hosting costs. Good question. I'd have to check. What were we paying, where was I, right let's jump on Google Calendar tomorrow at three to finalise the launch date. Best, Daniel.
Stock S1-mini:
Hi Maya, hope you are having a good week. I looked at the staging link you sent on WhatsApp for the photography portfolio. The question is really, how much we want to spend on it, because the gallery grid needs some custom work on GitHub before we push to production.
We also need to talk about hosting costs. Good question. I'd have to check. What were we paying, where was I? Right, let's jump on Google Calendar tomorrow at 3 to finalise the launch date.
Best, Daniel.
S1-ASR-Cleanup:
Hi Maya, hope you are having a good week.
I looked at the staging link you sent on WhatsApp for the photography portfolio. The question is really: how much we want to spend on it, because the gallery grid needs some custom work on GitHub before we push to production.
We also need to talk about hosting costs.
Good question. I'd have to check.
What were we paying? Where was I, right. Let's jump on Google Calendar tomorrow at 3:00 to finalise the launch date.
Best,
Daniel
repetition
Raw transcript:
Hey Maya, just letting you know I have to take Milo to the vet in Florentin tomorrow morning at ten. I put it on Google Calendar already so you can see when I am blocked. I put it on Google Calendar so you will see the block. His ear infection is back again so the vet needs to check his ear. We also need to get his yearly rabies shot and his pinuk after. I will keep my WhatsApp open if anything urgent comes up on the sprint. Thanks, Daniel.
Stock S1-mini:
Hey Maya, just letting you know I have to take Milo to the vet in Florentin tomorrow morning at 10. I put it on Google Calendar already, so you can see when I am blocked. I put it on Google Calendar so you will see the block. His ear infection is back again, so the vet needs to check his ear. We also need to get his yearly rabies shot and his pinuk after. I will keep my WhatsApp open if anything urgent comes up on the sprint. Thanks, Daniel.
S1-ASR-Cleanup:
Hey Maya,
Just letting you know I have to take Milo to the vet in Florentin tomorrow morning at 10:00. I put it on Google Calendar already, so you can see when I am blocked.
His ear infection is back again, so the vet needs to check his ear. We also need to get his yearly rabies shot and his pinuk after.
I will keep my WhatsApp open if anything urgent comes up on the sprint.
Thanks,
Daniel
selfcorrect
Raw transcript:
Hey Yoni, quick update on the the kitchen renovation. The contractor wants to come Thursday, no wait, Wednesday, actually let's stick with Thursday morning at nine. The blue car, I mean the silver one, is blocking the driveway so tell him to park on the street. Also he gave me the quote for the tiles, it was eleven, no wait, twelve thousand shekels, but he says he can drop it to ten if we pay cash. Put it on Google Calendar so we don't forget. Talk soon, Daniel.
Stock S1-mini:
Hey Yoni, quick update on the kitchen renovation. The contractor wants to come Wednesday, let's stick with Thursday morning at 9. The silver one is blocking the driveway, so tell him to park on the street.
Also, he gave me the quote for the tiles. It was 12,000 shekels, but he says he can drop it to 10 if we pay cash. Put it on Google Calendar so we don't forget.
Talk soon, Daniel.
S1-ASR-Cleanup:
Hey Yoni,
Quick update on the kitchen renovation.
The contractor wants to come Thursday morning at 9:00. The silver one is blocking the driveway, so tell him to park on the street.
Also, he gave me the quote for the tiles, 12,000 NIS, but he says he can drop it to 10,000 NIS if we pay cash. Put it on Google Calendar so we don't forget.
Talk soon,
Daniel
spelled
Raw transcript:
Put in my Google Calendar for next Tuesday at six in the morning tempo run with Brennan that's Brennan B R E double N A N we are meeting at the park near Hayarkon wait one sec Noa put that down sorry about that set a reminder thirty minutes before to do my dynamic stretches and update the interval split targets on TrainingPeaks the booking is under B R E double N A N if they ask at the track.
Stock S1-mini:
Put in my Google Calendar for next Tuesday at 6 in the morning. Tempo run with Brennan. That's Brennan, B-R-E, double N-A-N. We are meeting at the park near Hayarkon. Wait one sec.
Noa, put that down. Sorry about that. Set a reminder 30 minutes before to do my dynamic stretches, and update the interval split targets on TrainingPeaks. The booking is under B-R-E double N-A-N if they ask at the track.
S1-ASR-Cleanup:
Put in my Google Calendar for next Tuesday at 6:00 in the morning: tempo run with Brennan.
We are meeting at the park near Hayarkon. Set a reminder 30 minutes before to do my dynamic stretches.
Update the interval split targets on TrainingPeaks the booking is under Brennan if they ask at the track.
Status: v1.0, proof of concept. Trained on 405 before/after pairs: the user's own edits from
danielrosehill/ASR-Cleanup-Before-After
plus synthetic pairs targeting each behaviour above (LoRA rank 16, 4 epochs). Small data; expect rough edges.
Files
model.safetensorsand config: for transformers.s1-asr-cleanup-v1.0-Q8_0.gguf: for llama.cpp and anything built on it.s1-asr-cleanup-v1.0-Q4_K_M.gguf(added 2026-10-30): a smaller build for phones (397 MB), made from the same v1.0 weights with llama.cpp b11982. Not benchmarked against Q8_0.
Prompt
The same format as S1-mini, with the system prompt verbatim from its card, enable_thinking=False (the empty think
block below) and greedy decoding. The control line ends with [Layers: 1,2]:
<|im_start|>system
You are a text normalizer for speech-to-text transcripts. The input begins with a control line specifying the styling, structure, and context settings; clean the transcript to match those settings and output only the cleaned text.<|im_end|>
<|im_start|>user
[Styling: semi-formal] [Structure: prose] [Context: general] [Layers: 1,2]
<transcript><|im_end|>
<|im_start|>assistant
<think>
</think>
Keep inputs under about 1,000 tokens and set max_new_tokens to about 1.3 x input tokens + 32.
Licence
Apache 2.0 with the S1-mini naming term (see LICENSE and NOTICE): any use, distribution or integration must
identify the model as "S1-mini" by "Superwhisper". Training code:
danielrosehill/ASR-Post-Processor-Model.
- Downloads last month
- 112