Text Generation
MLX
Safetensors
English
qwen3_5
dictation
speech-to-text
post-processing
text-cleanup
orbal
conversational
4-bit precision
Instructions to use Orbalapp/orb-1-mini with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use Orbalapp/orb-1-mini with MLX:
# Make sure mlx-lm is installed # pip install --upgrade mlx-lm # Generate text with mlx-lm from mlx_lm import load, generate model, tokenizer = load("Orbalapp/orb-1-mini") prompt = "Write a story about Einstein" messages = [{"role": "user", "content": prompt}] prompt = tokenizer.apply_chat_template( messages, add_generation_prompt=True ) text = generate(model, tokenizer, prompt=prompt, verbose=True) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Pi
How to use Orbalapp/orb-1-mini with Pi:
Start the MLX server
# Install MLX LM: uv tool install mlx-lm # Start a local OpenAI-compatible server: mlx_lm.server --model "Orbalapp/orb-1-mini"
Configure the model in Pi
# Install Pi: npm install -g @earendil-works/pi-coding-agent # Add to ~/.pi/agent/models.json: { "providers": { "mlx-lm": { "baseUrl": "http://localhost:8080/v1", "api": "openai-completions", "apiKey": "none", "models": [ { "id": "Orbalapp/orb-1-mini" } ] } } }Run Pi
# Start Pi in your project directory: pi
- MLX LM
How to use Orbalapp/orb-1-mini with MLX LM:
Generate or start a chat session
# Install MLX LM uv tool install mlx-lm # Interactive chat REPL mlx_lm.chat --model "Orbalapp/orb-1-mini"
Run an OpenAI-compatible server
# Install MLX LM uv tool install mlx-lm # Start the server mlx_lm.server --model "Orbalapp/orb-1-mini" # Calling the OpenAI-compatible server with curl curl -X POST "http://localhost:8080/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Orbalapp/orb-1-mini", "messages": [ {"role": "user", "content": "Hello"} ] }' - Hermes Agent
How to use Orbalapp/orb-1-mini with Hermes Agent:
Start the MLX server
# Install MLX LM: uv tool install mlx-lm # Start a local OpenAI-compatible server: mlx_lm.server --model "Orbalapp/orb-1-mini"
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 Orbalapp/orb-1-mini
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use Orbalapp/orb-1-mini with OpenClaw:
Start the MLX server
# Install MLX LM: uv tool install mlx-lm # Start a local OpenAI-compatible server: mlx_lm.server --model "Orbalapp/orb-1-mini"
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 "Orbalapp/orb-1-mini" \ --custom-provider-id mlx-lm \ --custom-compatibility openai \ --custom-text-input \ --accept-risk \ --skip-health
Run OpenClaw
openclaw agent --local --agent main --message "Hello from Hugging Face"
orb-1-mini v2, round 5: numbers as people type them
Browse files- README.md +31 -22
- model.safetensors +1 -1
README.md
CHANGED
|
@@ -24,6 +24,14 @@ silicon.
|
|
| 24 |
|---|---|
|
| 25 |
| So um the meeting is on Tuesday, no, Thursday at, uh, three thirty, and it's like twenty five percent over budget. | The meeting is on Thursday at 3:30, and it's 25% over budget. |
|
| 26 |
| So um can you, can you move the the logo to, sorry, to this side? | Can you move the logo to this side? |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 27 |
|
| 28 |
## Prompts
|
| 29 |
|
|
@@ -63,29 +71,29 @@ print(generate(model, tokenizer, prompt, max_tokens=cap, sampler=make_sampler(te
|
|
| 63 |
|
| 64 |
## Results
|
| 65 |
|
| 66 |
-
Tested on 178 English dictations
|
| 67 |
-
7 made up
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
| |
|
| 71 |
-
|---|---|---|---|
|
| 72 |
-
| Passes Orbal's check that nothing was added | 100% |
|
| 73 |
-
| Meaning kept, blind check by Qwen 3.8 27B |
|
| 74 |
-
| Content words kept |
|
| 75 |
-
| Answers with a word added (of 178) | 0 | 0 to
|
| 76 |
-
| Names made up | 0 | 0 | 0 |
|
| 77 |
-
| Questions answered | 0 | 0 | 0 |
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
|
|
|
|
| 81 |
time grows with the length: a 347-word dictation took 5.7 s.
|
| 82 |
|
| 83 |
## What it does badly
|
| 84 |
|
| 85 |
-
-
|
| 86 |
-
|
| 87 |
-
|
| 88 |
-
as said.
|
| 89 |
- The four style prompts change very little. Email sometimes adds
|
| 90 |
paragraph breaks; otherwise they stay close to the clean-up prompt.
|
| 91 |
- Long dictations are slow, see above.
|
|
@@ -93,12 +101,13 @@ time grows with the length: a 347-word dictation took 5.7 s.
|
|
| 93 |
|
| 94 |
## How it was trained
|
| 95 |
|
| 96 |
-
Distilled from Qwen 3.8 27B. The teacher wrote
|
| 97 |
-
a spoken version of each, with fillers, false starts
|
|
|
|
| 98 |
A third were read aloud with macOS `say` and transcribed by Apple's speech
|
| 99 |
engine, for real mishearings, and all went through Orbal's own text rules.
|
| 100 |
The teacher then cleaned each one up five times, once per prompt. The
|
| 101 |
-
|
| 102 |
negation and pointing word kept) trained a LoRA, rank 16, for 2 epochs. It
|
| 103 |
was merged and quantised to 4-bit MLX. No user dictations were used for
|
| 104 |
training.
|
|
|
|
| 24 |
|---|---|
|
| 25 |
| So um the meeting is on Tuesday, no, Thursday at, uh, three thirty, and it's like twenty five percent over budget. | The meeting is on Thursday at 3:30, and it's 25% over budget. |
|
| 26 |
| So um can you, can you move the the logo to, sorry, to this side? | Can you move the logo to this side? |
|
| 27 |
+
| It costs twelve ninety nine a month, so that's about fifty dollars for four months. | It costs $12.99 a month, so that's about $50 for 4 months. |
|
| 28 |
+
|
| 29 |
+
## Versions
|
| 30 |
+
|
| 31 |
+
- `v2` (round 5, current): writes numbers, times, dates, money and percentages as people type them.
|
| 32 |
+
- `v1` (round 4): the first release.
|
| 33 |
+
|
| 34 |
+
Pin a version with `revision="v2"`.
|
| 35 |
|
| 36 |
## Prompts
|
| 37 |
|
|
|
|
| 71 |
|
| 72 |
## Results
|
| 73 |
|
| 74 |
+
Tested on 178 English dictations, 171 real ones from Orbal's developer and
|
| 75 |
+
7 made up, and on 15 made-up dictations with numbers said as words. None
|
| 76 |
+
were used in training.
|
| 77 |
+
|
| 78 |
+
| | v2, clean up | v2, the four other prompts | v1, clean up |
|
| 79 |
+
|---|---|---|---|
|
| 80 |
+
| Passes Orbal's check that nothing was added | 100% | 99% to 100% | 100% |
|
| 81 |
+
| Meaning kept, blind check by Qwen 3.8 27B | 99% | 93% to 98% | 100% |
|
| 82 |
+
| Content words kept | 99% | 98% to 99% | 100% |
|
| 83 |
+
| Answers with a word added (of 178) | 0 | 0 to 4 | 0 |
|
| 84 |
+
| Names made up | 0 | 0 | 0 |
|
| 85 |
+
| Questions answered | 0 | 0 | 0 |
|
| 86 |
+
| Numbers typed as digits (22 in the numbers test) | 22 | | 19 |
|
| 87 |
+
|
| 88 |
+
Speed with mlx-lm on an M1 Pro: it loads in 1.2 s; a typical dictation
|
| 89 |
+
takes 0.13 s, and 9 in 10 take under 0.47 s. It keeps every word, so the
|
| 90 |
time grows with the length: a 347-word dictation took 5.7 s.
|
| 91 |
|
| 92 |
## What it does badly
|
| 93 |
|
| 94 |
+
- Numbers over 99 and forms like "half past two" were left out of
|
| 95 |
+
training, because Orbal's check cannot yet verify their digits. The
|
| 96 |
+
model may leave them as said.
|
|
|
|
| 97 |
- The four style prompts change very little. Email sometimes adds
|
| 98 |
paragraph breaks; otherwise they stay close to the clean-up prompt.
|
| 99 |
- Long dictations are slow, see above.
|
|
|
|
| 101 |
|
| 102 |
## How it was trained
|
| 103 |
|
| 104 |
+
Distilled from Qwen 3.8 27B. The teacher wrote 8,323 example messages and
|
| 105 |
+
a spoken version of each, with fillers, false starts, self-corrections and,
|
| 106 |
+
for v2, numbers said as words.
|
| 107 |
A third were read aloud with macOS `say` and transcribed by Apple's speech
|
| 108 |
engine, for real mishearings, and all went through Orbal's own text rules.
|
| 109 |
The teacher then cleaned each one up five times, once per prompt. The
|
| 110 |
+
24,920 answers that passed every check (nothing added, every name, number,
|
| 111 |
negation and pointing word kept) trained a LoRA, rank 16, for 2 epochs. It
|
| 112 |
was merged and quantised to 4-bit MLX. No user dictations were used for
|
| 113 |
training.
|
model.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 424026224
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a3f74f3b8a9f64886644f8c285cfbf9f1006c45528756ba259d4c008954d02f7
|
| 3 |
size 424026224
|