Instructions to use evraldtah/Synapse-Alpha-1.0-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use evraldtah/Synapse-Alpha-1.0-4bit 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("evraldtah/Synapse-Alpha-1.0-4bit")

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

How to use evraldtah/Synapse-Alpha-1.0-4bit with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "evraldtah/Synapse-Alpha-1.0-4bit"

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": "evraldtah/Synapse-Alpha-1.0-4bit"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

MLX LM

How to use evraldtah/Synapse-Alpha-1.0-4bit with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "evraldtah/Synapse-Alpha-1.0-4bit"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "evraldtah/Synapse-Alpha-1.0-4bit"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "evraldtah/Synapse-Alpha-1.0-4bit",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Hermes Agent

How to use evraldtah/Synapse-Alpha-1.0-4bit 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 "evraldtah/Synapse-Alpha-1.0-4bit"

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 evraldtah/Synapse-Alpha-1.0-4bit

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use evraldtah/Synapse-Alpha-1.0-4bit with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "evraldtah/Synapse-Alpha-1.0-4bit"

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 "evraldtah/Synapse-Alpha-1.0-4bit" \
  --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"

Synapse Alpha 1.0 (4-bit MLX)

The on-device model of Synapse, a notes app with an Obsidian-style graph: Qwen3.5-2B fully fine-tuned on Synapse's own prompts, then quantized to 4 bits for MLX on the iPhone.

Trained for

turning speech-recognition transcripts, documents and photo text into meeting, lecture and summary notes, including Synapse's split-and-merge path for long recordings
cleaning up raw transcripts: mis-heard names and words fixed from context, punctuation, paragraphs, nothing summarised
the editor's AI actions, Ask with [[Page]] citations, and [[link]] suggestions for the graph
Synapse's text tool protocol (<tool name="…">…</tool>): web search and pages, a Linux shell on the phone or the computer, and the Agents tab's tools for pages, links, tags, the graph and files

Prompt format. One system message and one user message (earlier turns inlined in <conversation_so_far>), with the chat template's enable_thinking=false. Replies end with <|im_end|>.

Training. Full fine-tune of Qwen3.5-2B on one H100 (bf16, AdamW, lr 1e-5, 2 epochs).

Limits. An alpha trained on generated data from Synapse's own pipelines: check important facts, and use a larger model for hard reasoning. Speech recognition is done by Apple's SpeechTranscriber or Whisper; this model works on text.

Licence: Apache-2.0, as the base model. Training data included Wikipedia text (CC BY-SA).

Downloads last month: -

Safetensors

Model size

2B params

Tensor type

U32

BF16

F32

MLX

Hardware compatibility

4-bit

Model tree for evraldtah/Synapse-Alpha-1.0-4bit

Base model

Qwen/Qwen3.5-2B-Base

Finetuned

Qwen/Qwen3.5-2B

Quantized

(279)

this model