Instructions to use albertobarnabo/fiduciary-qwen3-4b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use albertobarnabo/fiduciary-qwen3-4b 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("albertobarnabo/fiduciary-qwen3-4b")

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 albertobarnabo/fiduciary-qwen3-4b with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "albertobarnabo/fiduciary-qwen3-4b"

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "albertobarnabo/fiduciary-qwen3-4b"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use albertobarnabo/fiduciary-qwen3-4b 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 "albertobarnabo/fiduciary-qwen3-4b"

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 albertobarnabo/fiduciary-qwen3-4b

Run Hermes

hermes

OpenClaw new

How to use albertobarnabo/fiduciary-qwen3-4b with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "albertobarnabo/fiduciary-qwen3-4b"

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 "albertobarnabo/fiduciary-qwen3-4b" \
  --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"

MLX LM

How to use albertobarnabo/fiduciary-qwen3-4b with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "albertobarnabo/fiduciary-qwen3-4b"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "albertobarnabo/fiduciary-qwen3-4b"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "albertobarnabo/fiduciary-qwen3-4b",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Fiduciary — Qwen3-4B (Senior Personal Financial Advisor)

Fiduciary is a fine-tuned Qwen3-4B-Instruct-2507 that acts as a senior personal financial advisor: it reviews your portfolio, checks the latest news, and reasons about how it affects your holdings — running entirely locally on Apple Silicon via MLX.

This is the fused, standalone 4-bit model (~~2.3GB) — runnable with one command, no adapter needed. The LoRA adapter alone (~~56MB) is at albertobarnabo/fiduciary-qwen3-4b-lora.

Quick start

pip install mlx-lm

python -m mlx_lm generate \
  --model albertobarnabo/fiduciary-qwen3-4b \
  --prompt "How's my portfolio doing?"

How it works

The split is deliberate: the model owns behavior, live tools own facts — because no fine-tune can know today's prices or this morning's headlines.

Senior-advisor voice — leads with the answer, exact per-position P&L, risk-first framing (sizing, concentration, time horizon), a "Bottom line" closer.
Disciplined tool use — emits <tool_call>{"name": ..., "arguments": ...}</tool_call> for get_portfolio, get_quotes, search_news, analyze_portfolio, reads the <tool_response>, and grounds every figure in it.
Knows when not to call tools — answers knowledge questions directly.
Guardrails — refuses "guaranteed winners", insider-trading bait, declines to execute trades (advice-only).

To use it as an agent, run a small loop: generate → parse the <tool_call> → execute the tool → append the <tool_response> → generate again. A reference runtime with live quote/news/portfolio tools (stdlib-only, no API keys) — plus the dataset generator and training config — is on GitHub: github.com/albertobarnabo/fiduciary.

Example (agent loop, live data)

you  Any news today that could affect my holdings?

Fiduciary
  → analyze_portfolio()  → search_news(query='market news', limit=6)

  Yes — there's news worth connecting to your holdings. ...
  **Impact on your portfolio:**
  - **Technology (51.6% of your equity — NVDA, AAPL, MSFT):** ...
  **Bottom line:** None of this justifies reactive trading. ...

Training

LoRA (rank 16) on ~1,200 fully synthetic advisor conversations, fused back into 4-bit weights (group size 64). Data is expanded one example per assistant turn so multi-turn tool calls are learned (mlx-lm's mask_prompt only trains the final assistant message). Every number in every answer is computed from the tool response in the same conversation. 1,200 iters, cosine LR 5e-5 → 5e-6, validation loss 0.046. Trained on an Apple M5 (16GB).

Limitations

Educational guidance, not individualized investment advice. Strong tool reflexes occasionally fire on a prompt that should be refused outright. News→impact mapping is sector-heuristic. Quote/news endpoints in the reference runtime are unofficial and degrade gracefully.

Downloads last month: 8,248

Safetensors

Model size

0.6B params

Tensor type

BF16

U32

MLX

Hardware compatibility

4-bit

Model tree for albertobarnabo/fiduciary-qwen3-4b

Base model

Qwen/Qwen3-4B-Instruct-2507

Quantized

mlx-community/Qwen3-4B-Instruct-2507-4bit

Finetuned

(2)

this model