Instructions to use v-vovk/medgemma-ios-q4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use v-vovk/medgemma-ios-q4 with MLX:
# Make sure mlx-lm is installed # pip install --upgrade mlx-lm # if on a CUDA device, also pip install mlx[cuda] # Generate text with mlx-lm from mlx_lm import load, generate model, tokenizer = load("v-vovk/medgemma-ios-q4") prompt = "Once upon a time in" text = generate(model, tokenizer, prompt=prompt, verbose=True) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- MLX LM
How to use v-vovk/medgemma-ios-q4 with MLX LM:
Generate or start a chat session
# Install MLX LM uv tool install mlx-lm # Generate some text mlx_lm.generate --model "v-vovk/medgemma-ios-q4" --prompt "Once upon a time"
- Atomic Chat
Update anonymous model card metadata
Browse files
README.md
ADDED
|
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Portable Nutrition Guidance SLM (Q4)
|
| 2 |
+
|
| 3 |
+
## Overview
|
| 4 |
+
|
| 5 |
+
This repository hosts a quantized on-device language model artifact for nutrition-guidance chat workflows.
|
| 6 |
+
The artifact is packaged for local iOS runtime loading and does not require backend inference.
|
| 7 |
+
|
| 8 |
+
## Intended use
|
| 9 |
+
|
| 10 |
+
- Personal nutrition guidance conversations
|
| 11 |
+
- Meal pattern coaching support
|
| 12 |
+
- Educational wellness tips for non-emergency scenarios
|
| 13 |
+
|
| 14 |
+
## Safety and medical disclaimer
|
| 15 |
+
|
| 16 |
+
This model is not a diagnostic or treatment system.
|
| 17 |
+
It should not be used for emergency decisions, medication dosing, or definitive medical advice.
|
| 18 |
+
Users should consult qualified clinicians for diagnosis and treatment decisions.
|
| 19 |
+
|
| 20 |
+
## Known limitations
|
| 21 |
+
|
| 22 |
+
- Outputs can still be incomplete, ambiguous, or incorrect.
|
| 23 |
+
- Responses may underperform on rare edge cases outside training distribution.
|
| 24 |
+
- Prompt phrasing can materially affect result quality and safety.
|
| 25 |
+
|
| 26 |
+
## Artifact details
|
| 27 |
+
|
| 28 |
+
- Version: `medgemma-ios-q4-v9-fixture-calib`
|
| 29 |
+
- Download URL: `https://huggingface.co/v-vovk/medgemma-ios-q4/resolve/main/medgemma-ios-q4-v9-fixture-calib.zip`
|
| 30 |
+
- SHA256: `dd794a634204557bec6639351d4e3fcfaa5717aca8599f5d7af7a3b29398684b`
|
| 31 |
+
- Size (bytes): `2594347078`
|
| 32 |
+
|
| 33 |
+
## Evaluation expectations
|
| 34 |
+
|
| 35 |
+
- Validate behavior against curated safety and hallucination checks before release.
|
| 36 |
+
- Track real-world mismatch reports and feed corrected examples into the next training cycle.
|