Instructions to use ThomsenDrake/ish-est-0.8b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ThomsenDrake/ish-est-0.8b with MLX:

# Make sure mlx-vlm is installed
# pip install --upgrade mlx-vlm

from mlx_vlm import load, generate
from mlx_vlm.prompt_utils import apply_chat_template
from mlx_vlm.utils import load_config

# Load the model
model, processor = load("ThomsenDrake/ish-est-0.8b")
config = load_config("ThomsenDrake/ish-est-0.8b")

# Prepare input
image = ["http://images.cocodataset.org/val2017/000000039769.jpg"]
prompt = "Describe this image."

# Apply chat template
formatted_prompt = apply_chat_template(
    processor, config, prompt, num_images=1
)

# Generate output
output = generate(model, processor, formatted_prompt, image)
print(output)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use ThomsenDrake/ish-est-0.8b with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "ThomsenDrake/ish-est-0.8b"

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": "ThomsenDrake/ish-est-0.8b"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent

How to use ThomsenDrake/ish-est-0.8b 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 "ThomsenDrake/ish-est-0.8b"

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 ThomsenDrake/ish-est-0.8b

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use ThomsenDrake/ish-est-0.8b with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "ThomsenDrake/ish-est-0.8b"

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 "ThomsenDrake/ish-est-0.8b" \
  --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"

Ish Est (`ish-est-0.8b`)

On-device MLX food / nutrition VLM for the Ish iPhone app.

Local fine-tune of Qwen/Qwen3.5-0.8B (MLX bf16 base: mlx-community/Qwen3.5-0.8B-MLX-bf16) on Nutrition5k (overhead RGB / depth-split family), with LoRA merged into the backbone plus small O(1) nutrition heads (ish_est_heads.safetensors + ish_est_heads_config.json).

MLX weights are 4-bit quantized for on-device size.

License

Apache License 2.0. Keep this card, LICENSE, and NOTICE with the weights.

Credit:

Qwen / Alibaba — Qwen3.5-0.8B base
Nutrition5k — Thames et al., CVPR 2021 (CC BY 4.0)
Recipe inspiration: Doses-AI food VLM work (GGUF release); this repo is an independent MLX derivative, not a conversion of unpublished third-party safetensors.

Do not drop the upstream copyright notices.

Eval (do not treat point estimates as exact)

Split	Cal MAE	Protein MAE
Full held-out 507	81.0 kcal	5.7 g
Locked-10 smoke (honesty floor)	151.3 kcal	20.6 g

Ish spoken Result uses honesty bands (half-width ≥ ~151 kcal / ~21 g), not ±80 off the 507 MAE. Rejects gate: 3/3 non-food on Vision classify smoke.

Heads emit normalized O(1); denorm is pred * scale with cal=100, pro=10, mass=100 (see ish_est_heads_config.json).

Files the app needs

config.json + model*.safetensors (+ index if sharded)
tokenizer / processor sidecars
ish_est_heads.safetensors + ish_est_heads_config.json (device head load)

Hub slug for the app: ish-est-0.8b (org/user prefix as published).

Downloads last month: 11

Safetensors

Model size

0.9B params

Tensor type

U32

BF16

F32

MLX

Hardware compatibility

4-bit

Model tree for ThomsenDrake/ish-est-0.8b

Base model

Qwen/Qwen3.5-0.8B-Base

Finetuned

Qwen/Qwen3.5-0.8B

Quantized

(252)

this model

Ish Est (ish-est-0.8b)

License

Eval (do not treat point estimates as exact)

Files the app needs

Model tree for ThomsenDrake/ish-est-0.8b

Ish Est (`ish-est-0.8b`)