Instructions to use Dellboy/chem_sage_7b_v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Dellboy/chem_sage_7b_v1 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("Dellboy/chem_sage_7b_v1")

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 Dellboy/chem_sage_7b_v1 with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Dellboy/chem_sage_7b_v1"

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": "Dellboy/chem_sage_7b_v1"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use Dellboy/chem_sage_7b_v1 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 "Dellboy/chem_sage_7b_v1"

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 Dellboy/chem_sage_7b_v1

Run Hermes

hermes

OpenClaw new

How to use Dellboy/chem_sage_7b_v1 with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Dellboy/chem_sage_7b_v1"

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 "Dellboy/chem_sage_7b_v1" \
  --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 Dellboy/chem_sage_7b_v1 with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "Dellboy/chem_sage_7b_v1"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "Dellboy/chem_sage_7b_v1"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "Dellboy/chem_sage_7b_v1",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

ChemSage 7B — Round 1 fused model

ChemSage is a QLoRA-tuned chemistry assistant for drug discovery, fine-tuned on Apple Silicon with MLX-LM.

This is the Round 1 fused model — the initial proof-of-concept on a 7B base; superseded by the 32B rounds.

Model details


Base model	`mlx-community/Qwen2.5-7B-Instruct-4bit`
Fine-tune method	QLoRA (RSLoRA, rank 32, scale 64.0)
Adapted layers	16 of 28 transformer layers
Trainable parameters	~26M (0.34% of 7.6B)
Training examples	1,200 (from 1,500 total; 80/10/10 split)
Behaviour classes	8
Training iters	750
Best val loss	0.389 (at iter 675)
Peak memory	~12 GB
Fused model size	~4.2 GB

Evaluation (5-round comparative, 2026-06-29)

Evaluated on 100 shared R5 test examples (seed=42) via eval/compare/eval_compare.py. Scores = examples where all instances correct / 100 examples (per-example pass/fail). Full report: eval/compare/results/compare_20260629_1928.html.

Metric	R1 (this)	R2	R3	R4	R5
SMILES validity	100%	100%	100%	100%	100%
SMARTS validity	N/A	0%	100%	55%	100%
Tool executability	0%	28%	79%	71%	95%
Code attempted	14%	41%	100%	97%	100%
Python extended	36%	34%	78%	69%	99%
Code-then-quote	N/A	0%	47%	19%	61%
Numerical fidelity	N/A	18%	57%	47%	89%
Rounding precision	100%	100%	98%	98%	99%
Refusal accuracy	98%	98%	97%	98%	100%
QED range	100%	100%	100%	100%	100%
PDB ID validity	100%	100%	100%	100%	100%
PyMOL syntax	77%	97%	89%	89%	90%
Degeneration-free	93%	96%	91%	98%	100%
Overall	72%	63%	87%	80%	95%

Usage

# Download from HuggingFace
huggingface-cli download Dellboy/chem_sage_7b_v1 --local-dir models/chem_sage_7b_v1

mlx_lm.server --model models/chem_sage_7b_v1 --port 8081
.venv/bin/python scripts/chat.py --model models/chem_sage_7b_v1

Training history

Round	Val loss	Improvement
Round 1 (7B, this model)	0.389	—
Round 2 (32B)	0.347	10.8% over R1
Round 3 (32B)	0.054	6.4x over R2
Round 4 (32B)	0.041	24% over R3
Round 5 (32B)	0.055	early stop; rank=64, 20k examples

Built by

Marc C. Deller, D.Phil. · marcdeller.com · marc@marcdeller.com

Downloads last month: 26

Safetensors

Model size

1B params

Tensor type

F16

U32

MLX

Hardware compatibility

4-bit

Model tree for Dellboy/chem_sage_7b_v1

Base model

Qwen/Qwen2.5-7B

Finetuned

mlx-community/Qwen2.5-7B-Instruct-4bit

Quantized

(3)

this model