v9ai/outreach-email-qwen3-1.7b-4bit-lora

Outreach Email — MLX LoRA Adapter

Fine-tuned LoRA adapter for B2B outreach email generation, trained on Apple Silicon via MLX.

Usage

import mlx_lm
from huggingface_hub import snapshot_download

# Download adapter from HF Hub to local cache
adapter_path = snapshot_download("v9ai/outreach-email-qwen3-1.7b-4bit-lora")

# Load base model with adapter
model, tokenizer = mlx_lm.load("mlx-community/Qwen3-1.7B-4bit", adapter_path=adapter_path)

# Generate
messages = [
    {"role": "system", "content": "You write B2B outreach emails..."},
    {"role": "user", "content": "Write an initial outreach email..."},
]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
output = mlx_lm.generate(model, tokenizer, prompt=prompt, max_tokens=512)

Serving

# Download adapter first, then serve via mlx_lm.server (OpenAI-compatible API)
python3 -c "from huggingface_hub import snapshot_download; print(snapshot_download('v9ai/outreach-email-qwen3-1.7b-4bit-lora'))"
# Use the printed path as --adapter-path:
python3 -m mlx_lm.server \
    --model mlx-community/Qwen3-1.7B-4bit \
    --adapter-path /path/to/cached/adapter \
    --port 8080

Training Details

Parameter	Value
Base model	mlx-community/Qwen3-1.7B-4bit
Fine-tune type	LoRA
Rank	8
Scale	4.0
Dropout	0.1
Max sequence length	512
Training iterations	1170
Learning rate	1e-05
Hardware	Apple M1 MacBook Pro, 16GB

Task

Generates personalized B2B outreach emails (initial + 3 follow-up stages) as JSON:

{"subject": "...", "body": "..."}

Trained on synthetic examples generated via DeepSeek teacher model, validated against quality rubrics (JSON validity, word count compliance, personalization, CTA presence).

About

Part of the lead-gen B2B sales intelligence platform. See also the SalesCue ML modules for lead scoring, intent detection, sentiment analysis, and more.