Text Generation
MLX
Safetensors
English
lemonseed
mixture-of-experts
gated-deltanet
mixture-of-depths
rocm
base-model
Instructions to use lemonade-sdk/lemonseed-1.5b-base with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use lemonade-sdk/lemonseed-1.5b-base 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("lemonade-sdk/lemonseed-1.5b-base") 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 lemonade-sdk/lemonseed-1.5b-base with MLX LM:
Generate or start a chat session
# Install MLX LM uv tool install mlx-lm # Generate some text mlx_lm.generate --model "lemonade-sdk/lemonseed-1.5b-base" --prompt "Once upon a time"
- Atomic Chat
| """LemonSeed 1.5B — minimal generation template. | |
| LemonSeed is a custom hybrid (Gated-DeltaNet + attention + MoE + Mixture-of-Depths) | |
| architecture and is NOT loadable via `transformers`. | |
| Primary engine (C++/ROCm): https://github.com/Geramy/LSE | |
| Reference docs: https://github.com/Geramy/lemonseed-docs | |
| This file is the simple MLX reference path (needs the lemonseed Python model code). | |
| pip install mlx fastokens huggingface_hub | |
| git clone https://github.com/Geramy/lemonseed-docs && cd lemonseed | |
| Then download this repo's files (model.safetensors, config.json, tokenizer.json) | |
| next to each other and run: | |
| python example_generate.py | |
| """ | |
| import sys | |
| sys.path.insert(0, ".") # so `lemonseed` package is importable | |
| from lemonseed.model import load_model | |
| from lemonseed.tokenizer import load_tokenizer | |
| from lemonseed.generate import generate_text | |
| MODEL = "model.safetensors" # config.json must sit alongside it | |
| model = load_model(MODEL) | |
| tok = load_tokenizer("qwen3.6") # uses the Qwen3.6 tokenizer (bundled tokenizer.json) | |
| PROMPTS = [ | |
| "The capital of France is", | |
| "def fibonacci(n):", | |
| "Question: What is 8 × 6?\nLet's think step by step.\nAnswer:", | |
| "Once upon a time", | |
| ] | |
| for p in PROMPTS: | |
| out = generate_text(model, tok, p, max_new_tokens=64, greedy=True) | |
| print("=" * 60) | |
| print(out) | |