mirror of
https://github.com/zenlm/zen-coder.git
synced 2026-11-10 21:30:01 +00:00
- Rename Qwen2.5-Coder-* examples to zen-coder-* (consistent Zen branding) - Add LLM.md knowledge base (symlinked as CLAUDE.md/AGENTS.md/etc.) - Update .gitignore: exclude model weights and AI symlink files - Add training configs, inference scripts, test suite - Add MLX README, model card, Makefile for common operations
60 lines
1.7 KiB
Makefile
60 lines
1.7 KiB
Makefile
.PHONY: help train train-mlx train-full fuse test serve clean
|
|
|
|
MODEL_NAME = zen-coder
|
|
BASE_MODEL_HF = Qwen/Qwen3-Coder-Next
|
|
BASE_MODEL_MLX = lmstudio-community/Qwen3-Coder-Next-MLX-4bit
|
|
HF_REPO = zenlm/$(MODEL_NAME)
|
|
OUTPUT_DIR = ./training/output
|
|
ADAPTER_DIR = ./training/output/mlx-adapters
|
|
FUSED_DIR = ./training/output/zen-coder-mlx
|
|
|
|
help:
|
|
@echo "Zen Coder - Flagship Coding Model"
|
|
@echo "=================================="
|
|
@echo " make train LoRA fine-tune on MLX (Apple Silicon)"
|
|
@echo " make train-full Full training on agentic dataset"
|
|
@echo " make fuse Fuse LoRA adapters into model"
|
|
@echo " make test Test model identity + coding"
|
|
@echo " make serve Start local MLX inference server"
|
|
@echo " make clean Clean training artifacts"
|
|
|
|
train:
|
|
@echo "Training zen-coder on MLX..."
|
|
python training/train_zen_coder_4b_mlx.py --epochs 3 --batch-size 4
|
|
|
|
train-mlx:
|
|
@echo "MLX LoRA training on Qwen3-Coder-Next..."
|
|
python -m mlx_lm.lora \
|
|
--model $(BASE_MODEL_MLX) \
|
|
--train \
|
|
--data training/data \
|
|
--adapter-path $(ADAPTER_DIR) \
|
|
--iters 500 \
|
|
--batch-size 2 \
|
|
--num-layers 16 \
|
|
--learning-rate 1e-5
|
|
|
|
train-full:
|
|
python training/train_full.py
|
|
|
|
fuse:
|
|
python -m mlx_lm.fuse \
|
|
--model $(BASE_MODEL_MLX) \
|
|
--adapter-path $(ADAPTER_DIR) \
|
|
--save-path $(FUSED_DIR)
|
|
|
|
test:
|
|
@echo "Testing zen-coder..."
|
|
python -c "from mlx_lm import load, generate; \
|
|
model, tok = load('$(BASE_MODEL_MLX)'); \
|
|
print(generate(model, tok, prompt='Who are you?', max_tokens=100))"
|
|
|
|
serve:
|
|
@echo "Starting zen-coder server on port 3690..."
|
|
python -m mlx_lm.server --model $(BASE_MODEL_MLX) --port 3690
|
|
|
|
clean:
|
|
rm -rf $(OUTPUT_DIR)
|
|
|
|
.DEFAULT_GOAL := help
|