Files
zeekay 89da15d070 refactor: zen-coder branding - rename examples, add LLM.md and training configs
- 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
2026-02-27 19:06:12 -08:00

64 lines
1.8 KiB
Python

"""
Zen Coder Unified Training Framework
Supports fine-tuning 5 model architectures:
Tier 1: Qwen3 4B (lightweight, 8GB VRAM) $17-52
Tier 2: Devstral 24B (mid-range, 24GB VRAM) $70-140
Tier 3: Devstral 123B (large, 128GB VRAM) $140-280
Tier 4: GLM-4.5 358B (MoE, 128GB VRAM) $210-420
Tier 5: Kimi K2 1T (MoE, 400GB VRAM) $420-840
Costs based on 8xH200 @ $35/hr, 10K training samples.
Backends:
- MLX (Apple Silicon, local) - Tier 1-2
- Unsloth/PyTorch (NVIDIA GPUs, 2x faster) - Tier 1-4
- DeepSpeed ZeRO-3 (Multi-GPU clusters) - All tiers
- HuggingFace Autotrain (Cloud) - All tiers
Benchmarks (12 ARC from GLM-4.5):
Agentic: TAU-Bench, BFCL V3, BrowseComp
Reasoning: MMLU-Pro, AIME-24, MATH-500, SciCode, GPQA, HLE, LiveCodeBench
Coding: SWE-bench Verified, Terminal-Bench
Usage:
from zen_trainer import ZenTrainer, ZEN_MODELS, ZenBenchmark
# Train
trainer = ZenTrainer(
model_key="qwen3-4b",
dataset_path="hanzoai/zen-agentic-dataset",
output_dir="./output/zen-coder-4b",
)
trainer.train()
# Benchmark
bench = ZenBenchmark(
model_path="./output/zen-coder-4b",
model_key="qwen3-4b",
)
bench.run_all()
bench.compare_to_baseline()
"""
__version__ = "0.1.0"
from .models import ZEN_MODELS, get_model_config, estimate_training_cost, COST_SUMMARY
from .trainer import ZenTrainer, train_all_models
from .benchmark import ZenBenchmark, benchmark_all_models, generate_leaderboard
__all__ = [
# Models
"ZEN_MODELS",
"get_model_config",
"estimate_training_cost",
"COST_SUMMARY",
# Training
"ZenTrainer",
"train_all_models",
# Benchmarks
"ZenBenchmark",
"benchmark_all_models",
"generate_leaderboard",
]