Supported Models

Choose a model to open its interactive import/export and training configurations. Every option comes directly from an authoritative YAML verification card; independent fields are never combined into synthetic commands.

Current inventory: 24 model cards and 345 concrete configurations.

Each model page lets you select import/export, pretraining, benchmarking, SFT, LoRA, or long-context SFT and immediately see the exact command and expected result for the recorded precision and GPU.

Models

Choose a model to inspect its recorded import/export, training, and precision combinations.

DeepSeek 2 models

DeepSeek-V3 ◐ Partial DeepSeek-V4-Flash ◐ Partial

Google 2 models

gemma-3-4b-it ◐ Partial gemma-4-26B-A4B-it ◐ Partial

LG AI EXAONE 1 model

K-EXAONE-2.0-750B-A37B ◐ Partial

MiniMax 1 model

MiniMax-M3 ◐ Partial

Moonshot AI 2 models

Kimi-K3 ◐ Partial Moonlight-16B-A3B ✓ Verified

NVIDIA 6 models

Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 ◐ Partial NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 ◐ Partial NVIDIA-Nemotron-3-Nano-4B-BF16 ✓ Verified NVIDIA-Nemotron-3-Super-120B-A12B-BF16 ◐ Partial NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 ◐ Partial NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 ✓ Verified

OpenAI 2 models

gpt-oss-120b ◐ Partial gpt-oss-20b ◐ Partial

Qwen 5 models

Qwen3-235B-A22B ◐ Partial Qwen3-30B-A3B ✓ Verified Qwen3-8B ✓ Verified Qwen3.6-35B-A3B ✓ Verified Qwen3.8-27B ◐ Partial

StepFun 1 model

Step-3.5-Flash ◐ Partial

Z.ai 2 models

GLM-5 ◐ Partial GLM-5.2 ◐ Partial

Verification status

Statuses are explicit: ✅ verified, ⚪ unverified, ❌ unsupported, and ➖ not applicable. A model page may summarize mixed concrete states as partially verified.

A partially verified model contains a mix of verified and pending or unsupported configurations. Selectors expose only exact card leaves and never fabricate combinations.

The versioned normalized data is available as catalog-v1.json.