Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.baseten.co/llms.txt

Use this file to discover all available pages before exploring further.

Loops supports a curated set of Hugging Face base models. Each entry below has a verified LoRA configuration on Baseten infrastructure. Coverage expands as new models are validated end to end.
ModelInference dtypeTrainer dtypeLoRA dtypeMax sequence length
Qwen/Qwen3.6-35B-A3Bfp8bf16bf16131,072
Qwen/Qwen3.6-27Bfp8bf16bf16131,072
Qwen/Qwen3.5-9Bbf16bf16bf16131,072
Qwen/Qwen3.5-4Bbf16bf16bf16131,072
Qwen/Qwen3.5-2Bbf16bf16bf16131,072
Qwen/Qwen3.5-0.8Bbf16bf16bf16131,072
Qwen/Qwen3.5-122B-A10Bfp8bf16bf16Contact support
Qwen/Qwen3.5-397B-A17Bfp8bf16bf16Contact support
moonshotai/Kimi-K2.6nvfp4bf16bf16Contact support
Qwen/Qwen3-30B-Instruct-2507bf16bf16bf16131,072
deepseek-ai/DeepSeek-V4-Pronvfp4bf16bf16Contact support
deepseek-ai/DeepSeek-V4-Flashfp8bf16bf16Contact support
zai-org/GLM-5.1nvfp4bf16bf16Contact support
MiniMaxAI/MiniMax-M2.7fp8bf16bf16Contact support
The model identifier is the Hugging Face path you pass as base_model when creating a trainer (POST /v1/trainer_sessions/{session_id}/trainers) or to truss loops push. The trainer dtype controls the precision used for forward, backward, and optimizer steps; the LoRA dtype controls the adapter weights; the inference dtype is what the paired sampling server runs the served checkpoint at. To request a model not listed here, contact support.