Software1 hOctober 21, 2026, 21:19

Ollama 0.34.1 drops LoRA adapter support

Since version 0.34.1 (September 2026), Ollama rejects any Modelfile with an ADAPTER line and no longer converts safetensors to GGUF or quantizes GGUF files. Users must now either merge the adapter into the base model and import a single GGUF, or serve it with llama.cpp's llama-server --lora.

DEV Community

Open source

Ollama 0.34.1 drops LoRA adapter support

Version 0.40.7 from October 18, 2026 still rejects ADAPTER
GGUF conversion and quantization moved to llama.cpp tooling
The merged GGUF generated about 13% faster on CPU
Older models with an adapter layer still run with a deprecation warning

#Ollama#Llama.cpp

llama-server prompt cache reuses KV computed under a different LoRA scale

Software

Ollama 0.34.1 drops LoRA adapter support

llama-server prompt cache reuses KV computed under a different LoRA scale

Superfluid: open-source local LLM server for multi-agent inference

DMAD: LoRA adapters cut MiniMax-H3 audio-video generation from 50 to 4 steps

vLLM ignores LoRA rank_pattern and alpha_pattern, mis-scaling adapters