Supported models

The packaged recipes currently support the following model, training method, and engine combinations.

A recipe is a packaged preset of training parameters for a specific model, training method, engine, and hardware configuration. Its compatibility rules determine when it can be selected, and its default parameters fill in values that are not provided in the training request.

Supervised fine-tuning (SFT)

aimfttk engine

The aimfttk engine provides the following SFT recipes:

Model	Model ID	Training method	engine	Training mode	Recipe ID
Gemma 3 27B IT	google/gemma-3-27b-it	SFT	aimfttk	LoRA, 1 GPU	aimfttk/sft/google-gemma-3-27b-it/lora/1gpu/default
Llama 3.2 1B Instruct	meta-llama/Llama-3.2-1B-Instruct	SFT	aimfttk	Full parameter, 1 GPU	aimfttk/sft/meta-llama-llama-3.2-1b-instruct/full/1gpu/default
Llama 3.2 3B Instruct	meta-llama/Llama-3.2-3B-Instruct	SFT	aimfttk	Full parameter, 1 GPU	aimfttk/sft/meta-llama-llama-3.2-3b-instruct/full/1gpu/default
Mistral Small 24B Instruct 2501	mistralai/Mistral-Small-24B-Instruct-2501	SFT	aimfttk	LoRA, 1 GPU	aimfttk/sft/mistralai-mistral-small-24b-instruct-2501/lora/1gpu/default
Qwen3 32B	Qwen/Qwen3-32B	SFT	aimfttk	LoRA, 1 GPU	aimfttk/sft/qwen-qwen3-32b/lora/1gpu/default

Select a supported model

Pass the model ID when running a job. The CLI uses it, together with the available accelerator and requested training mode, to select a compatible recipe:

aim_fine_tune train \
  --model-path /model \
  --train-data-path /data/train.jsonl \
  --output-path /output \
  --aim-model-id google/gemma-3-27b-it

To inspect the recipes included in the image, or to see only recipes compatible with the current accelerator setup, run:

aim_fine_tune list-recipes # List all recipes
aim_fine_tune list-recipes --compatible-only
aim_fine_tune list-recipes --aim-id google/gemma-3-27b-it

Recipe compatibility and defaults are defined in the packaged recipe files. See Recipes for selection behavior and the recipe schema reference for the metadata fields.