Customization Target Values#
When creating a customization target via the API, you must provide the following fields:
name: A unique name for your target (see table below).namespace: The namespace in which to create the target (see table below).model_uri: The URI of the base model to customize (see table below).num_parameters: The number of parameters in the model (see table below).precision: The model’s precision (see table below).
Tip
For more information on configuring and managing GPU types for fine-tuning jobs, refer to the GPU Managment documentation.
Supported Model Values#
Use the following values to fill out your customization target details for a given model.
name |
namespace |
model_uri |
num_parameters |
precision |
|---|---|---|---|---|
llama-3.2-1b@2.0 |
meta |
ngc://nvidia/nemo/llama-3_2-1b:2.0 |
1000000000 |
bf16-mixed |
llama-3.2-1b-instruct@2.0 |
meta |
ngc://nvidia/nemo/llama-3_2-1b-instruct:2.0 |
1000000000 |
bf16-mixed |
llama-3.2-3b-instruct@2.0 |
meta |
ngc://nvidia/nemo/llama-3_2-3b-instruct:2.0 |
3000000000 |
bf16-mixed |
llama3-70b-instruct@2.0 |
meta |
ngc://nvidia/nemo/llama-3-70b-instruct-nemo:2.0 |
70000000000 |
bf16-mixed |
llama-3.1-8b-instruct@2.0 |
meta |
ngc://nvidia/nemo/llama-3_1-8b-instruct-nemo:2.0 |
8000000000 |
bf16-mixed |
llama-3.1-70b-instruct@2.0 |
meta |
ngc://nvidia/nemo/llama-3_1-70b-instruct-nemo:2.0 |
70000000000 |
bf16-mixed |
phi-4@1.0 |
microsoft |
ngc://nvidia/nemo/phi-4:1.0 |
14659507200 |
bf16 |
llama-3.3-70b-instruct@2.0 |
meta |
ngc://nvidia/nemo/llama-3_3-70b-instruct:2.0 |
70000000000 |
bf16-mixed |
nemotron-nano-llama-3.1-8b@1.0 |
nvidia |
ngc://nvidia/nemo/nemotron-nano-3_1-8b:0.0.1 |
8000000000 |
bf16-mixed |
nemotron-super-llama-3.3-49b@1.0 |
nvidia |
ngc://nvidia/nemo/nemotron-super-3_3-49b:v1 |
49000000000 |
bf16-mixed |
llama-3.2-1b-embedding@0.0.1 |
meta |
ngc://nvidia/nemo/llama-3_2-1b-embedding-base:0.0.1 |
1000000000 |
bf16-mixed |
Note
This table may not always reflect the latest supported models. For the canonical list, see the NeMo Customizer values.yaml.