Text Generation
Transformers
Safetensors
PyTorch
nemotron_h
nvidia
conversational
custom_code
Eval Results

How do i fix the Configuration Parsing issue with the Invalid JSON for config file config.json

#3
by mtcl - opened

As the title says: How do i fix the Configuration Parsing issue with the Invalid JSON for config file config.json

I am using vllm and had the same issue with the invalid JSON. I used 'hf download nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8' to save the model locally, then changed the config.json. You need to change line 58:
"time_step_limit": [
0.0,
Infinity
],

to:
"time_step_limit": [
0.0,
1e30
],

Infinity is not an allowed value in JSON. Then I can serve the model.

NVIDIA org

The config file has been updated, and the issue is resolved.

bkartal changed discussion status to closed

Sign up or log in to comment