Spaces:
Running on CPU Upgrade
Running on CPU Upgrade
Add Nemotron 3.5 Transcribe evaluation scripts
#3
by nithinraok - opened
Add setup and custom evaluator scripts for nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B. Uses BF16 with NeMo Speech and vLLM 0.32.0. Requires an H100 80 GB. Syntax and simulated-response checks passed. GPU inference and the full installation remain untested.
nithinraok changed pull request status to open
Accepted after review. The setup pins match NeMo Speech 040574f (vLLM 0.32.0) and the model revision on the card. Python syntax, bash -n, the FFASR evaluate() contract, and a mock server check of the request body all passed. GPU inference was not run here.
whojavumusic changed pull request status to merged