How nvidia-medtech/nv-reason-ct works
NV-Reason-CT provides example inference and fine-tuning workflows for a hosted generative 3D CT vision-language model; model weights, architecture, and processor implementation live in the external Hugging Face package. Inference accepts a NIfTI volume and prompt, then returns generated text. Training examples cover supervised fine-tuning and GRPO, including region-aware label and reward logic. The graph reflects the README and shown source excerpts; implementation details of the remote model and unsampled training code are not inferred.
Main entry points: vlm_grpo_train.py GRPO entrypoint · inference.py Inference CLI · vlm_sft_train.py SFT entrypoint · vlm_rewards.py Reward functions
Components 15
Inference
- Inference CLI
inference.py - NV-Reason-CT package
- Generated response
Supervised tuning
- SFT entrypoint
train/vlm_sft_train.py - SFT records
datalists - SFT configuration
configs - Processor save callback
train/vlm_callbacks.py
Policy optimization
- GRPO entrypoint
train/vlm_grpo_train.py - 3D GRPO trainer
train/vlm_grpo_trainer.py - GRPO records
datalists - GRPO configuration
configs
CT evaluation
- Reward functions
train/vlm_rewards.py - CT label schemas
train/vlm_labels.py
Other components
- Clinician or researcher
- NIfTI CT volume
Connections 16
- Clinician or researcher → Inference CLI: runs
README.md - NIfTI CT volume → Inference CLI: provides scan
inference.py - Inference CLI → NV-Reason-CT package: loads model and processor
inference.py - Inference CLI → NV-Reason-CT package: submits CT and prompt
inference.py - NV-Reason-CT package → Generated response: generates response
- SFT records → SFT entrypoint: loads records
train/vlm_sft_train.py - SFT configuration → SFT entrypoint: configures run
README.md - SFT entrypoint → NV-Reason-CT package: loads model and processor
train/vlm_sft_train.py - SFT entrypoint → Processor save callback: uses callback
train/vlm_sft_train.py - GRPO records → GRPO entrypoint: loads records
train/vlm_grpo_train.py - GRPO configuration → GRPO entrypoint: configures run
README.md - GRPO entrypoint → NV-Reason-CT package: loads model and processor
train/vlm_grpo_train.py - GRPO entrypoint → 3D GRPO trainer: constructs trainer
train/vlm_grpo_train.py - GRPO entrypoint → Reward functions: selects reward functions
train/vlm_grpo_train.py - GRPO entrypoint → Processor save callback: uses save helper
train/vlm_grpo_train.py - Reward functions → CT label schemas: uses label schemas
train/vlm_rewards.py