CondonFM Skills

[ohadmo]

This repository is read-only and does not accept Pull Requests. Please open an Issue for reproducible bugs.

[github-actions]

Thanks for your interest. This repository is read-only and does not accept Pull Requests. Please open an Issue for reproducible bugs.

[github-actions]

Thanks for your interest. This repository is read-only and does not accept Pull Requests. Please open an Issue for reproducible bugs.

[ohadmo]

/nvskills-ci

[ohadmo]

@greptileai

[greptile-apps]

Greptile Summary

The PR adds four public CodonFM skills covering setup, embedding extraction, variant scoring, and fine-tuning, together with static contract tests and NVSkills CI integration.

• Documents supported public Encodon workflows, required inputs, commands, outputs, and unsupported features.
• Updates evaluation checkpoint handling so safetensors files are not passed to torch.load.
• Adds direct-host dependencies and configures the CUDA 12.4 PyTorch package index.
• Adds tests validating skill metadata, runner arguments, references, and safetensors evaluation behavior.

Confidence Score: 5/5

The PR appears safe to merge, with no concrete blocking or independently actionable non-blocking defects established in the changed code.

The documented commands align with the checked-in runner, datasets, task wiring, and outputs, while the checkpoint change avoids deserializing safetensors through PyTorch without an established regression.

Important Files Changed

Filename	Overview
src/tasks.py	Narrows dataset-state loading to Lightning .ckpt files, uses CPU deserialization, and extracts an optional datamodule state before prediction.
requirements.txt	Adds dotenv and makes the CUDA 12.4 PyTorch repository an extra package index so the pinned xFormers requirement can resolve from it.
skills/codonfm-setup/SKILL.md	Documents container and direct-host setup, public checkpoint downloads, GPU verification, and supported public-v1 boundaries.
skills/codonfm-score/SKILL.md	Documents the reachable masked-codon mutation-prediction workflow and its input, output, and interpretation contracts.
skills/codonfm-embed/SKILL.md	Documents frozen CLS embedding extraction using the public Encodon evaluation path.
skills/codonfm-finetune/SKILL.md	Documents supported fine-tuning strategies, dataset requirements, checkpoint arguments, and runtime constraints.
tests/skills/test_public_skills.py	Adds standard-library contract tests for skill structure, parser compatibility, required flags, references, and safetensors handling.
.github/workflows/request-nvskills-ci.yml	Adds reusable NVSkills CI dispatch for pull requests, authorized issue comments, and signed automation pushes.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  U[User request] --> S{CodonFM skill}
  S -->|Setup| A[Environment and checkpoint setup]
  S -->|Embed| B[Encodon embedding prediction]
  S -->|Score| C[Masked-codon variant scoring]
  S -->|Fine-tune| D[Encodon training workflow]
  B --> V[Dry-run configuration check]
  C --> V
  D --> V
  V --> R[Runner execution]
  R --> O[Checkpoints, metrics, or NumPy outputs]

Loading

_{Reviews (1): Last reviewed commit: "nvskills workflow file" | Re-trigger Greptile}

[ohadmo]

/nvskills-ci

[ohadmo]

/nvskills-ci