Improve Evo2 generation execution and result validation - #41
Merged
Merged
Conversation
Signed-off-by: Ohad Mosafi <omosafi@nvidia.com>
Signed-off-by: Ohad Mosafi <omosafi@nvidia.com>
ohadmo
marked this pull request as ready for review
September 29, 2026 19:26
Collaborator
Author
|
/nvskills-ci |
|
Signed-off-by: nvskills-svc-account <svc-nvskills-signing@nvidia.com>
Signed-off-by: Ohad Mosafi <omosafi@nvidia.com>
Collaborator
Author
Collaborator
Author
|
/nvskills-ci |
Signed-off-by: Ohad Mosafi <omosafi@nvidia.com>
Collaborator
Author
|
/nvskills-ci |
Signed-off-by: nvskills-svc-account <svc-nvskills-signing@nvidia.com>
Signed-off-by: Ohad Mosafi <omosafi@nvidia.com>
Collaborator
Author
|
/nvskills-ci |
Signed-off-by: nvskills-svc-account <svc-nvskills-signing@nvidia.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The Evo2 generation example required agents to reconstruct an API client, leaving request execution, validation, and saved outputs inconsistent. This adds a reusable client for hosted and local generation that validates returned DNA, probabilities, and timing; saves request/response JSON, the raw response body, FASTA, and metrics; and reports values from the actual response. The skill instructions now direct agents to execute the client when generation is requested.
The evaluation dataset matches the September 29 neutral run, including the hosted task and assertions. The model configuration and grading thresholds are unchanged. Source and distributed skill copies are synchronized.
Client reliability
response.rawbefore checking HTTP status, parsing JSON, or serializing it. Malformed JSON and non-finite values retain diagnostics without producing success artifacts.requests>=2.28directly inpyproject.tomland regenerateuv.lock; resolved package versions remain unchanged.Validation
uv tree --no-dedupe --frozenpasses withrequestsdeclared as a direct dependency.ef0e5bb: Tier 1 and Tier 2 passed with observations; Tier 3 passed with Claude Code and Codex on the hosted generation task. The signing service attached the benchmark, skill card, and signature inb589a53. The subsequent client reliability fixes in93d4978pass local checks and require a fresh NV-CARPS evaluation and signature.Last completed Tier 3 results (
ef0e5bb)The Tier 3 job reports PASS, skill lift +0.24, for source commit
ef0e5bb.The benchmark reports the following effectiveness scores (50% goal accuracy + 50% behavior check):
Both conditions retain 100% security and correctness scores. This run covers one hosted-generation task, with one executed attempt per agent and condition; it does not establish coverage of the deferred local-Docker and forward-pass cases.
Evidence: completed evaluation for ef0e5bb, validation of the signed evaluation commit, and generated benchmark.