Merge(ia) : train model on medical dataset - #17
Open
estellebriollet13 wants to merge 55 commits into
Open
Conversation
Infra/ollama mvp
Stop tracking the committed __pycache__ bytecode and add the usual Python/HF-cache/ruff ignore patterns.
Sort imports and use `dict | None` over `Optional[dict]` so the merged infra script passes the lint gate.
ci: add CI/CD pipeline, linting and repo hygiene
The hardcoded file list omitted models/phi3_financial/special_tokens_map.json, which .gitattributes also tracks via LFS. With the other files materialized, the script exited 0 and reported success while that artifact could still be a pointer, falsely validating the inherited model. Discover the set from `git lfs ls-files`, falling back to the .gitattributes filter when git-lfs is absent, so the list can no longer drift.
fix: derive LFS check list dynamically (close Codex P2)
* ci: run gitleaks via CLI instead of licensed action The repository moved under an organization, and gitleaks-action requires a paid license for org-owned repos, breaking the secret-scan job. Run the open-source gitleaks CLI directly, which needs no license. * ci: allowlist intentional challenge artifacts in gitleaks Full-history scanning flags the planted 'compromised inheritance' indicators in logs/ (audited by the Cyber track). Add a documented .gitleaks.toml allowlist for logs/ so CI passes while still scanning every other path for real secrets.
* docs: add professional README and relocate brief to docs * chore: add MIT license * docs: add code of conduct and security policy * docs: add changelog * chore: add editorconfig * chore: add issue/PR templates and codeowners
* chore: remove Triton deployment path Team standardized on Ollama; drop the unused tritton_server/ Dockerfile and the model_repository/ Triton Python backend. * ci: drop Dockerfile lint job for removed Triton path * docs: drop Triton references after removal
* ci: exclude cyber audit evidence from gitleaks and LFS The cyber deliverables quote the planted challenge secrets as evidence; allowlist rendu/cyber/ in gitleaks and keep its small JSON proofs out of Git LFS. * feat: add cyber audit scripts and proof artifacts Dataset audit/cleaning tool, backdoor PoC, robustness test suite, the hardened test Modelfile, and the captured evidence (poisoning samples, live backdoor output, robustness results). * docs: add cyber security audit and robustness reports Security audit and robustness reports in Markdown with professional PDF exports, plus the deliverable index and reproduction guide. * style: satisfy ruff on cyber audit scripts
* chore: security hardening, Triton bonus dockerization and documentation updates * perf: share host HuggingFace cache with Triton container for instant loading and persistence * docs: fix shape dimension in Triton curl example * fix: import sys in train_finance_model.py to resolve Ruff F821 lint error * fix: resolve PR reviews: fix Dockerfile copy path context, use python healthcheck, add Mac unverified warning, and soften pilotage status * docs: clarify macOS hardware limitation for Triton bonus in pilotage status
* feat: débur ajout front interface - switch avec hugo (plus de token) * fix(devweb): stop page jumping to top on load Le scroll-vers-le-bas se déclenchait au montage à cause du double-rendu StrictMode (le garde didMountRef était contourné) et utilisait scrollIntoView, qui fait défiler la fenêtre entière. - garde robuste basée sur le nombre de messages (jamais au montage) - défilement limité au conteneur des messages, sans toucher la fenêtre - reset de scroll ciblant le bon conteneur scrollable Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * feat: branchement sur ia + améliroation front passage a hugo * refactor(devweb): split App into components, fix duplicate title - extrait Sidebar, ChatHeader, MessageList, MessageBubble, WelcomeScreen, Composer, SettingsPanel, BoxSelector, StatusBadge dans src/components/ - types partagés dans types.ts, données statiques dans data.ts - App.tsx réduit à l'orchestration de l'état - supprime le titre "Phi-3.5-Financial" en double sur l'écran d'accueil (on garde le nom du modèle actif) - supprime les fichiers morts ChatInput.tsx et hooks/useOllama.ts Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * feat: réglage front fini * fix: correction après rewiew * fix: build --------- Co-authored-by: atesta103 <alexandre.testa.at@gmail.com> Co-authored-by: BarretHugo <121606866+BarretHugo@users.noreply.github.com> Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
data: validation dataset finance (audit anomalies) + nettoyage dataset medical pour fine-tuning LoRA
Merge (ia-estelle) : test and validate phi3.5-financial
…ssier de rendu ia
# Conflicts: # rendu/ia/phi3_analysis.md # rendu/ia/run_phi3_api.py
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
4 participants
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.
No description provided.