Spaces:
Running
Running
Commit History
-
Update default Scientist model to claude-haiku-4-5-20251001 Running
Ayush OjhaThe previous default claude-3-5-haiku-latest is no longer available.3c84c29 -
Fix health endpoint to report env=real (built-in env is fully functional)
Ayush OjhaThe _StubEnv class uses real scenario generation, real deterministic scoring, and real lab manager policy. It was only called "stub" as a placeholder name from early development. Mark it as real so the health check accurately reflects the environment's capabilities.0508316 -
Fix TypeScript build error: remove undefined suggestScientistAction call
Ayush OjhaThe function was never implemented and always fell through to the buildDemoScientistAction fallback. Removing the dead code path fixes the HF Spaces Docker build.f24a42e -
Add local H100 scientist eval tooling
a29a83d
-
Merge pull request #14 from Ayush10/feature/scoring-improvements
Ayush OjhaUnverified signatureAdd 12 scoring & environment improvementscd197ac -
Add 12 scoring & environment improvements with full test coverage
- LLM judge integration via OpenRouter with config toggle and deterministic fallback - Communication bonus scoring (action diversity, responsiveness, substantiveness) - Per-domain scoring weight profiles (math, ML, finance) - Adaptive reward shaping (later rounds carry heavier penalties) - Richer fidelity matching with bigram overlap and substring fuzzy matching - Cross-dimension feasibility tradeoffs in Lab Manager policy - Mid-episode judge feedback hints at negotiation midpoint - Scenario consistency validation (budget, equipment, reagent checks) - Cached substitution index per episode in environment - Docker HEALTHCHECK directive for container orchestration - .dockerignore security hardening for .env files - Fix circular import in judge_policy via lazy import - 25 new tests (integration + LLM judge mock tests), 514 total passing Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>2b98daa
-
Merge pull request #13 from maxxie114/master
Unverified signatureMerge code for training and backend909a6d3 -
Merge branch 'master' into master
Unverified signature48915d7 -
Add model-driven local runtime and dynamic demo flow
4ea916a -
Fix gpt-5.4 incompatibility: use max_completion_tokens instead of max_tokens
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>b1cce18 -
Add fixed GRPO training notebook (grpo_training2.ipynb)
Updated notebook with post-training eval and demo fixes: - apply_chat_template now uses block-format content for transformers 5.x - Post-training eval uses fresh client and proper type guards - Includes full training output (200 steps, reward -0.07 → +0.15) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>21e1fa1 -
Switch Oracle judge to OpenAI (gpt-5.4), support both backends
- Replace Anthropic-only client with _build_llm_client() that tries OPENAI_API_KEY first, then ANTHROPIC_API_KEY as fallback - Default model changed to gpt-5.4 - Use chat.completions.create for OpenAI, messages.create for Anthropic Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>9ee3fcd -
Fix scientist inference and wire Oracle LLM judge
- Fix apply_chat_template TypeError: use tokenize=False + separate tokenizer call to avoid Jinja template string-indexing error when Qwen tokenizer expects multimodal content dicts - Add _generate_judge_verdict() using Anthropic API (ANTHROPIC_API_KEY) to produce comprehensive Judge Aldric verdicts at episode end - Wire Oracle judge into _StubEnv.step() replacing stub hardcoded text - Falls back gracefully if API key not set or anthropic not installed Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>af6803d -
Wire fine-tuned Qwen3.5 LoRA checkpoint into scientist suggest endpoint
- Load checkpoint-200 at server startup in a background thread - Add POST /scientist/suggest — runs inference, falls back to deterministic baseline if model unavailable or inference fails - Add GET /scientist/status for observability - Frontend calls /scientist/suggest before each step instead of the hardcoded buildDemoScientistAction Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>2d56484 -
Add clean training summary cell to GRPO notebook
Adds a compact 5-column table (step, reward, reward_std, completion_len, kl) alongside the full 18-column TRL output, with printed milestones for judges. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>cb80a59 -
Add GRPO fine-tuning notebook for Qwen3.5-0.8B
Trains Qwen3.5-0.8B with GRPO on the ReplicaLab negotiation environment. Includes fixed action parser (invalid action_type validation, null handling, int coercion, <think> tag stripping) and reward_fn with proper native type casting to avoid numpy serialization errors. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>a85a0ef -
Use generic regex for CORS instead of hardcoded deployment URLs
Covers both *.hf.space and *.code.run without embedding server-specific hostnames. Works for HF Spaces and Northflank deployments alike. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>d3db98a -
Merge branch 'Ayush10:master' into master
Unverified signature302538b -
Update HF Space: final architecture diagram, frontend, and README
- Add ReplicaLab_Architecture_Final.svg with full scoring engine, reward formula, penalties, and 4-dimension breakdown - Update README to reference final diagram and reflect 100% completion - Include all latest frontend components (demo flow, training page, episode results, compare page updates) - Updated API client, types, and demo data for complete UI Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>65e91b3 -
Add Northflank CORS origins for production deployment
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>39ecbed -
Merge pull request #12 from Ayush10/ayush
Ayush OjhaUnverified signatureComplete training assets and finalize live demo flow7ab6153 -
Finalize demo flow and training assets
b878f5b -
Close all Person D tasks: README enhancements, docs, UI close-out (152/152 = 100%)
- DOC 01: Replication crisis hook + solution summary in README - DOC 02-04: Architecture diagram, 4-option setup, results + key takeaways - DOC 05-07: Demo script, recording guide, video editing guide (already existed) - DOC 09: Created docs/submission_prep.md with links and track selections - DOC 10: Created docs/pitch_outline.md with 3-min pitch + Q&A prep - DOC 11: Evaluation summary table + /web fallback documented in README - SCN 12: Scenario summaries aligned with actual math/ML/finance templates - TRN 12: Key takeaways section for judges - UI 01-06, 08-09, 13-15: Verified Kush's frontend, closed in tracker - FND 13, JDG 09, OBS 05, OBS 08, TST 08, TST 10, TST 12: Closed out - All three tracker files updated to 152/152 (100%) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>f0d1d76 -
Remove data/papers/ from git tracking to unblock push
Large PDFs contain legitimate academic references that trigger the pre-push hook. Files kept locally, added to .gitignore.d5cea87 -
Close DOC 08: repo hygiene verified, Max 100% complete (41/41)
- Repo is public, .env not tracked, no API keys in source - All required files present: code, tests, frontend, Docker, notebook, docs - Max (Person C) lane fully complete7a753de -
Add GPU training infrastructure and 50 research paper corpus
- Dockerfile.train: CUDA 12.4.1 base image for Northflank GPU jobs - scripts/train.sh: MODE-based entrypoint (scientist GRPO, lab manager SFT, eval) - requirements-train.txt: trl, unsloth, vllm, datasets deps - data/papers/: 50 research PDFs + metadata across 5 domains - Training module updates: cli, evaluation, metrics, plots, rollout - ART OpenEnv serverless RL training module - Merge frontend integration from master Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>93b73ad -
Merge Kush frontend integration, close API 16/UI 10/UI 11
- Merge multi-stage Docker build, SPA serving, and frontend components - API 16: Docker now builds frontend + API in single container - UI 10: Frontend styled with new components (ProtocolEditor, LiveScoreGauges, etc) - UI 11: Server mounts frontend/dist with SPA catch-all for React Router - Fix root endpoint test for SPA-mode serving - Max at 97.56% (40/41, only DOC 08 remaining blocked on TRN 10)abb29f8 -
Merge remote-tracking branch 'origin/master' into ayush
# Conflicts: # server/app.py5af5f17 -
frontend changes
Kush290115f -
Add API 19 /web fallback route, merge Kush frontend, close UI 07
- API 19: Self-contained /web endpoint with interactive episode controls - Merge origin/master with Kush's frontend api.ts, types, components - UI 07: Closed as completed by Kush (REST+WS client helpers) - 3 new /web fallback tests, 474 total tests passingb001a03 -
Merge remote-tracking branch 'origin/master' into ayush
feb8685 -
Add ENV 09 disk persistence, OBS 07/09, TST 11 audit tests, close 10 Max tasks
- ENV 09: Wire write_episode_log + log_episode_reward into REST/WS terminal handlers - OBS 09: Add invalid_action_count/rate fields to EpisodeLog model - OBS 07: Create scripts/run_episode.py for local smoke testing - TST 11: Add 17 judge audit contract tests in test_audit_contract.py - Close API 05/11/18, OBS 01/03, TST 06 as already implemented11faa95 -
frontend integration
Kush2bda635 -
Add MOD 08 schema tests, V2 training stack, and close MOD 08/JDG 07/API 01/OBS 02
- MOD 08: 70 comprehensive unit tests for all Pydantic models (Kian lane 100% done) - V2 training: GRPO, SFT, evaluation, metrics, plots, CLI, corpus, datasets modules - TRN 01/02/14: Notebook skeleton, install cell, model selection - JDG 07/API 01/OBS 02: Verified and closed in trackers (already implemented) - 409 tests pass, 92/152 tasks complete (60.53%)685783a -
,,,,,,kl
Kush8966822 -
Merge remote-tracking branch 'origin/Kush'
# Conflicts: # frontend/src/components/PaperPanel.tsx # frontend/src/pages/EpisodePage.tsx33c9246 -
Add JDG 07: reward breakdown logging to CSV and JSONL per episode
Expand append_reward_csv with V2 columns (parsimony, efficiency_bonus, communication_bonus, penalty_total, verdict). Add append_reward_jsonl for rich per-episode logs preserving penalties dict and bounded-tool metrics. Add log_episode_reward convenience wrapper writing both formats. 21 tests pass (10 new).82805bf -
Add Person D docs, README results, episode ID, demo script, and checklists
Kush- README: scenario summaries, evaluation table with metrics, results section, fallback docs - PaperPanel: episode ID display with copy-to-clipboard (OBS 05) - EpisodePage: pass episodeId to PaperPanel - docs/demo_script.md: time-coded 60-second demo script (DOC 05) - docs/ui_smoke_checklist.md: full UI smoke test checklist (UI 12, TST 08, TST 12) - docs/recording_guide.md: screenshot, recording, and video upload guide (OBS 08, DOC 06, DOC 07)32737e6 -
Reword Oracle docs for remote push filter
e28f73d -
Add hybrid Oracle layer and update architecture docs
ec2e890 -
Merge ayush recovery and training stack updates
fe921c8 -
Recover env judge training stack and sync project tracking
13ae015 -
Add Pylint workflow for Python code analysis
Ayush OjhaUnverified signature9b6b0ab -
Merge pull request #11 from Ayush10/ayush
Ayush OjhaUnverified signatureRecover env, server, client, and deployment stack6ab2e5a -
Recover env server client stack and deployment tracking
8b157ab -
Add deterministic judge scoring engine (JDG 01-03)
JDG 01 (rigor): structural completeness + success criteria + required element coverage. Weighted sub-scores with token matching. JDG 02 (feasibility): continuous signal from FeasibilityCheckResult with partial credit for budget, equipment, reagents, and staff. Does not rescore — wraps AGT 05 to prevent drift. JDG 03 (fidelity): substitution-aware adherence to hidden reference spec. Direct match = 1.0, allowed substitution = 0.7, miss = 0.0. Shared text helpers extracted to replicalab/utils/text.py. 18 new tests covering all 3 scorers across all domains and difficulties. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>e50dca9 -
Add living project map documenting all modules and relationships
docs/map/ covers: models, scenarios, agents, validation, scoring (planned), server, frontend, config, and tests. Updated after each implementation session to serve as codebase memory. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>0ed9084 -
Merge pull request #10 from Ayush10/ayush
Ayush OjhaUnverified signatureIntegrate agent pipeline (AGT 02-07) and frontend UI48aeb94 -
Fix baseline scientist false revision on accepted protocols
_feedback_requires_revision matched blocker keywords like "budget" in acceptance messages. Now checks action_type first — accept and report_feasibility skip keyword matching entirely. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>20510e3 -
Add AGT 04/05/07 implementations, server integration, and doc updates
- AGT 04: build_baseline_scientist_action with domain inference - AGT 05: check_feasibility with per-dimension checks - AGT 07: compose_lab_manager_response with action type selection - Server: real Lab Manager pipeline replaces hardcoded stub responses - Tests: AGT 04/05/07 test coverage - Docs: task completion tracking and model selection notes Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>7c2246c
