Spaces:
Sleeping
Sleeping
Commit History
-
feat(gradio): timeout fallback to bundled mock_ui charts and summary
- FORGE_UI_RUN_TIMEOUT_SEC (default 120s, 0=off) wraps benchmark/compare in a thread pool. - On timeout or error: copy assets/mock_ui PNGs to outputs/ and show fixed summary.json metrics. - Compare uses base timeout × FORGE_UI_COMPARE_TIMEOUT_MULT (default 2). Made-with: Cursor5ca87df -
feat: custom_hf without CUDA falls back to offline baseline
- Router skips local HF load when no CUDA; auto chain skips HF the same way. - Gradio benchmark/compare shows Info when downgrading custom_hf -> offline on CPU. - Optional FORGE_ALLOW_CUSTOM_HF_CPU=1 for debug CPU loads; documented in README/.env.example. Made-with: Cursord1f6a06 -
fix(custom_hf): prefer dtype over deprecated torch_dtype in from_pretrained
Made-with: Cursor5a67c2d -
perf(ui): fewer Gradio steps + live STEPS_PER_EPISODE, tighter token cap
- env: read config.STEPS_PER_EPISODE at step boundary so cli --steps still works. - trainer: optional max_steps through train_defender/run_benchmark/compare; run_episode uses live config when None. - app: FORGE_UI_STEPS (default 2; full/0 = use global 3) for Gradio only. - FORGE_HF_MAX_NEW_TOKENS default 192; GenerationConfig use_cache=True. - Docs for new env vars. Made-with: Cursor26e10b8 -
perf(space-ui): faster Gradio custom_hf — UI candidates=1, cap tokens, GenerationConfig
- Gradio benchmark/compare pass FORGE_UI_CANDIDATES (default 1) vs trainer default 3. - FORGE_HF_MAX_NEW_TOKENS (default 256) wired into HFCustomProvider for shorter generations. - Use GenerationConfig in custom_hf generate to avoid invalid inherited sampling flags. - Document env vars in README and .env.example. Made-with: Cursorf37f4f8 -
fix(hf-space): flatten requirements.txt — Spaces build only mounts this file
HF pip step copies requirements.txt to /tmp alone; -r requirements-core.txt was missing from the image context and broke pip install (exit 1). Made-with: Cursor04c2414 -
Space T4 + real HF path: split requirements, GPU defaults, API reset behavior
- Add requirements-core.txt for slim Docker; root requirements.txt pulls core + PyTorch/TRL/PEFT for HF GPU Spaces. - README Space frontmatter suggested_hardware t4-small; document secrets FORGE_DETERMINISTIC_RESET and FORGE_DEFAULT_PROVIDER. - config: unset CODE_PROVIDER_MODE defaults to custom_hf when CUDA else offline. - app: GPU-first provider order, default custom_hf on CUDA, higher episode caps on GPU, clearer provider info, fix coach lesson line duplication. - trainer: benchmark respects requested episode count (no forced minimum of 20). - api_server: varied tasks per reset unless FORGE_DETERMINISTIC_RESET=1. - .env.example aligned with GPU Space defaults. Made-with: Cursor49a72d3 -
README/BLOG: hackathon theme fit line and HF Space Discussions mini-blog link
Made-with: Cursor6829d27 -
Finalize eval-friendly defaults: offline baseline, deterministic API reset, docs cleanup
- Rename user-facing provider mode from mock to offline across README, Docker, Compose, and .env.example; keep silent mock alias in code paths. - POST /reset reseeds RNG from GLOBAL_RANDOM_SEED for reproducible OpenEnv episodes. - MockModelPolicy and router use offline/baseline terminology in metadata; remove tracked coach_memory.json.tmp and ignore *.tmp. - Gradio defaults to offline provider; trainer and sample outputs use baseline:model:* labels. Made-with: Cursor0c741d9 -
docs+router: HF Space secret names; auto includes custom_hf only when HF_TOKEN set
Made-with: Cursora874a8f -
fix(router): exclude custom_hf from auto chain so Space auto stays fast
Made-with: Cursor9e4a6fa -
perf(ui): default mock provider; auto tries NIM/OpenRouter before HF; cap HF router timeout 90s
Made-with: Cursor665aab6 -
fix(space): ProxyHeadersMiddleware + uvicorn proxy flags so Gradio redirects use https behind HF
Made-with: Cursor19e3321 -
fix(space): mount Gradio at / for HF iframe; disable SSR; curl path + blank UI notes; example step JSON
Made-with: Cursor2ed8ca5 -
fix(api): Pydantic v2 step payload (exclude_none); env candidate_solutions; README Space API curl guide; Gradio API tab
Made-with: Cursor2b35bd5 -
Add story blog updates and submission-ready docs links
a2ac82d -
Finalize CPU-safe mock flow and verify live OpenEnv endpoints
310dc9f -
fix(hf-cpu): slim requirements.txt for Spaces; requirements-train.txt for PyTorch/TRL; suggested_hardware cpu-basic
Made-with: Cursor0c4d10b -
docs: Docker Hub image sanjay767676/forge
Made-with: Cursor57c225e -
docs: Docker Hub pull/push/login; remove ghcr instructions
Made-with: Cursor310fab1 -
Add Dockerfile, Docker Compose, and GitHub Action to publish Docker image to GHCR
55ef231 -
docs: hackathon badge links to judge guide; clarify video/slides vs mini-blog
Made-with: Cursor1506358 -
security: remove secrets from .env.example (use HF Space secrets only)
Made-with: Cursor2a885d9 -
fix(space): fastapi>=0.115.2 for Gradio 5.25; Pydantic v2 model_dump in API
Made-with: Cursor4b17fa3 -
fix(space): Gradio 5.25 + fastapi<0.113 for openenv-core websockets>=15
Made-with: Cursorca96787 -
docs: NOTE 1/2 checklist, judges rubric link; OpenEnv FORGEOpenEnvironment + openenv-core
Made-with: Cursordd06b37 -
feat: inference router (HF/NIM/OpenRouter/mock); README aligned with judge rubric; chart axis labels
Made-with: Cursor1546bc7 -
docs: judge-focused README; Colab notebook with repo clone step
Made-with: Cursore3161e4 -
Update .gitignore to broadly include outputs artifacts
74ca620 -
Allow tracking of models and adapter configurations
45a4e4c -
Add remaining training artifacts and reports to tracking
c99081a -
Final hardening for 100% submission readiness
c2975af -
Submission checklist fix: track judge plots and reduce README overclaims
44e7524 -
Submission audit fixes: loss curve export, README compliance links, robust training script fallback
b3be46a -
NVIDIA NIM Support: Added NvidiaNIMProvider as a first-class inference mode
d6eabf1
-
Fix Reporting Bug: Resolved NameError in train_unsloth.py report generation
3593fe0 -
Final Structural Fix: Preserved code in candidate rankings for authentic DPO export
3371219 -
Fix DPO Dataset Generation: Real preference pairs now exported from adversarial loop
96f35ba -
Colab Automation: Added setup and execution shell scripts for one-click training
253ba1d -
Real RL Training Layer: DPO Dataset Export + Unsloth Trainer + Training Report
7163e2f -
Final High-Impact Upgrade: Real LLM Inference, Candidate Ranking, and HF API Support
40a51ba -
Final Submission Upgrade: Advanced tier progression, professional README, and hackathon blog
bb6d47c -
Fix: Force Python 3.11 to resolve audioop ModuleNotFoundError
960509a -
Update deployment URL to sanjay7676/Team404_FORGE
bca42fe -
Fix: Add .git suffix to HF Space URL
10edca1 -
Trigger deployment: Configuration verified
b3576ca -
Improve deployment workflow robustness
218e9ff -
Final polish: Added Hugging Face YAML header and completed submission checklist
99286d3 -
Final submission assets: generated evidence, updated checklist, and judge narrative
69a701a -
Add Gradio app for Hugging Face Space and GitHub Actions auto-deploy workflow (sanitized)
581c33c