Hugging Face
Search models, datasets, users...
Ctrl K
Models
Datasets
Spaces
Buckets
Docs
Enterprise
Pricing
Website
Tasks
HuggingChat
Collections
Languages
Organizations
Community
Blog
Posts
Daily Papers
Hardware
Learn
Discord
Forum
GitHub
Solutions
Team & Enterprise
Hugging Face PRO
Enterprise Support
Inference Providers
Inference Endpoints
Storage Buckets
Log In
Sign Up
Spaces:
cazyundee
/
training
Like
0
Running
on
Zero
App
Files
Files
Community
Fetching metadata from the HF Docker repository...
main
training
795 kB
Ctrl+K
Ctrl+K
2 contributors
History:
120 commits
cazyundee
RESEARCH_LOG: maxk2 final verdict + the three-arm 100K arc closes (mechanisms measured, escalation is the operator's; 1700-rung is the one cross-scale gain)
1a4dbbd
15 minutes ago
__pycache__
bench_learn: set logp_floor/kl_hard/min_updates attrs expected by train_step after logp_floor fix
3 days ago
docs
RESEARCH_LOG: maxk2 final verdict + the three-arm 100K arc closes (mechanisms measured, escalation is the operator's; 1700-rung is the one cross-scale gain)
15 minutes ago
experiments
SF ladder fix (was lost in the endgame manual sequence): ensure_stockfish() + hub-pinned SF binary download for Space in-run SF-Elo anchors; train_selfplay.py:565 imports it - container lacked it, so C1's in-run SF rows failed harmlessly (ImportError caught); PUSH DEFERRED until c1_attack completes (a push rebuilds the container and would kill the running job)
1 day ago
runs
fast_v1: speed-optimized training recipe on the Space
3 days ago
scripts
bench_learn: set logp_floor/kl_hard/min_updates attrs expected by train_step after logp_floor fix
3 days ago
tests
TD(lambda)/GAE return targets (--gae-lambda), per-outcome-class value diagnostics, SF-Elo absolute benchmark (experiments/exp_sf_elo.py, eval-only) + --sf-elo-every integration; tests/test_gae.py (7 tests)
2 days ago
tinychess
maxk config (NEW MANDATE: 100K budget): baseline arch at d72/dff160/d_attn48/d_move36 = 99,374 params - the 2000-Elo chassis absorbing all 18 proven feature channels + jitter; forward-verified
1 day ago
.gitattributes
1.52 kB
initial commit
2 months ago
.gitignore
20 Bytes
TD(lambda)/GAE return targets (--gae-lambda), per-outcome-class value diagnostics, SF-Elo absolute benchmark (experiments/exp_sf_elo.py, eval-only) + --sf-elo-every integration; tests/test_gae.py (7 tests)
2 days ago
README.md
2 kB
fix: gradio 5 for huggingface_hub>=1.0 compat
2 months ago
app.py
20.2 kB
exp_capture_probe: Space-side free-capture probe (eval-only), allowlisted; faithful to local probe (iter269 0.60 local vs 0.56-0.61 historical)
2 days ago
inspect_trace.py
6.24 kB
tinychess: self-play research substrate (phase 1-4)
2 months ago
requirements.txt
120 Bytes
gemma_decoder pilot: --checkpoint flag (trained-model source), textiser fixes (router-less guard, share clamp), requirements + transformers/huggingface_hub for Space
2 days ago
sf_curriculum.py
5.37 kB
Escalation option (i) implemented, DEFAULT OFF (--sf-opponent-frac 0.0): sf_curriculum module plays colour-balanced ours-vs-SF games with the league protocol (only OUR moves learned - logp/entropy/value on our plies, zeros on SF's); e2e smoke green (2 games vs SF@1320); the no-SF-in-training constraint is fully intact until the operator flips the flag
about 24 hours ago
train_selfplay.py
41.4 kB
terminal-anneal set-line (the restore existed but the SET never landed - cycle-64's patch printed instead of wrote; caught by maxk2's e2e launch FATAL at the restore line; now structurally verified set<loop<restore in main scope)
about 3 hours ago