ceselder
/

maxact-fast

interpretability

activation-steering

max-activating-examples

Model card Files Files and versions

116 kB

Ctrl+K

Ctrl+K

2 contributors

History: 18 commits

ceselder's picture

SPEED.md: MFU state + open speed levers for hill-climbing; make full README canonical

8e0dddc 1 day ago

scripts
out-of-core rework: embed_cluster_acts streams resids to disk memmap + streaming multi-GPU k-means (--kmeans-sample); build_data reads unified emb.f32 + GPU-batched Newton/IRLS probes 1 day ago
src
pretrain: sequence packing (block-diag attn, per-marker injection) — no-leak + inject proofs pass; MFU bench pending 2 days ago
.gitignore

64 Bytes
maxact-fast: clean repo — DESIGN + pretraining pipeline (embed/cluster/probe-cache/build-data/pretrain), Dr. GRPO RL to follow 2 days ago
DESIGN.md

3.89 kB
maxact-fast: clean repo — DESIGN + pretraining pipeline (embed/cluster/probe-cache/build-data/pretrain), Dr. GRPO RL to follow 2 days ago
README.md

2.74 kB
SPEED.md: MFU state + open speed levers for hill-climbing; make full README canonical 1 day ago
README_HF.md

2.74 kB
rl: pin TRITON_ATTN backend (FLASHINFER silently breaks vllm-lens injection on Blackwell) 2 days ago
SPEED.md

4.59 kB
SPEED.md: MFU state + open speed levers for hill-climbing; make full README canonical 1 day ago
pyproject.toml

763 Bytes
maxact-fast: hardening fixes (6 bugs) + Dr.GRPO vllm-lens rl.py + sbatch; Probe = universal direction 2 days ago