Gahow Wang

gahow pushed to feat/two-stop at gahow/aituner

2026-06-15 08:03:18 +00:00

9f52812753 Document Stop-A validation: calibration + GPU fidelity check

gahow pushed to main at gahow/xtrain

2026-06-15 07:54:23 +00:00

777f3c7949 docs: Phase T4 — autograd engine

e7ce504b1f ops: differentiable autograd nodes + per-op grad-check tests

Compare 2 commits »

gahow pushed to main at gahow/xtrain

2026-06-15 07:52:36 +00:00

59e33f60e3 docs: Phase T4 — autograd engine

f3a1188f0e ops: differentiable autograd nodes + per-op grad-check tests

Compare 2 commits »

gahow pushed to main at gahow/xtrain

2026-06-15 07:46:52 +00:00

2e68942032 docs: Phase T4 — autograd engine

7de172a2cf ops: differentiable autograd nodes + per-op grad-check tests

224f750ee4 autograd: tape engine + grad accumulation

5aef3742d6 ops: transformer op fwd/bwd CUDA kernels + Tensor wrappers

Compare 4 commits »

gahow pushed to main at gahow/xtrain

2026-06-15 07:29:09 +00:00

88fbe0a85d gemm: realistic f32 tolerances in GEMM acceptance tests

gahow pushed to main at gahow/xtrain

2026-06-15 07:27:12 +00:00

dde2fde297 docs: Phase T3 — GEMM fwd/bwd + finite-diff

1384044f27 gemm: GPU acceptance tests vs cuBLAS + finite-diff

08c88bf360 gemm: tiled F32 forward + transpose + backward (dA/dB)

9ca98efd98 autodiff: finite-diff gradient-check harness

Compare 4 commits »

gahow pushed to feat/two-stop at gahow/aituner

2026-06-15 07:22:50 +00:00

958739027a Fix Stop-A validation config: system vllm, cap max-model-len

gahow pushed to feat/two-stop at gahow/aituner

2026-06-15 07:19:49 +00:00

0f57ee96a9 Drop LLM endpoint from Stop-A full-data config (baseline-only run)

gahow pushed to feat/two-stop at gahow/aituner

2026-06-15 07:19:10 +00:00

43125f48cf Address review of two-stop branch

3af1d84ac0 Add Stop-A full-data validation config (real-time replay, no cap)

Compare 2 commits »

gahow pushed to main at gahow/xtrain

2026-06-15 07:18:01 +00:00

fbd07a578c tensor: minimal Tensor crate over xtrain-cuda

63dc05fd10 tensor: add scale elementwise CUDA kernel + FFI

8557a289a2 docs: Phase T2 — tensor abstraction

c1b204296b docs: backfill T1 build-chain

Compare 4 commits »

gahow pushed to feat/two-stop at gahow/aituner

2026-06-15 07:10:21 +00:00

08e53fd897 Add Stop-A calibration script (CPU-only convergence curve)

gahow pushed to feat/two-stop at gahow/aituner

2026-06-15 06:45:34 +00:00

a8f903498d Add Stop-B authority: deterministic validator overrides LLM stop

gahow created branch main in gahow/xtrain

2026-06-15 06:42:48 +00:00

gahow pushed to main at gahow/xtrain

2026-06-15 06:42:48 +00:00

92acf9f413 T1: scaffold repo + Rust/CUDA build chain (vecadd smoke test)

gahow created repository gahow/xtrain

2026-06-15 06:28:09 +00:00

gahow created branch feat/two-stop in gahow/aituner

2026-06-15 06:24:11 +00:00

gahow pushed to feat/two-stop at gahow/aituner

2026-06-15 06:24:11 +00:00

51a9e4a007 Add Stop-A: offered-L-C-A convergence early-stop for replay

0f15bbc3f1 Make the offered-load axis session-coherent

6f8e3c95c1 Unify harness L-C-A on the canonical lca.WorkloadProfile

Compare 3 commits »

gahow pushed to main at gahow/aituner

2026-06-15 06:02:37 +00:00

8b4116fad0 Add reference paper and qwen27b tpot25 16-iter notes

27d1c8fa92 Add L-C-A workload profile metric and CLI profile commands

984eb1f325 Document 8-GPU harness ablation results for qwen27b and qwen235b prefill

Compare 3 commits »

gahow pushed to main at gahow/xserv

2026-06-12 07:01:43 +00:00

cf1e9e41db tools: single-stream decode benchmark vs llama.cpp

d33220498a quantization: MXFP4 W4A16 expert weights (memory-optimization foundation)

Compare 2 commits »

gahow pushed to main at gahow/xserv

2026-06-11 17:23:31 +00:00

e631a71b68 quantization: single strided-batched FP8 MoE GEMM — cut per-token launches ~768→48