Gahow Wang

gahow deleted branch t14-flash-attention from gahow/xtrain

2026-06-18 11:51:55 +00:00

gahow deleted branch t15-gqa from gahow/xtrain

2026-06-18 11:51:55 +00:00

gahow deleted branch t16-grad-accum from gahow/xtrain

2026-06-18 11:51:55 +00:00

gahow deleted branch t17-process-per-gpu from gahow/xtrain

2026-06-18 11:51:55 +00:00

gahow pushed to main at gahow/xtrain

2026-06-18 11:51:44 +00:00

db70abe450 docs: T20 — Phase-2 systems-depth capstone (reframe README to two phases)

gahow pushed to main at gahow/xserv

2026-06-18 10:12:07 +00:00

531cd3fe08 style: format Rust workspace

013465fc06 docs: Phase 21 — decode CUDA graph + GPU argmax results

8414f8d1e6 sampling: GPU argmax fast path for greedy decode

34224c7c93 gpt-oss: replay the whole batch=1 decode step as one CUDA graph

4088f49b7d cuda: infrastructure for whole-step CUDA graph capture

Compare 10 commits »

gahow created branch t20-capstone in gahow/xtrain

2026-06-18 10:11:54 +00:00

gahow pushed to t20-capstone at gahow/xtrain

2026-06-18 10:11:54 +00:00

db70abe450 docs: T20 — Phase-2 systems-depth capstone (reframe README to two phases)

gahow pushed to main at gahow/xtrain

2026-06-18 10:05:08 +00:00

71b0a1621f docs: T17 process-per-GPU results — measured throughput-neutral

4abb17383a test: process-per-GPU DDP correctness (ddp_proc.rs)

a188c8a277 distributed: train_ddp_mp bin (process-per-GPU launcher/worker)

ffd548b80b distributed: process-per-GPU launcher + worker (proc.rs)

c470c627a7 docs: Phase T17 — process-per-GPU DDP design

Compare 5 commits »

gahow pushed to t17-process-per-gpu at gahow/xtrain

2026-06-18 10:03:21 +00:00

71b0a1621f docs: T17 process-per-GPU results — measured throughput-neutral

gahow created branch t17-process-per-gpu in gahow/xtrain

2026-06-18 09:48:58 +00:00

gahow pushed to t17-process-per-gpu at gahow/xtrain

2026-06-18 09:48:58 +00:00

4abb17383a test: process-per-GPU DDP correctness (ddp_proc.rs)

a188c8a277 distributed: train_ddp_mp bin (process-per-GPU launcher/worker)

ffd548b80b distributed: process-per-GPU launcher + worker (proc.rs)

c470c627a7 docs: Phase T17 — process-per-GPU DDP design

Compare 4 commits »

gahow pushed to main at gahow/xtrain

2026-06-18 09:39:26 +00:00

2ff4573a31 docs: T15 GQA results + evolution row (模型架构) + README build-journey row

39df0b40c1 gqa: fix kv-proj shape test param indices (embed,attn_norm precede wq)

830d06ad01 gqa: real grouped-query attention (repeat_kv op + both SDPA paths + wiring + tests)

62b1cb5dc7 docs: Phase T15 — GQA design (repeat_kv broadcast op + backward grad-sum)

Compare 4 commits »

gahow pushed to feat/fig18-real-output-lca-substrate at gahow/aituner

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

95c02d7dd9 Fig-18: chained driver for 2 extra naive runs (n=3 nondeterminism)

gahow pushed to t15-gqa at gahow/xtrain

2026-06-17 17:45:00 +00:00

2ff4573a31 docs: T15 GQA results + evolution row (模型架构) + README build-journey row

gahow pushed to t15-gqa at gahow/xtrain

2026-06-17 17:38:43 +00:00

39df0b40c1 gqa: fix kv-proj shape test param indices (embed,attn_norm precede wq)

gahow created branch t15-gqa in gahow/xtrain

2026-06-17 17:37:44 +00:00

gahow pushed to t15-gqa at gahow/xtrain

2026-06-17 17:37:44 +00:00

830d06ad01 gqa: real grouped-query attention (repeat_kv op + both SDPA paths + wiring + tests)

62b1cb5dc7 docs: Phase T15 — GQA design (repeat_kv broadcast op + backward grad-sum)

Compare 2 commits »

gahow deleted branch integration-t14-t16-t18 from gahow/xtrain

2026-06-17 17:24:04 +00:00

gahow pushed to main at gahow/xtrain

2026-06-17 17:23:57 +00:00

4b6d3e0a79 test: flash+dropout cross-feature grad-check (Phase-2 integration)

c36cdf74d1 Merge t18-dropout into main

f26db882e5 Merge t16-grad-accum into main

9e958cb0f9 Merge t14-flash-attention into main

80fafa1914 docs: T18 evolution row + README build-journey row (dropout)

Compare 26 commits »