filter --category=<reading|kernel|paper|infra|coding|writing|other>search <text>TAB to autocompleteESC to clear
activity --period=52w
streak=2dlongest=6d71 entries /
52 days
JulAugSepOctNovDecJanFebMarAprMayJunJul
lessmore
──wednesday, 8 july 2026────────────────────────────────────────────────────────────────────────────────2 entries
21:30[TRACE]Continuing work on it.
09:01[INFO]Reading about numerical quantized formats + the B300 datasheet (including Blackwell features like tcgen, TMA, CLC, 2CTA, etc.) to build an intuition.
──tuesday, 7 july 2026────────────────────────────────────────────────────────────────────────────────1 entry
21:30[TRACE]Starting v2 of the mega kernel implementation.
──friday, 3 july 2026────────────────────────────────────────────────────────────────────────────────1 entry
10:00[NOTE]Officially started :D lets see us in the final day of September with a blogpost :D
──thursday, 2 july 2026────────────────────────────────────────────────────────────────────────────────1 entry
20:00[NOTE]Preparing the summer work.
──monday, 22 june 2026────────────────────────────────────────────────────────────────────────────────2 entries
23:20[TRACE]More mega kernel work - iterating on the fused pipeline and closing the gap on the baseline.
17:53[POST]Finished blogpost on the mega kernel.
──sunday, 21 june 2026────────────────────────────────────────────────────────────────────────────────1 entry
23:33[TRACE]Mega kernel: scheduling the stages and balancing warp specialization for better overlap.
──saturday, 20 june 2026────────────────────────────────────────────────────────────────────────────────1 entry
23:48[TRACE]Weekend mega kernel grind: benchmarking the fused path vs the multi-launch baseline.
──friday, 19 june 2026────────────────────────────────────────────────────────────────────────────────1 entry
23:26[TRACE]Mega kernels - cutting HBM traffic by keeping intermediates on-chip across stages.
──thursday, 18 june 2026────────────────────────────────────────────────────────────────────────────────2 entries
23:55[TRACE]Mega kernel result: 1.23x vLLM on Llama 1B (B300).
23:38[TRACE]Tuning the mega kernel - register allocation and async copy pipelining for the persistent loop.
──wednesday, 17 june 2026────────────────────────────────────────────────────────────────────────────────1 entry
23:12[TRACE]Mega kernel: first end-to-end run fused, profiling occupancy and tail effects.
──monday, 15 june 2026────────────────────────────────────────────────────────────────────────────────1 entry
23:50[TRACE]Mega kernels - applying ideas from the Event Tensor paper to the fused dataflow.
──sunday, 14 june 2026────────────────────────────────────────────────────────────────────────────────1 entry
23:05[DEBUG]Reading the Event Tensor paper.
──saturday, 13 june 2026────────────────────────────────────────────────────────────────────────────────1 entry
23:30[TRACE]Debugging the mega kernel - SMEM pressure and barrier sync between the fused stages.
──friday, 12 june 2026────────────────────────────────────────────────────────────────────────────────1 entry
22:55[TRACE]Mega kernel work: producer/consumer warp specialization to overlap GEMM and the routing stage.
──thursday, 11 june 2026────────────────────────────────────────────────────────────────────────────────1 entry
23:42[TRACE]Mega kernels - sketching the persistent grid + on-chip pipeline between stages.
──wednesday, 10 june 2026────────────────────────────────────────────────────────────────────────────────1 entry
23:18[TRACE]Started working on mega kernels - fusing the whole forward pass into one persistent kernel.
──thursday, 4 june 2026────────────────────────────────────────────────────────────────────────────────1 entry
20:00[SUCCESS]Starting the LeetCode marathon - grinding problems.
──wednesday, 3 june 2026────────────────────────────────────────────────────────────────────────────────1 entry
23:50[SUCCESS]Working on the timeseries project (late night).
──tuesday, 2 june 2026────────────────────────────────────────────────────────────────────────────────2 entries
23:01[DEBUG]Reading the PoD Attention paper.
15:00[SUCCESS]Worked on adding a lot of features to markviz and the CC integration in the afternoon.
──friday, 29 may 2026────────────────────────────────────────────────────────────────────────────────1 entry
22:11[TRACE]Tried prefill MoE - reached c=64 at 0.77 of vLLM.
──thursday, 28 may 2026────────────────────────────────────────────────────────────────────────────────1 entry
11:23[TRACE]Pushed new flash-moe code to the main branch.
──wednesday, 27 may 2026────────────────────────────────────────────────────────────────────────────────1 entry
20:00[DEBUG]Re-reading Mirage.
──tuesday, 26 may 2026────────────────────────────────────────────────────────────────────────────────2 entries
11:10[DEBUG]Reading TVM paper.
09:45[DEBUG]Reading Ansor paper.
──monday, 25 may 2026────────────────────────────────────────────────────────────────────────────────1 entry
20:00[POST]Reworked the /log page: streak counter + expandable notes per entry.
──tuesday, 19 may 2026────────────────────────────────────────────────────────────────────────────────1 entry
13:30[NOTE]WON BEST PAPER at the faculty competition with flash-moe! Paper now moving on to compete at the university level :D
──sunday, 17 may 2026────────────────────────────────────────────────────────────────────────────────1 entry
16:00[TRACE]More flash-moe work : tests + improvements. Grinding hours on it before the competition.
──saturday, 16 may 2026────────────────────────────────────────────────────────────────────────────────1 entry
15:00[TRACE]Weekend work on flash-moe : adding more tests and iterating on it. Putting in hours to polish it up.
──tuesday, 12 may 2026────────────────────────────────────────────────────────────────────────────────1 entry
23:08[SUCCESS]Finished scheduled of modulo scheduling and printing of it. Working on next step because its a harder problem and i like it. I wanna build it and with llms i can do it correctly.
──monday, 11 may 2026────────────────────────────────────────────────────────────────────────────────2 entries
23:45[NOTE]A bit inactive but got burnt out by uni and lost my path :((
23:37[SUCCESS]Started working on modulo scheduling to understand it on toy problems.
──sunday, 10 may 2026────────────────────────────────────────────────────────────────────────────────1 entry
13:30[DEBUG]Re-reading Mirage superoptimizer paper and reading Twill compiler for scheduling + modulo scheduling.
──friday, 8 may 2026────────────────────────────────────────────────────────────────────────────────1 entry
23:00[TRACE]Refactored attention to use more different stuff like: scheduler policy and traits.
──saturday, 2 may 2026────────────────────────────────────────────────────────────────────────────────1 entry
23:11[NOTE]Enjoying the extended weekend :)
──thursday, 30 april 2026────────────────────────────────────────────────────────────────────────────────1 entry
23:12[TRACE]More attention library work . Refining the implementation.
──wednesday, 29 april 2026────────────────────────────────────────────────────────────────────────────────1 entry
23:14[TRACE]Continuing work on the attention library ,iterating on kernel internals primitives ( mma , async , warp p).
──tuesday, 28 april 2026────────────────────────────────────────────────────────────────────────────────1 entry
23:08[TRACE]Working on the attention library . FInished main kernel implementation.
──monday, 27 april 2026────────────────────────────────────────────────────────────────────────────────2 entries
23:55[TRACE]Category kernel - writing attention config structs and working on the main kernel implementation for it. Working on helpers methods
17:00[NOTE]Stuck with uni homeworks, wanna do more cooler stuff. Had idea to do some dlcompiler :)
──sunday, 26 april 2026────────────────────────────────────────────────────────────────────────────────1 entry