Register and share your invite link to earn from video plays and referrals.

Search results for speeditup
speeditup community
One keyword maps to one global community path.
Create community
People
Not Found
Tweets including speeditup
Since when do you slow down 20 miles when it rains #speeditup#
Unlocking Lossless Speedups in LLMs via Discrete Diffusion paper:
Google used LLM agents to discover new quantum algorithms with 290x speedups over human baselines. 25 researchers from Google dropped an insane paper on discovering formally verified quantum programs with AI. Normally, if an LLM writes a complex algorithm, you have to worry about hallucinations. they can cheat, weaken theorems, or fake their performance numbers. so deepmind built a zero-trust environment. they unleashed 7 different autonomous agent teams to write quantum programs AND simultaneously write the mathematical proofs in the Lean theorem prover to prove they actually work. every submission was checked in a clean sandbox against a frozen specification. the ai couldn't lie.. it either worked and mathematically proved its efficiency, or it failed. to make it even crazier, they gave the agents a shared version-controlled wiki. if one agent team hit a dead end, they posted their notes to the wiki so the other ai teams wouldn't repeat the mistake. the results of this multi-agent collaboration are actually mindblowing: - they cut the computational cost of the arcsin quantum subroutine by over 290x - they reduced the cost of simulating nuclear fusion stopping power, photochemistry, and drugs by 2.49x to 10.9x - on one photochemistry problem, the ai teams iterated and beat each other's records 7 different times, eventually combining techniques from three separate teams to get the final verified circuit fault-tolerant quantum computers are still years away because the algorithms require way too much compute. but these ai agents are drastically shrinking the size of the software needed to run them. we are literally watching autonomous ai agents mathematically optimize the software for quantum computers that don't even fully exist yet.. the future is moving so fast.
Show more
Catching up with the chain after being offline will see massive speedups with the upcoming Glamsterdam hard fork. Here's how it works: Today, nodes that come back online enter what's called "full-sync" mode. This means fetching blocks from peers and executing them. This guarantees that nodes verify each state transition and produce transaction receipts (=outputs that tell you "how" your tx executed, the amount of gas it used, logs, etc). With EIP-7928: Block-Level Access Lists (BALs), each block comes with a BAL (=state diff) attached, allowing nodes to apply the changes listed in each BAL to their local state and arrive at the correct post-state, validated against the state root in the block header. This has already been useful for snap sync, and Glamsterdam ships snap/2, improving upon classical snap sync by increasing robustness and speed. It turns out, it's also useful for catching up with the chain after being offline for a while. Instead of entering full-sync mode, nodes can download blocks and BALs from peers and then use the BALs to derive the latest finalized state, and the state diffs between blocks needed for reorgs. Instead of re-executing all transactions, they simply apply the BALs. This works fine for most nodes but some might also want to have the receipts for finalized transactions. In such cases, nodes can either re-execute or fetch receipts from peers. This is safe because execution is only skipped for finalized blocks, as reported by the attached CL client, and it's fast. Nethermind has been experimenting with this, and the first results are pretty wild: 2,000 blocks synced in 40 seconds instead of 76 minutes. That's 115x faster. These are early devnet benchmarks, so take them with a grain of salt, but this is very much the kind of improvement that could make life easier for RPCs, full-node/desktop wallets. Check out @Nethermind 's PR:
Show more
Finally, a speedup method for Minimax H3 reference to video! ComfyUI-Ref2VA-VSA adds video sparse attention to Minimax, generating 5-second clips in ~72s on an RTX 4090. It keeps reference tokens dense while pruning video attention—2.24× faster than VDN-H3 at ~13.5GB VRAM.
Show more
⚡ New RTX local-agent optimizations include a vLLM speedup on Blackwell. @NVIDIARTXSpark reports: 🛠️ 1.2x vLLM performance on RTX PRO 6000 Blackwell ⚡ Up to 1.4x on a two-system DGX Spark cluster Great to see the local serving path getting faster across RTX and DGX Spark.
Show more
Lossless quantization has usually meant giving up inference speedup. This paper changes that. SLQ (Statistically-Lossless Quantization) reaches task-lossless compression at 3.3 bits per parameter, and distribution-lossless at 5-6 bpp where the output distribution is practically indistinguishable from the original. 1.7 to 3.6x throughput over BF16 in @vllm_project. Beats FP8 while staying lossless. From Michael Helcig, @_EldarKurtic, and @DAlistarh.
Show more
MOSS Image 8 step finetune is here. ~5x speedup compared with the base model (50 steps).
🚀 Introducing FlashQLA: high-performance linear attention kernels built on TileLang. ⚡ 2–3× forward speedup. 2× backward speedup. 💻 Purpose-built for agentic AI on your personal devices. 💡Key insights: 1. Gate-driven automatic intra-card CP. 2. Hardware-friendly algebraic reformulation. 3. TileLang fused warp-specialized kernels. FlashQLA boosts SM utilization via automatic intra-device CP. The gains are especially pronounced for TP setups, small models, and long-context workloads. Instead of fusing the entire GDN flow into a single kernel, we split it into two kernels optimized for CP and backward efficiency. At large batch sizes this incurs extra memory I/O overhead vs. a fully fused approach, but it delivers better real-world performance on edge devices and long-context workloads. The backward pass was the hardest part: we built a 16-stage warp-specialized pipeline under extremely tight on-chip memory constraints, ultimately achieving 2×+ kernel-level speedups. We hope this is useful to the community!🫶🫶 Learn more: 📖 Blog: 💻 Code:
Show more
0
33
1.3K
149
Forward to community
at 4096 lanes per call, across the set of proposed operations we see a wallclock speedup of 11.7x-47.6x for 64-bit integers and 10.3x-45.1x for 128-bit integers. the proposed cost model results in a 6.1x-20.0x reduction in cu consumption for 64-bit operations and an 11.1x-60.5x reduction for 128-bit operations
Show more