building world's fastest retrieval & inference algorithms
you can visit www dot srswti dot com to know more.
SRSWTI Research Labs
moonshot to machine god.
- 20 followers
- United States of America
- @srswti_ai
- https://www.srswti.com
- team@srswti.com
Pinned Loading
Repositories
Showing 10 of 19 repositories
- knivesysl-xe Public
Native Intel Xe2 LLM inference: SYCL/DPAS kernels, paged KV, prefix caching, continuous batching, and XMX packed prefill
- knivesysl Public
bare-cuda fp6 inference engine for qwen3.8-27b on one rtx 5090 — sm120 tensor-core kernels, 256k context, near-lossless perf too!
- cutlass-ysl Public
nvfp4 gemm specialized for llm decode on consumer blackwell (sm120). 1.42x over cutlass 4.8 at batch 1, bit-exact.
- shooting-brake Public
run a 118b moe on one 5090 and two intel b70s. 88 tok/s. yes cuda and sycl both runtimes talking to each other. beating rtx pro 6000 on 120k ctx plus.
- vllm Public Forked from vllm-project/vllm
A high-throughput and memory-efficient inference and serving engine for LLMs
- cradle-codec Public
gpu-native kv cache compression and streaming infrastructure for long-context inference. ummm basically, nccl but through ethernet
- matryoshka Public
- axe Public
axe - a precision agentic coder. large codebases. zero bloat. terminal-native. precise retrieval. powerful inference.
People
This organization has no public members. You must be a member to see who’s a part of this organization.
Top languages
Loading…
Most used topics
Loading…