<feed xmlns='http://www.w3.org/2005/Atom'>
<title>ffwd/pplx_embed_cuda.cu, branch master</title>
<subtitle>Fastforward embedding inference engine</subtitle>
<id>https://git.onurgunduz.com/ffwd/atom/pplx_embed_cuda.cu?h=master</id>
<link rel='self' href='https://git.onurgunduz.com/ffwd/atom/pplx_embed_cuda.cu?h=master'/>
<link rel='alternate' type='text/html' href='https://git.onurgunduz.com/ffwd/'/>
<updated>2026-06-14T06:27:03Z</updated>
<entry>
<title>gitignore: track new binary and library names</title>
<updated>2026-06-14T06:27:03Z</updated>
<author>
<name>Onur Gündüz</name>
<email>ogunduz@gmail.com</email>
</author>
<published>2026-06-10T06:24:39Z</published>
<link rel='alternate' type='text/html' href='https://git.onurgunduz.com/ffwd/commit/?id=2c37727e653d2340d8977dd5db4addc767dd81ea'/>
<id>urn:sha1:2c37727e653d2340d8977dd5db4addc767dd81ea</id>
<content type='text'>
</content>
</entry>
<entry>
<title>cli: rename flags, add --stream, split serving into pplx-embed-server</title>
<updated>2026-06-14T06:27:03Z</updated>
<author>
<name>Onur Gündüz</name>
<email>ogunduz@gmail.com</email>
</author>
<published>2026-06-10T05:30:34Z</published>
<link rel='alternate' type='text/html' href='https://git.onurgunduz.com/ffwd/commit/?id=c14151290df20e1a2ebc9fb06a77b075ca4ad959'/>
<id>urn:sha1:c14151290df20e1a2ebc9fb06a77b075ca4ad959</id>
<content type='text'>
</content>
</entry>
<entry>
<title>cuda: store BF16 activations directly in norm and SiLU kernels</title>
<updated>2026-06-14T06:27:03Z</updated>
<author>
<name>Onur Gündüz</name>
<email>ogunduz@gmail.com</email>
</author>
<published>2026-06-10T04:11:26Z</published>
<link rel='alternate' type='text/html' href='https://git.onurgunduz.com/ffwd/commit/?id=399e35879492269584c8aa18eccb3b1b10527a74'/>
<id>urn:sha1:399e35879492269584c8aa18eccb3b1b10527a74</id>
<content type='text'>
</content>
</entry>
<entry>
<title>cuda: add optional BF16 CUDA weight storage (--cuda-weights bf16)</title>
<updated>2026-06-14T06:27:03Z</updated>
<author>
<name>Onur Gündüz</name>
<email>ogunduz@gmail.com</email>
</author>
<published>2026-06-08T23:05:01Z</published>
<link rel='alternate' type='text/html' href='https://git.onurgunduz.com/ffwd/commit/?id=3613657e678c54500f122c5b305f9cc9f5032d21'/>
<id>urn:sha1:3613657e678c54500f122c5b305f9cc9f5032d21</id>
<content type='text'>
</content>
</entry>
<entry>
<title>cuda: unify attention on one path; drop the max_seq&gt;=64 dispatch</title>
<updated>2026-06-14T06:27:03Z</updated>
<author>
<name>Onur Gündüz</name>
<email>ogunduz@gmail.com</email>
</author>
<published>2026-06-08T22:10:27Z</published>
<link rel='alternate' type='text/html' href='https://git.onurgunduz.com/ffwd/commit/?id=f4bc5ad86de55257c1f2a3a9de0a0e9fc91bfd3e'/>
<id>urn:sha1:f4bc5ad86de55257c1f2a3a9de0a0e9fc91bfd3e</id>
<content type='text'>
</content>
</entry>
<entry>
<title>cuda: add fast GEMM mode (--cuda-fast-gemm mode)</title>
<updated>2026-06-14T06:27:03Z</updated>
<author>
<name>Onur Gündüz</name>
<email>ogunduz@gmail.com</email>
</author>
<published>2026-06-08T21:27:02Z</published>
<link rel='alternate' type='text/html' href='https://git.onurgunduz.com/ffwd/commit/?id=ccceb5d0594e6879d4e2e4236980d233ea76c7ed'/>
<id>urn:sha1:ccceb5d0594e6879d4e2e4236980d233ea76c7ed</id>
<content type='text'>
</content>
</entry>
<entry>
<title>cuda: clarify comment gemm_compute</title>
<updated>2026-06-14T06:27:03Z</updated>
<author>
<name>Onur Gündüz</name>
<email>ogunduz@gmail.com</email>
</author>
<published>2026-06-08T20:48:11Z</published>
<link rel='alternate' type='text/html' href='https://git.onurgunduz.com/ffwd/commit/?id=940303fe2fa3d55746e86fc199d8900f0634568a'/>
<id>urn:sha1:940303fe2fa3d55746e86fc199d8900f0634568a</id>
<content type='text'>
</content>
</entry>
<entry>
<title>cuda: add opt-in BF16 tensor-core GEMM mode</title>
<updated>2026-06-14T06:27:03Z</updated>
<author>
<name>Onur Gündüz</name>
<email>ogunduz@gmail.com</email>
</author>
<published>2026-06-08T20:41:06Z</published>
<link rel='alternate' type='text/html' href='https://git.onurgunduz.com/ffwd/commit/?id=26aa937f023652648f201ad849d8dd644ca0261c'/>
<id>urn:sha1:26aa937f023652648f201ad849d8dd644ca0261c</id>
<content type='text'>
</content>
</entry>
<entry>
<title>cuda: replace long-seq attention with cuBLAS GEMM attention</title>
<updated>2026-06-14T06:27:03Z</updated>
<author>
<name>Onur Gündüz</name>
<email>ogunduz@gmail.com</email>
</author>
<published>2026-06-08T19:27:37Z</published>
<link rel='alternate' type='text/html' href='https://git.onurgunduz.com/ffwd/commit/?id=0b8fb1d850ffb0b1002767dbc5b533158c318e4f'/>
<id>urn:sha1:0b8fb1d850ffb0b1002767dbc5b533158c318e4f</id>
<content type='text'>
</content>
</entry>
<entry>
<title>cuda: replace tiled attention with warp-per-query reduction</title>
<updated>2026-06-14T06:27:03Z</updated>
<author>
<name>Onur Gündüz</name>
<email>ogunduz@gmail.com</email>
</author>
<published>2026-06-08T17:24:20Z</published>
<link rel='alternate' type='text/html' href='https://git.onurgunduz.com/ffwd/commit/?id=6e5065044750afb6e933f0b289fdd2041ac3dc55'/>
<id>urn:sha1:6e5065044750afb6e933f0b289fdd2041ac3dc55</id>
<content type='text'>
</content>
</entry>
</feed>
