Simsimd

Add

* `cos_f16_sve` common metric kind for Arm SVE ([7b29466](https://github.com/ashvardanian/SimSIMD/commit/7b2946665a409f468963fc4ffb622bd1107ea8a7))
* `simsimd_cos_i8x16neon` ([4dfe9d5](https://github.com/ashvardanian/SimSIMD/commit/4dfe9d577ffd5cf34c12e6b26a99726740dbe50c))
* `to_int` method for capsules ([c0ca719](https://github.com/ashvardanian/SimSIMD/commit/c0ca719cd84917c0c31dd2ee4536cf60e25c15ea))
* Arm Neon f16 dot product, refactor visibility ([23567db](https://github.com/ashvardanian/SimSIMD/commit/23567db33400f897d5559f295da86dd21059ccf2))
* Arm SVE acceleration for f16 dot-product, L2 ([de34a04](https://github.com/ashvardanian/SimSIMD/commit/de34a04a6be464edc9a1eeab498c63bc8eaebc98))
* AVX-512 for dot-product on Sapphire Rapids ([401ed6b](https://github.com/ashvardanian/SimSIMD/commit/401ed6b285823f421ea01a19baf54c6e603f9d2c))
* Benchmarks for cosine similarity ([e29c177](https://github.com/ashvardanian/SimSIMD/commit/e29c177a328e705c4bd9278cad9a219e51572a51))
* C99 version ([06fb120](https://github.com/ashvardanian/SimSIMD/commit/06fb120681f60115ade39d9e6a2648740e2273ff))
* capsule for dot_f32sve ([b5cb7b1](https://github.com/ashvardanian/SimSIMD/commit/b5cb7b16e0080d9ec68c2d71b1abbb94ac619656))
* capsule for dot_f32sve ([ff23d94](https://github.com/ashvardanian/SimSIMD/commit/ff23d94cc7a27fd0f3b8b3f94699056e6de9ed98))
* Capsules for Chemistry research ([bf6ec11](https://github.com/ashvardanian/SimSIMD/commit/bf6ec11a6052ee12277fd51919b689daaa6b5be6))
* configuration for test debug ([8a82485](https://github.com/ashvardanian/SimSIMD/commit/8a82485302359071b73b285d966fcd641e927abb))
* Graviton 3 results ([8ea912a](https://github.com/ashvardanian/SimSIMD/commit/8ea912ada6cf45b13864f5a7039d93f6c714e8b9))
* Hamming distance with scalar, AVX-512, NEON ([cbd2b85](https://github.com/ashvardanian/SimSIMD/commit/cbd2b85c946e02cad1cb11c5961251220869c331))
* Initial benchmark and AVX2, Neon, SVE ([fb0aa11](https://github.com/ashvardanian/SimSIMD/commit/fb0aa1136e155ec23410bc9dc0e8520dec04c207))
* Separate chemsitry-oriented measures ([dc00183](https://github.com/ashvardanian/SimSIMD/commit/dc00183bf9e413044a778d4c21f43041c5a6e818))
* SVE cosine distance of dynamic length vecs ([3162f7d](https://github.com/ashvardanian/SimSIMD/commit/3162f7df364f311021622c603716a7f6f9424d75))
* SVE-accelerated Hamming distance ([e407ab3](https://github.com/ashvardanian/SimSIMD/commit/e407ab353c2b277e8da81255eee52f1664feb69c))
* Tanimoto distances ([168f026](https://github.com/ashvardanian/SimSIMD/commit/168f026aa4007c39c6bc3f26a97905cd16bb44d7))
* test for module using ([c5afa3a](https://github.com/ashvardanian/SimSIMD/commit/c5afa3af53c30267e7efa678548b966e10ffc4ae))

Build

* AVX-512 code for Hamming distance ([d6a7820](https://github.com/ashvardanian/SimSIMD/commit/d6a782061a6b42e473d558279143556c851f867f))
* cpython module and test ([ac5985e](https://github.com/ashvardanian/SimSIMD/commit/ac5985e3dcbb46b0e08fed8279fa841031d42f89))

Create

* CPython wrapper ([54b5643](https://github.com/ashvardanian/SimSIMD/commit/54b5643b3eacbea1d924e6b5559aa7209f81b32b))
* CPython wrapper ([9612b04](https://github.com/ashvardanian/SimSIMD/commit/9612b042e068ea335626019a1cfe5835a7707bf1))

Docs

* Add functionality table ([8b08703](https://github.com/ashvardanian/SimSIMD/commit/8b087039241a8eb92717622b1643b3402e8423ff))
* Add functor annotation ([a523a7b](https://github.com/ashvardanian/SimSIMD/commit/a523a7b801fbbda679f58e59cc7fa5400cd62093))
* Added AVX-FMA benchmarks on Threadripper ([b683bf0](https://github.com/ashvardanian/SimSIMD/commit/b683bf0a59dc634bc4a3d5eb17fcaf42f8b767e9))
* Indent ([649e6a0](https://github.com/ashvardanian/SimSIMD/commit/649e6a0bf9d2a68a51a90a4803717f9557c72dfb))
* L2 annotation ([2577dfc](https://github.com/ashvardanian/SimSIMD/commit/2577dfc8357c9aa5f2dc3d8c8882fb12155f86f8))
* Reordered ([dde94a8](https://github.com/ashvardanian/SimSIMD/commit/dde94a81ac01014bea62f18ac874c0423185428c))

Fix

* `simsimd_cos_i8x16neon` for Gravitons ([ccad0b4](https://github.com/ashvardanian/SimSIMD/commit/ccad0b41f5e88742d1276016ad1e9a5d636d2589))
* Arm Neon ordering FMA arguments ([2ab1293](https://github.com/ashvardanian/SimSIMD/commit/2ab12935669229b2cbf8609f83794528cdf2584d))
* Benchmark compilation ([9209bfb](https://github.com/ashvardanian/SimSIMD/commit/9209bfbc7a65058fc61d1c2e3b2adaa0f28871b7))
* Capsule destructors ([12f756e](https://github.com/ashvardanian/SimSIMD/commit/12f756e4552b65fb3f25f4a111df4aa141652e14))
* Casting to `f16` ([709e346](https://github.com/ashvardanian/SimSIMD/commit/709e346bff72061cabf1a009ec18cbce43d403f9))
* Casting to f16 ([6245890](https://github.com/ashvardanian/SimSIMD/commit/62458907c69ca3a119e3e076aa9553ca377315d5))
* Compilation ([905942a](https://github.com/ashvardanian/SimSIMD/commit/905942a0d2ccb5eacb7028a8b3303c833ca80045))
* Compilation error - unnamed argument ([1311da4](https://github.com/ashvardanian/SimSIMD/commit/1311da476dd251e1780d81be9ba1f11bde5fe279))
* Cosine similarity with Arm Neon ([d2d8ad5](https://github.com/ashvardanian/SimSIMD/commit/d2d8ad5dae3f8c8ac2181c08bc1afb44d87ba2b2))
* Duplicate symbol in cosine similarity ([d791376](https://github.com/ashvardanian/SimSIMD/commit/d7913763316b865c54e722940d24b3cd042c30f8))
* Hamming iteration step with SVE ([e0194fa](https://github.com/ashvardanian/SimSIMD/commit/e0194facfcc366b324c7922400b084b58d3158a2))
* IP error compensation with normalization ([b577ad8](https://github.com/ashvardanian/SimSIMD/commit/b577ad882eac54f9ad61a69979489557b8b1257e))
* Missing includes for `size_t` ([74760c7](https://github.com/ashvardanian/SimSIMD/commit/74760c7d9a018bb6c959db75e7898186d79a9faa))
* Missing MSVC intrinsics ([0fe4c1e](https://github.com/ashvardanian/SimSIMD/commit/0fe4c1edff00d6f13eb7bc52fd7a8fe153346dd0))
* Module name ([8a08231](https://github.com/ashvardanian/SimSIMD/commit/8a08231024576b0ef53e9be7ab0032e56edcaa0c))
* Quantized i8 dot-product on AVX ([725152d](https://github.com/ashvardanian/SimSIMD/commit/725152d733986e49a90212de3473d82591c11965))
* Separate similarity and distance measures ([3e922ce](https://github.com/ashvardanian/SimSIMD/commit/3e922ceb783c6caa4097941de10c163abec2e929))

Improve

* Extended benchmarks to include `chem.h` ([b222949](https://github.com/ashvardanian/SimSIMD/commit/b222949e99bfc2d95123dd0b32955aa5a42f7f4a))
* Using PyTest ([8c29c45](https://github.com/ashvardanian/SimSIMD/commit/8c29c4503e3ca03fdf3cf645b7ec532a1b564b7b))

Make

* Downgrading to C++11, CMake imterface ([add627d](https://github.com/ashvardanian/SimSIMD/commit/add627d3a6b873b57d85d8b5e127c833b8ba6c00))
* Fix tests ([a034a5c](https://github.com/ashvardanian/SimSIMD/commit/a034a5cdea79063ac86c463c7dc6bdb82ec3ef72))
* Script to update version ([9b12d67](https://github.com/ashvardanian/SimSIMD/commit/9b12d67282eac7d812a9b90574d129fe233a2e54))
* Semantic Releases ([208290e](https://github.com/ashvardanian/SimSIMD/commit/208290e22947e620e007ce76f658f24a8cadf01e))
* Specifying the header sources ([ce95f7c](https://github.com/ashvardanian/SimSIMD/commit/ce95f7ceb297a18e32db12fc179ebf2ed49cb05f))
* Using different debuggers on MacOS ([7390b7d](https://github.com/ashvardanian/SimSIMD/commit/7390b7df73e00e7458e38c3cefdc0e7039dba4d6))

Refactor

* From `vmlaq_f32` to `vfmaq_f32` ([16f6f34](https://github.com/ashvardanian/SimSIMD/commit/16f6f34191732b4173e09068e5ce67aa90ceec84))
* From similarities to distances ([5b02e49](https://github.com/ashvardanian/SimSIMD/commit/5b02e49ad01850a60e99100806f239b1f0465b77))
* move to python folder ([8ec854f](https://github.com/ashvardanian/SimSIMD/commit/8ec854fba7ea188e6e5f1a1a1440cdba526570b0))
* move to python folder ([a507c56](https://github.com/ashvardanian/SimSIMD/commit/a507c56c04cb1aa08a172f69822316199779d3e2))
* Project name ([8648ee0](https://github.com/ashvardanian/SimSIMD/commit/8648ee03635f32f5843a85400dd41acb1dad5033))
* Shorter function names ([1d20d4d](https://github.com/ashvardanian/SimSIMD/commit/1d20d4dc8305f7838a00f82e528effe4f13307a4))
* Switching C++11 -> C99 ([3bd535f](https://github.com/ashvardanian/SimSIMD/commit/3bd535f18b44b7dc6967316108e96192bcad1c62))
* Type names ([63a27ec](https://github.com/ashvardanian/SimSIMD/commit/63a27ec223c2399eb9ff268fa23b7229fd794b6f))

| `avx512_f32_js_1536d` | 1.127 M/s | 13.84 G/s | 0.001 | 345u |
| `avx512_f16_js_1536d` | 2.139 M/s | 13.14 G/s | 0.070 | 0.020 |
| `avx2_f16_js_1536d` | 0.547 M/s | 3.36 G/s | 0.011 | 0.003 |

Of course, the results will vary depending on the vector size. I generally use 1536 dimensions, matching the size of OpenAI Ada embeddings, standard in NLP workloads. The Jensen Shannon divergence, however, is used broadly in other domains of statistics, bio-informatics, and chem-informatics, so I'm adding it as a new out-of-the-box supported metric into [USearch](https://github.com/unum-cloud/usearch) today 🥳

This further accelerates the k-approximate Nearest Neighbors Search and the [clustering of Billions of different protein sequences](https://github.com/unum-cloud/usearch#clustering) without alignment procedures. [Expect one more "Less Slow" post soon!](https://ashvardanian.com/tags/less-slow/) 🤗