node-llama-cpp 2.8.14 → 2.8.16

package/README.md

@@ -26,10 +26,10 @@
 * Up-to-date with the latest version of `llama.cpp`. Download and compile the latest release with a single CLI command.
 * Force a model to generate output in a parseable format, like JSON, or even force it to follow a specific JSON schema
 
-## [Documentation](https://withcatai.github.io/node-llama-cpp/)
-* [Getting started guide](https://withcatai.github.io/node-llama-cpp/guide/)
-* [API reference](https://withcatai.github.io/node-llama-cpp/api/classes/LlamaModel)
-* [CLI help](https://withcatai.github.io/node-llama-cpp/guide/cli/)
+## [Documentation](https://node-llama-cpp.withcat.ai/)
+* [Getting started guide](https://node-llama-cpp.withcat.ai/guide/)
+* [API reference](https://node-llama-cpp.withcat.ai/api/classes/LlamaModel)
+* [CLI help](https://node-llama-cpp.withcat.ai/guide/cli/)
 * [Changelog](https://github.com/withcatai/node-llama-cpp/releases)
 * [Roadmap](https://github.com/orgs/withcatai/projects/1)
 
@@ -72,10 +72,10 @@ const a2 = await session.prompt(q2);
 console.log("AI: " + a2);
 ```
 
-> For more examples, see the [getting started guide](https://withcatai.github.io/node-llama-cpp/guide/)
+> For more examples, see the [getting started guide](https://node-llama-cpp.withcat.ai/guide/)
 
 ## Contributing
-To contribute to `node-llama-cpp` read the [contribution guide](https://withcatai.github.io/node-llama-cpp/guide/contributing).
+To contribute to `node-llama-cpp` read the [contribution guide](https://node-llama-cpp.withcat.ai/guide/contributing).
 
 ## Acknowledgements
 * llama.cpp: [ggerganov/llama.cpp](https://github.com/ggerganov/llama.cpp)

package/dist/config.js

@@ -53,7 +53,7 @@ export const defaultChatSystemPrompt = "You are a helpful, respectful and honest
     "If you don't know the answer to a question, please don't share false information.";
 export const cliBinName = "node-llama-cpp";
 export const npxRunPrefix = "npx --no ";
-const documentationUrl = "https://withcatai.github.io/node-llama-cpp";
+const documentationUrl = "https://node-llama-cpp.withcat.ai";
 export const documentationPageUrls = {
     CUDA: documentationUrl + "/guide/CUDA"
 };

package/dist/config.js.map

@@ -1 +1 @@
-{"version":3,"file":"config.js","sourceRoot":"","sources":["../src/config.ts"],"names":[],"mappings":"AAAA,OAAO,EAAC,aAAa,EAAC,MAAM,KAAK,CAAC;AAClC,OAAO,KAAK,IAAI,MAAM,MAAM,CAAC;AAC7B,OAAO,KAAK,EAAE,MAAM,IAAI,CAAC;AACzB,OAAO,OAAO,MAAM,SAAS,CAAC;AAC9B,OAAO,MAAM,MAAM,SAAS,CAAC;AAC7B,OAAO,KAAK,IAAI,MAAM,MAAM,CAAC;AAC7B,OAAO,EAAC,wBAAwB,EAAC,MAAM,kCAAkC,CAAC;AAE1E,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAE/D,MAAM,GAAG,GAAG,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;AAGrC,MAAM,CAAC,MAAM,cAAc,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC;AAClE,MAAM,CAAC,MAAM,wBAAwB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,YAAY,CAAC,CAAC;AAChF,MAAM,CAAC,MAAM,kBAAkB,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,EAAE,WAAW,CAAC,CAAC;AAC1E,MAAM,CAAC,MAAM,0BAA0B,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,EAAE,OAAO,EAAE,UAAU,CAAC,CAAC;AAC1F,MAAM,CAAC,MAAM,iBAAiB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,WAAW,CAAC,CAAC;AACxE,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,WAAW,EAAE,UAAU,CAAC,CAAC;AAC5F,MAAM,CAAC,MAAM,qBAAqB,GAAG,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,MAAM,EAAE,EAAE,gBAAgB,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC;AACzF,MAAM,CAAC,MAAM,0BAA0B,GAAG,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,EAAE,mCAAmC,CAAC,CAAC;AACvG,MAAM,CAAC,MAAM,mBAAmB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,cAAc,CAAC,CAAC;AAC7E,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,4BAA4B,CAAC,CAAC;AACjG,MAAM,CAAC,MAAM,4BAA4B,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,oBAAoB,CAAC,CAAC;AAC5F,MAAM,CAAC,MAAM,2BAA2B,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,mBAAmB,CAAC,CAAC;AAC1F,MAAM,CAAC,MAAM,cAAc,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,CAAC,CAAC;AACjE,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,CAAC,CAAC;AAC5E,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,CAAC,CAAC;AAC5E,MAAM,CAAC,MAAM,UAAU,GAAG,SAAS,CAAC;AAEpC,MAAM,CAAC,MAAM,IAAI,GAAG,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC;KAC5B,OAAO,CAAC,OAAO,CAAC;KAChB,MAAM,EAAE,CAAC;AACd,MAAM,CAAC,MAAM,yBAAyB,GAAG,GAAG,CAAC,GAAG,CAAC,qBAAqB,CAAC;KAClE,OAAO,CAAC,qBAAqB,CAAC;KAC9B,QAAQ,EAAE,CAAC;AAChB,MAAM,CAAC,MAAM,sBAAsB,GAAG,GAAG,CAAC,GAAG,CAAC,6BAA6B,CAAC;KACvE,OAAO,CAAC,MAAM,wBAAwB,EAAE,CAAC;KACzC,QAAQ,EAAE,CAAC;AAChB,MAAM,CAAC,MAAM,2BAA2B,GAAG,GAAG,CAAC,GAAG,CAAC,sBAAsB,CAAC;KACrE,OAAO,CAAC,OAAO,CAAC,QAAQ,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC;KACzD,MAAM,EAAE,CAAC;AACd,MAAM,CAAC,MAAM,0BAA0B,GAAG,GAAG,CAAC,GAAG,CAAC,qBAAqB,CAAC;KACnE,OAAO,CAAC,OAAO,CAAC;KAChB,MAAM,EAAE,CAAC;AACd,MAAM,CAAC,MAAM,mBAAmB,GAAG,GAAG,CAAC,GAAG,CAAC,8BAA8B,CAAC;KACrE,OAAO,CAAC,OAAO,CAAC;KAChB,MAAM,EAAE,CAAC;AACd,MAAM,CAAC,MAAM,2BAA2B,GAAG,GAAG,CAAC,GAAG,CAAC,oCAAoC,CAAC;KACnF,OAAO,CAAC,yBAAyB,CAAC;KAClC,QAAQ,EAAE,CAAC;AAChB,MAAM,CAAC,MAAM,2BAA2B,GAAG,GAAG,CAAC,GAAG,CAAC,oCAAoC,CAAC;KACnF,OAAO,CAAC,yBAAyB,CAAC;KAClC,QAAQ,EAAE,CAAC;AAChB,MAAM,CAAC,MAAM,8BAA8B,GAAG,8BAA8B,CAAC;AAC7E,MAAM,CAAC,MAAM,uBAAuB,GAAG,+FAA+F;IAClI,+HAA+H;IAC/H,mFAAmF,CAAC;AACxF,MAAM,CAAC,MAAM,UAAU,GAAG,gBAAgB,CAAC;AAC3C,MAAM,CAAC,MAAM,YAAY,GAAG,WAAW,CAAC;AAExC,MAAM,gBAAgB,GAAG,4CAA4C,CAAC;AACtE,MAAM,CAAC,MAAM,qBAAqB,GAAG;IACjC,IAAI,EAAE,gBAAgB,GAAG,aAAa;CAChC,CAAC"}
+{"version":3,"file":"config.js","sourceRoot":"","sources":["../src/config.ts"],"names":[],"mappings":"AAAA,OAAO,EAAC,aAAa,EAAC,MAAM,KAAK,CAAC;AAClC,OAAO,KAAK,IAAI,MAAM,MAAM,CAAC;AAC7B,OAAO,KAAK,EAAE,MAAM,IAAI,CAAC;AACzB,OAAO,OAAO,MAAM,SAAS,CAAC;AAC9B,OAAO,MAAM,MAAM,SAAS,CAAC;AAC7B,OAAO,KAAK,IAAI,MAAM,MAAM,CAAC;AAC7B,OAAO,EAAC,wBAAwB,EAAC,MAAM,kCAAkC,CAAC;AAE1E,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAE/D,MAAM,GAAG,GAAG,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;AAGrC,MAAM,CAAC,MAAM,cAAc,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC;AAClE,MAAM,CAAC,MAAM,wBAAwB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,YAAY,CAAC,CAAC;AAChF,MAAM,CAAC,MAAM,kBAAkB,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,EAAE,WAAW,CAAC,CAAC;AAC1E,MAAM,CAAC,MAAM,0BAA0B,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,EAAE,OAAO,EAAE,UAAU,CAAC,CAAC;AAC1F,MAAM,CAAC,MAAM,iBAAiB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,WAAW,CAAC,CAAC;AACxE,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,WAAW,EAAE,UAAU,CAAC,CAAC;AAC5F,MAAM,CAAC,MAAM,qBAAqB,GAAG,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,MAAM,EAAE,EAAE,gBAAgB,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC;AACzF,MAAM,CAAC,MAAM,0BAA0B,GAAG,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,EAAE,mCAAmC,CAAC,CAAC;AACvG,MAAM,CAAC,MAAM,mBAAmB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,cAAc,CAAC,CAAC;AAC7E,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,4BAA4B,CAAC,CAAC;AACjG,MAAM,CAAC,MAAM,4BAA4B,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,oBAAoB,CAAC,CAAC;AAC5F,MAAM,CAAC,MAAM,2BAA2B,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,mBAAmB,CAAC,CAAC;AAC1F,MAAM,CAAC,MAAM,cAAc,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,CAAC,CAAC;AACjE,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,CAAC,CAAC;AAC5E,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,CAAC,CAAC;AAC5E,MAAM,CAAC,MAAM,UAAU,GAAG,SAAS,CAAC;AAEpC,MAAM,CAAC,MAAM,IAAI,GAAG,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC;KAC5B,OAAO,CAAC,OAAO,CAAC;KAChB,MAAM,EAAE,CAAC;AACd,MAAM,CAAC,MAAM,yBAAyB,GAAG,GAAG,CAAC,GAAG,CAAC,qBAAqB,CAAC;KAClE,OAAO,CAAC,qBAAqB,CAAC;KAC9B,QAAQ,EAAE,CAAC;AAChB,MAAM,CAAC,MAAM,sBAAsB,GAAG,GAAG,CAAC,GAAG,CAAC,6BAA6B,CAAC;KACvE,OAAO,CAAC,MAAM,wBAAwB,EAAE,CAAC;KACzC,QAAQ,EAAE,CAAC;AAChB,MAAM,CAAC,MAAM,2BAA2B,GAAG,GAAG,CAAC,GAAG,CAAC,sBAAsB,CAAC;KACrE,OAAO,CAAC,OAAO,CAAC,QAAQ,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC;KACzD,MAAM,EAAE,CAAC;AACd,MAAM,CAAC,MAAM,0BAA0B,GAAG,GAAG,CAAC,GAAG,CAAC,qBAAqB,CAAC;KACnE,OAAO,CAAC,OAAO,CAAC;KAChB,MAAM,EAAE,CAAC;AACd,MAAM,CAAC,MAAM,mBAAmB,GAAG,GAAG,CAAC,GAAG,CAAC,8BAA8B,CAAC;KACrE,OAAO,CAAC,OAAO,CAAC;KAChB,MAAM,EAAE,CAAC;AACd,MAAM,CAAC,MAAM,2BAA2B,GAAG,GAAG,CAAC,GAAG,CAAC,oCAAoC,CAAC;KACnF,OAAO,CAAC,yBAAyB,CAAC;KAClC,QAAQ,EAAE,CAAC;AAChB,MAAM,CAAC,MAAM,2BAA2B,GAAG,GAAG,CAAC,GAAG,CAAC,oCAAoC,CAAC;KACnF,OAAO,CAAC,yBAAyB,CAAC;KAClC,QAAQ,EAAE,CAAC;AAChB,MAAM,CAAC,MAAM,8BAA8B,GAAG,8BAA8B,CAAC;AAC7E,MAAM,CAAC,MAAM,uBAAuB,GAAG,+FAA+F;IAClI,+HAA+H;IAC/H,mFAAmF,CAAC;AACxF,MAAM,CAAC,MAAM,UAAU,GAAG,gBAAgB,CAAC;AAC3C,MAAM,CAAC,MAAM,YAAY,GAAG,WAAW,CAAC;AAExC,MAAM,gBAAgB,GAAG,mCAAmC,CAAC;AAC7D,MAAM,CAAC,MAAM,qBAAqB,GAAG;IACjC,IAAI,EAAE,gBAAgB,GAAG,aAAa;CAChC,CAAC"}

package/dist/utils/compileLLamaCpp.js

@@ -18,34 +18,30 @@ export async function compileLlamaCpp({ arch = process.arch, nodeTarget = proces
         const toolchainFile = await getToolchainFileForArch(arch);
         const runtimeVersion = nodeTarget.startsWith("v") ? nodeTarget.slice("v".length) : nodeTarget;
         const cmakeCustomOptions = new Map();
-        if ((metal && process.platform === "darwin") || process.env.LLAMA_METAL === "1")
-            cmakeCustomOptions.set("LLAMA_METAL", "1");
+        if ((metal && process.platform === "darwin") || process.env.GGML_METAL === "1")
+            cmakeCustomOptions.set("GGML_METAL", "1");
         else
-            cmakeCustomOptions.set("LLAMA_METAL", "OFF");
-        if (cuda || process.env.LLAMA_CUBLAS === "1")
-            cmakeCustomOptions.set("LLAMA_CUBLAS", "1");
-        if (process.env.LLAMA_MPI === "1")
-            cmakeCustomOptions.set("LLAMA_MPI", "1");
-        if (process.env.LLAMA_OPENBLAS === "1")
-            cmakeCustomOptions.set("LLAMA_OPENBLAS", "1");
-        if (process.env.LLAMA_BLAS_VENDOR != null)
-            cmakeCustomOptions.set("LLAMA_BLAS_VENDOR", process.env.LLAMA_BLAS_VENDOR);
-        if (process.env.LLAMA_CUDA_FORCE_DMMV != null)
-            cmakeCustomOptions.set("LLAMA_CUDA_FORCE_DMMV", process.env.LLAMA_CUDA_FORCE_DMMV);
-        if (process.env.LLAMA_CUDA_DMMV_X != null)
-            cmakeCustomOptions.set("LLAMA_CUDA_DMMV_X", process.env.LLAMA_CUDA_DMMV_X);
-        if (process.env.LLAMA_CUDA_MMV_Y != null)
-            cmakeCustomOptions.set("LLAMA_CUDA_MMV_Y", process.env.LLAMA_CUDA_MMV_Y);
-        if (process.env.LLAMA_CUDA_F16 != null)
-            cmakeCustomOptions.set("LLAMA_CUDA_F16", process.env.LLAMA_CUDA_F16);
-        if (process.env.LLAMA_CUDA_KQUANTS_ITER != null)
-            cmakeCustomOptions.set("LLAMA_CUDA_KQUANTS_ITER", process.env.LLAMA_CUDA_KQUANTS_ITER);
-        if (process.env.LLAMA_CUDA_PEER_MAX_BATCH_SIZE != null)
-            cmakeCustomOptions.set("LLAMA_CUDA_PEER_MAX_BATCH_SIZE", process.env.LLAMA_CUDA_PEER_MAX_BATCH_SIZE);
-        if (process.env.LLAMA_HIPBLAS === "1")
-            cmakeCustomOptions.set("LLAMA_HIPBLAS", "1");
-        if (process.env.LLAMA_CLBLAST === "1")
-            cmakeCustomOptions.set("LLAMA_CLBLAST", "1");
+            cmakeCustomOptions.set("GGML_METAL", "OFF");
+        if (cuda || process.env.GGML_CUDA === "1")
+            cmakeCustomOptions.set("GGML_CUDA", "1");
+        if (process.env.GGML_OPENBLAS === "1")
+            cmakeCustomOptions.set("GGML_OPENBLAS", "1");
+        if (process.env.GGML_BLAS_VENDOR != null)
+            cmakeCustomOptions.set("GGML_BLAS_VENDOR", process.env.GGML_BLAS_VENDOR);
+        if (process.env.GGML_CUDA_FORCE_DMMV != null)
+            cmakeCustomOptions.set("GGML_CUDA_FORCE_DMMV", process.env.GGML_CUDA_FORCE_DMMV);
+        if (process.env.GGML_CUDA_DMMV_X != null)
+            cmakeCustomOptions.set("GGML_CUDA_DMMV_X", process.env.GGML_CUDA_DMMV_X);
+        if (process.env.GGML_CUDA_MMV_Y != null)
+            cmakeCustomOptions.set("GGML_CUDA_MMV_Y", process.env.GGML_CUDA_MMV_Y);
+        if (process.env.GGML_CUDA_F16 != null)
+            cmakeCustomOptions.set("GGML_CUDA_F16", process.env.GGML_CUDA_F16);
+        if (process.env.GGML_CUDA_KQUANTS_ITER != null)
+            cmakeCustomOptions.set("GGML_CUDA_KQUANTS_ITER", process.env.GGML_CUDA_KQUANTS_ITER);
+        if (process.env.GGML_CUDA_PEER_MAX_BATCH_SIZE != null)
+            cmakeCustomOptions.set("GGML_CUDA_PEER_MAX_BATCH_SIZE", process.env.GGML_CUDA_PEER_MAX_BATCH_SIZE);
+        if (process.env.GGML_HIPBLAS === "1")
+            cmakeCustomOptions.set("GGML_HIPBLAS", "1");
         if (toolchainFile != null)
             cmakeCustomOptions.set("CMAKE_TOOLCHAIN_FILE", toolchainFile);
         for (const key in process.env) {

package/dist/utils/compileLLamaCpp.js.map

@@ -1 +1 @@
-{"version":3,"file":"compileLLamaCpp.js","sourceRoot":"","sources":["../../src/utils/compileLLamaCpp.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,MAAM,CAAC;AACxB,OAAO,EAAC,aAAa,EAAC,MAAM,KAAK,CAAC;AAClC,OAAO,OAAO,MAAM,SAAS,CAAC;AAC9B,OAAO,EAAE,MAAM,UAAU,CAAC;AAC1B,OAAO,KAAK,MAAM,OAAO,CAAC;AAC1B,OAAO,EACH,8BAA8B,EAAE,qBAAqB,EAAE,iBAAiB,EAAE,cAAc,EAAE,wBAAwB,EACrH,MAAM,cAAc,CAAC;AACtB,OAAO,EAAC,eAAe,EAAC,MAAM,sBAAsB,CAAC;AACrD,OAAO,EAAC,cAAc,EAAC,MAAM,kBAAkB,CAAC;AAChD,OAAO,EAAC,YAAY,EAAC,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAC,mBAAmB,EAAE,YAAY,EAAE,eAAe,EAAC,MAAM,YAAY,CAAC;AAE9E,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAE/D,MAAM,CAAC,KAAK,UAAU,eAAe,CAAC,EAClC,IAAI,GAAG,OAAO,CAAC,IAAI,EAAE,UAAU,GAAG,OAAO,CAAC,OAAO,EAAE,cAAc,EAAE,iBAAiB,GAAG,IAAI,EAAE,KAAK,GAAG,OAAO,CAAC,QAAQ,KAAK,QAAQ,EAClI,IAAI,GAAG,KAAK,EAGf;IACG,IAAI;QACA,IAAI,CAAC,CAAC,MAAM,EAAE,CAAC,UAAU,CAAC,iBAAiB,CAAC,CAAC,EAAE;YAC3C,MAAM,IAAI,KAAK,CAAC,IAAI,iBAAiB,4BAA4B,CAAC,CAAC;SACtE;QAED,MAAM,aAAa,GAAG,MAAM,gBAAgB,EAAE,CAAC;QAC/C,MAAM,aAAa,GAAG,MAAM,uBAAuB,CAAC,IAAI,CAAC,CAAC;QAC1D,MAAM,cAAc,GAAG,UAAU,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,KAAK,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC;QAC9F,MAAM,kBAAkB,GAAG,IAAI,GAAG,EAAkB,CAAC;QAErD,IAAI,CAAC,KAAK,IAAI,OAAO,CAAC,QAAQ,KAAK,QAAQ,CAAC,IAAI,OAAO,CAAC,GAAG,CAAC,WAAW,KAAK,GAAG;YAAE,kBAAkB,CAAC,GAAG,CAAC,aAAa,EAAE,GAAG,CAAC,CAAC;;YACvH,kBAAkB,CAAC,GAAG,CAAC,aAAa,EAAE,KAAK,CAAC,CAAC;QAElD,IAAI,IAAI,IAAI,OAAO,CAAC,GAAG,CAAC,YAAY,KAAK,GAAG;YAAE,kBAAkB,CAAC,GAAG,CAAC,cAAc,EAAE,GAAG,CAAC,CAAC;QAE1F,IAAI,OAAO,CAAC,GAAG,CAAC,SAAS,KAAK,GAAG;YAAE,kBAAkB,CAAC,GAAG,CAAC,WAAW,EAAE,GAAG,CAAC,CAAC;QAC5E,IAAI,OAAO,CAAC,GAAG,CAAC,cAAc,KAAK,GAAG;YAAE,kBAAkB,CAAC,GAAG,CAAC,gBAAgB,EAAE,GAAG,CAAC,CAAC;QACtF,IAAI,OAAO,CAAC,GAAG,CAAC,iBAAiB,IAAI,IAAI;YAAE,kBAAkB,CAAC,GAAG,CAAC,mBAAmB,EAAE,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC,CAAC;QACtH,IAAI,OAAO,CAAC,GAAG,CAAC,qBAAqB,IAAI,IAAI;YAAE,kBAAkB,CAAC,GAAG,CAAC,uBAAuB,EAAE,OAAO,CAAC,GAAG,CAAC,qBAAqB,CAAC,CAAC;QAClI,IAAI,OAAO,CAAC,GAAG,CAAC,iBAAiB,IAAI,IAAI;YAAE,kBAAkB,CAAC,GAAG,CAAC,mBAAmB,EAAE,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC,CAAC;QACtH,IAAI,OAAO,CAAC,GAAG,CAAC,gBAAgB,IAAI,IAAI;YAAE,kBAAkB,CAAC,GAAG,CAAC,kBAAkB,EAAE,OAAO,CAAC,GAAG,CAAC,gBAAgB,CAAC,CAAC;QACnH,IAAI,OAAO,CAAC,GAAG,CAAC,cAAc,IAAI,IAAI;YAAE,kBAAkB,CAAC,GAAG,CAAC,gBAAgB,EAAE,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC,CAAC;QAC7G,IAAI,OAAO,CAAC,GAAG,CAAC,uBAAuB,IAAI,IAAI;YAAE,kBAAkB,CAAC,GAAG,CAAC,yBAAyB,EAAE,OAAO,CAAC,GAAG,CAAC,uBAAuB,CAAC,CAAC;QACxI,IAAI,OAAO,CAAC,GAAG,CAAC,8BAA8B,IAAI,IAAI;YAAE,kBAAkB,CAAC,GAAG,CAAC,gCAAgC,EAAE,OAAO,CAAC,GAAG,CAAC,8BAA8B,CAAC,CAAC;QAC7J,IAAI,OAAO,CAAC,GAAG,CAAC,aAAa,KAAK,GAAG;YAAE,kBAAkB,CAAC,GAAG,CAAC,eAAe,EAAE,GAAG,CAAC,CAAC;QACpF,IAAI,OAAO,CAAC,GAAG,CAAC,aAAa,KAAK,GAAG;YAAE,kBAAkB,CAAC,GAAG,CAAC,eAAe,EAAE,GAAG,CAAC,CAAC;QAEpF,IAAI,aAAa,IAAI,IAAI;YACrB,kBAAkB,CAAC,GAAG,CAAC,sBAAsB,EAAE,aAAa,CAAC,CAAC;QAElE,KAAK,MAAM,GAAG,IAAI,OAAO,CAAC,GAAG,EAAE;YAC3B,IAAI,GAAG,CAAC,UAAU,CAAC,8BAA8B,CAAC,EAAE;gBAChD,MAAM,MAAM,GAAG,GAAG,CAAC,KAAK,CAAC,8BAA8B,CAAC,MAAM,CAAC,CAAC;gBAChE,MAAM,KAAK,GAAG,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;gBAC/B,kBAAkB,CAAC,GAAG,CAAC,MAAM,EAAE,KAAM,CAAC,CAAC;aAC1C;SACJ;QAED,MAAM,eAAe,EAAE,CAAC;QAExB,MAAM,YAAY,CAAC,KAAK,EAAE,CAAC,KAAK,EAAE,IAAI,EAAE,gBAAgB,EAAE,IAAI,EAAE,OAAO,EAAE,aAAa,EAAE,MAAM,EAAE,GAAG,aAAa,CAAC,EAAE,SAAS,CAAC,CAAC;QAE9H,MAAM,YAAY,CACd,KAAK,EACL,CAAC,KAAK,EAAE,IAAI,EAAE,gBAAgB,EAAE,IAAI,EAAE,SAAS,EAAE,aAAa,EAAE,MAAM,EAAE,SAAS,GAAG,IAAI,EAAE,oBAAoB,GAAG,cAAc,EAAE,GAAG,aAAa,CAAC;aAC7I,MAAM,CAAC,CAAC,GAAG,kBAAkB,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,MAAM,GAAG,GAAG,GAAG,GAAG,GAAG,KAAK,CAAC,CAAC,EACtF,SAAS,CACZ,CAAC;QAEF,MAAM,gBAAgB,GAAG;YACrB,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,EAAE,KAAK,CAAC;YACzC,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,EAAE,WAAW,EAAE,KAAK,CAAC;SACzD,CAAC;QACF,MAAM,qBAAqB,GAAG,MAAM,oBAAoB,CAAC,IAAI,CAAC,CAAC;QAE/D,KAAK,MAAM,eAAe,IAAI,gBAAgB,EAAE;YAC5C,IAAI,MAAM,EAAE,CAAC,UAAU,CAAC,eAAe,CAAC,EAAE;gBACtC,MAAM,KAAK,GAAG,MAAM,EAAE,CAAC,OAAO,CAAC,eAAe,CAAC,CAAC;gBAEhD,MAAM,OAAO,CAAC,GAAG,CACb,KAAK,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CACpB,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,eAAe,EAAE,QAAQ,CAAC,EAAE,IAAI,CAAC,IAAI,CAAC,qBAAqB,EAAE,QAAQ,CAAC,EAAE;oBACtF,SAAS,EAAE,KAAK;iBACnB,CAAC,CACL,CAAC,CACL,CAAC;aACL;SACJ;QAED,mBAAmB,CAAC,qBAAqB,EAAE,IAAI,CAAC,IAAI,CAAC,qBAAqB,EAAE,QAAQ,CAAC,CAAC,CAAC;QAEvF,IAAI,iBAAiB,EAAE;YACnB,MAAM,cAAc,CAAC,sBAAsB,CAAC,CAAC;SAChD;KACJ;IAAC,OAAO,GAAG,EAAE;QACV,IAAI,iBAAiB;YACjB,MAAM,cAAc,CAAC,kBAAkB,CAAC,CAAC;QAE7C,IAAI,IAAI;YACJ,OAAO,CAAC,IAAI,CAAC,IAAI;gBACb,KAAK,CAAC,IAAI,CAAC,mBAAmB,CAAC;gBAC/B,KAAK,CAAC,MAAM,CAAC,qEAAqE,CAAC;gBACnF,qBAAqB,CAAC,IAAI,CAC7B,CAAC;QAEN,MAAM,GAAG,CAAC;KACb;YAAS;QACN,MAAM,mBAAmB,EAAE,CAAC;KAC/B;AACL,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,6BAA6B;IAC/C,MAAM,qBAAqB,GAAG,MAAM,oBAAoB,CAAC,KAAK,CAAC,CAAC;IAEhE,IAAI,qBAAqB,IAAI,IAAI;QAC7B,OAAO,IAAI,CAAC;IAEhB,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,qBAAqB,EAAE,kBAAkB,CAAC,CAAC;IAExE,IAAI,MAAM,EAAE,CAAC,UAAU,CAAC,UAAU,CAAC;QAC/B,OAAO,UAAU,CAAC;IAEtB,OAAO,IAAI,CAAC;AAChB,CAAC;AAID,KAAK,UAAU,oBAAoB,CAAC,iBAA0B,KAAK;IAC/D,IAAI,MAAM,EAAE,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC,EAAE;QACpE,OAAO,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;KACxD;SAAM,IAAI,MAAM,EAAE,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC,EAAE;QACzE,OAAO,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;KACtD;IAED,IAAI,cAAc;QACd,MAAM,IAAI,KAAK,CAAC,2CAA2C,CAAC,CAAC;IAEjE,OAAO,IAAI,CAAC;AAChB,CAAC;AAED,KAAK,UAAU,gBAAgB;IAC3B,IAAI,MAAM,eAAe,EAAE;QACvB,OAAO,EAAE,CAAC;IAEd,MAAM,SAAS,GAAG,MAAM,YAAY,EAAE,CAAC;IAEvC,IAAI,SAAS,IAAI,IAAI;QACjB,OAAO,EAAE,CAAC;IAEd,OAAO,CAAC,cAAc,EAAE,SAAS,CAAC,CAAC;AACvC,CAAC;AAED,KAAK,UAAU,uBAAuB,CAAC,UAAkB;IACrD,IAAI,OAAO,CAAC,IAAI,KAAK,UAAU;QAC3B,OAAO,IAAI,CAAC;IAEhB,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,CAAC;IAClC,MAAM,QAAQ,GAAG,OAAO,CAAC,IAAI,CAAC;IAE9B,MAAM,iBAAiB,GAAG,GAAG,QAAQ,SAAS,QAAQ,WAAW,UAAU,QAAQ,CAAC;IAEpF,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,wBAAwB,EAAE,iBAAiB,CAAC,CAAC;IAExE,IAAI,MAAM,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC;QAC7B,OAAO,QAAQ,CAAC;IAEpB,OAAO,IAAI,CAAC;AAChB,CAAC;AAED,KAAK,UAAU,mBAAmB,CAAC,aAAqB,EAAE,WAAmB;IACzE,MAAM,cAAc,GAAG,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,SAAS,CAAC,CAAC;IAE3D,IAAI,MAAM,EAAE,CAAC,UAAU,CAAC,cAAc,CAAC,EAAE;QACrC,MAAM,EAAE,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC;QAC7B,MAAM,EAAE,CAAC,IAAI,CAAC,cAAc,EAAE,WAAW,CAAC,CAAC;QAE3C,MAAM,SAAS,GAAG,MAAM,EAAE,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC;QAEhD,MAAM,OAAO,CAAC,GAAG,CACb,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CACxB,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,QAAQ,CAAC,EAAE,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,QAAQ,CAAC,EAAE;YAC1E,SAAS,EAAE,IAAI;SAClB,CAAC,CACL,CAAC,CACL,CAAC;QAEF,MAAM,EAAE,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC;KAChC;AACL,CAAC"}
+{"version":3,"file":"compileLLamaCpp.js","sourceRoot":"","sources":["../../src/utils/compileLLamaCpp.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,MAAM,CAAC;AACxB,OAAO,EAAC,aAAa,EAAC,MAAM,KAAK,CAAC;AAClC,OAAO,OAAO,MAAM,SAAS,CAAC;AAC9B,OAAO,EAAE,MAAM,UAAU,CAAC;AAC1B,OAAO,KAAK,MAAM,OAAO,CAAC;AAC1B,OAAO,EACH,8BAA8B,EAAE,qBAAqB,EAAE,iBAAiB,EAAE,cAAc,EAAE,wBAAwB,EACrH,MAAM,cAAc,CAAC;AACtB,OAAO,EAAC,eAAe,EAAC,MAAM,sBAAsB,CAAC;AACrD,OAAO,EAAC,cAAc,EAAC,MAAM,kBAAkB,CAAC;AAChD,OAAO,EAAC,YAAY,EAAC,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAC,mBAAmB,EAAE,YAAY,EAAE,eAAe,EAAC,MAAM,YAAY,CAAC;AAE9E,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAE/D,MAAM,CAAC,KAAK,UAAU,eAAe,CAAC,EAClC,IAAI,GAAG,OAAO,CAAC,IAAI,EAAE,UAAU,GAAG,OAAO,CAAC,OAAO,EAAE,cAAc,EAAE,iBAAiB,GAAG,IAAI,EAAE,KAAK,GAAG,OAAO,CAAC,QAAQ,KAAK,QAAQ,EAClI,IAAI,GAAG,KAAK,EAGf;IACG,IAAI;QACA,IAAI,CAAC,CAAC,MAAM,EAAE,CAAC,UAAU,CAAC,iBAAiB,CAAC,CAAC,EAAE;YAC3C,MAAM,IAAI,KAAK,CAAC,IAAI,iBAAiB,4BAA4B,CAAC,CAAC;SACtE;QAED,MAAM,aAAa,GAAG,MAAM,gBAAgB,EAAE,CAAC;QAC/C,MAAM,aAAa,GAAG,MAAM,uBAAuB,CAAC,IAAI,CAAC,CAAC;QAC1D,MAAM,cAAc,GAAG,UAAU,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,KAAK,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC;QAC9F,MAAM,kBAAkB,GAAG,IAAI,GAAG,EAAkB,CAAC;QAErD,IAAI,CAAC,KAAK,IAAI,OAAO,CAAC,QAAQ,KAAK,QAAQ,CAAC,IAAI,OAAO,CAAC,GAAG,CAAC,UAAU,KAAK,GAAG;YAAE,kBAAkB,CAAC,GAAG,CAAC,YAAY,EAAE,GAAG,CAAC,CAAC;;YACrH,kBAAkB,CAAC,GAAG,CAAC,YAAY,EAAE,KAAK,CAAC,CAAC;QAEjD,IAAI,IAAI,IAAI,OAAO,CAAC,GAAG,CAAC,SAAS,KAAK,GAAG;YAAE,kBAAkB,CAAC,GAAG,CAAC,WAAW,EAAE,GAAG,CAAC,CAAC;QAEpF,IAAI,OAAO,CAAC,GAAG,CAAC,aAAa,KAAK,GAAG;YAAE,kBAAkB,CAAC,GAAG,CAAC,eAAe,EAAE,GAAG,CAAC,CAAC;QACpF,IAAI,OAAO,CAAC,GAAG,CAAC,gBAAgB,IAAI,IAAI;YAAE,kBAAkB,CAAC,GAAG,CAAC,kBAAkB,EAAE,OAAO,CAAC,GAAG,CAAC,gBAAgB,CAAC,CAAC;QACnH,IAAI,OAAO,CAAC,GAAG,CAAC,oBAAoB,IAAI,IAAI;YAAE,kBAAkB,CAAC,GAAG,CAAC,sBAAsB,EAAE,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC,CAAC;QAC/H,IAAI,OAAO,CAAC,GAAG,CAAC,gBAAgB,IAAI,IAAI;YAAE,kBAAkB,CAAC,GAAG,CAAC,kBAAkB,EAAE,OAAO,CAAC,GAAG,CAAC,gBAAgB,CAAC,CAAC;QACnH,IAAI,OAAO,CAAC,GAAG,CAAC,eAAe,IAAI,IAAI;YAAE,kBAAkB,CAAC,GAAG,CAAC,iBAAiB,EAAE,OAAO,CAAC,GAAG,CAAC,eAAe,CAAC,CAAC;QAChH,IAAI,OAAO,CAAC,GAAG,CAAC,aAAa,IAAI,IAAI;YAAE,kBAAkB,CAAC,GAAG,CAAC,eAAe,EAAE,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC;QAC1G,IAAI,OAAO,CAAC,GAAG,CAAC,sBAAsB,IAAI,IAAI;YAAE,kBAAkB,CAAC,GAAG,CAAC,wBAAwB,EAAE,OAAO,CAAC,GAAG,CAAC,sBAAsB,CAAC,CAAC;QACrI,IAAI,OAAO,CAAC,GAAG,CAAC,6BAA6B,IAAI,IAAI;YAAE,kBAAkB,CAAC,GAAG,CAAC,+BAA+B,EAAE,OAAO,CAAC,GAAG,CAAC,6BAA6B,CAAC,CAAC;QAC1J,IAAI,OAAO,CAAC,GAAG,CAAC,YAAY,KAAK,GAAG;YAAE,kBAAkB,CAAC,GAAG,CAAC,cAAc,EAAE,GAAG,CAAC,CAAC;QAElF,IAAI,aAAa,IAAI,IAAI;YACrB,kBAAkB,CAAC,GAAG,CAAC,sBAAsB,EAAE,aAAa,CAAC,CAAC;QAElE,KAAK,MAAM,GAAG,IAAI,OAAO,CAAC,GAAG,EAAE;YAC3B,IAAI,GAAG,CAAC,UAAU,CAAC,8BAA8B,CAAC,EAAE;gBAChD,MAAM,MAAM,GAAG,GAAG,CAAC,KAAK,CAAC,8BAA8B,CAAC,MAAM,CAAC,CAAC;gBAChE,MAAM,KAAK,GAAG,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;gBAC/B,kBAAkB,CAAC,GAAG,CAAC,MAAM,EAAE,KAAM,CAAC,CAAC;aAC1C;SACJ;QAED,MAAM,eAAe,EAAE,CAAC;QAExB,MAAM,YAAY,CAAC,KAAK,EAAE,CAAC,KAAK,EAAE,IAAI,EAAE,gBAAgB,EAAE,IAAI,EAAE,OAAO,EAAE,aAAa,EAAE,MAAM,EAAE,GAAG,aAAa,CAAC,EAAE,SAAS,CAAC,CAAC;QAE9H,MAAM,YAAY,CACd,KAAK,EACL,CAAC,KAAK,EAAE,IAAI,EAAE,gBAAgB,EAAE,IAAI,EAAE,SAAS,EAAE,aAAa,EAAE,MAAM,EAAE,SAAS,GAAG,IAAI,EAAE,oBAAoB,GAAG,cAAc,EAAE,GAAG,aAAa,CAAC;aAC7I,MAAM,CAAC,CAAC,GAAG,kBAAkB,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,MAAM,GAAG,GAAG,GAAG,GAAG,GAAG,KAAK,CAAC,CAAC,EACtF,SAAS,CACZ,CAAC;QAEF,MAAM,gBAAgB,GAAG;YACrB,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,EAAE,KAAK,CAAC;YACzC,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,EAAE,WAAW,EAAE,KAAK,CAAC;SACzD,CAAC;QACF,MAAM,qBAAqB,GAAG,MAAM,oBAAoB,CAAC,IAAI,CAAC,CAAC;QAE/D,KAAK,MAAM,eAAe,IAAI,gBAAgB,EAAE;YAC5C,IAAI,MAAM,EAAE,CAAC,UAAU,CAAC,eAAe,CAAC,EAAE;gBACtC,MAAM,KAAK,GAAG,MAAM,EAAE,CAAC,OAAO,CAAC,eAAe,CAAC,CAAC;gBAEhD,MAAM,OAAO,CAAC,GAAG,CACb,KAAK,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CACpB,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,eAAe,EAAE,QAAQ,CAAC,EAAE,IAAI,CAAC,IAAI,CAAC,qBAAqB,EAAE,QAAQ,CAAC,EAAE;oBACtF,SAAS,EAAE,KAAK;iBACnB,CAAC,CACL,CAAC,CACL,CAAC;aACL;SACJ;QAED,mBAAmB,CAAC,qBAAqB,EAAE,IAAI,CAAC,IAAI,CAAC,qBAAqB,EAAE,QAAQ,CAAC,CAAC,CAAC;QAEvF,IAAI,iBAAiB,EAAE;YACnB,MAAM,cAAc,CAAC,sBAAsB,CAAC,CAAC;SAChD;KACJ;IAAC,OAAO,GAAG,EAAE;QACV,IAAI,iBAAiB;YACjB,MAAM,cAAc,CAAC,kBAAkB,CAAC,CAAC;QAE7C,IAAI,IAAI;YACJ,OAAO,CAAC,IAAI,CAAC,IAAI;gBACb,KAAK,CAAC,IAAI,CAAC,mBAAmB,CAAC;gBAC/B,KAAK,CAAC,MAAM,CAAC,qEAAqE,CAAC;gBACnF,qBAAqB,CAAC,IAAI,CAC7B,CAAC;QAEN,MAAM,GAAG,CAAC;KACb;YAAS;QACN,MAAM,mBAAmB,EAAE,CAAC;KAC/B;AACL,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,6BAA6B;IAC/C,MAAM,qBAAqB,GAAG,MAAM,oBAAoB,CAAC,KAAK,CAAC,CAAC;IAEhE,IAAI,qBAAqB,IAAI,IAAI;QAC7B,OAAO,IAAI,CAAC;IAEhB,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,qBAAqB,EAAE,kBAAkB,CAAC,CAAC;IAExE,IAAI,MAAM,EAAE,CAAC,UAAU,CAAC,UAAU,CAAC;QAC/B,OAAO,UAAU,CAAC;IAEtB,OAAO,IAAI,CAAC;AAChB,CAAC;AAID,KAAK,UAAU,oBAAoB,CAAC,iBAA0B,KAAK;IAC/D,IAAI,MAAM,EAAE,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC,EAAE;QACpE,OAAO,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;KACxD;SAAM,IAAI,MAAM,EAAE,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC,EAAE;QACzE,OAAO,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;KACtD;IAED,IAAI,cAAc;QACd,MAAM,IAAI,KAAK,CAAC,2CAA2C,CAAC,CAAC;IAEjE,OAAO,IAAI,CAAC;AAChB,CAAC;AAED,KAAK,UAAU,gBAAgB;IAC3B,IAAI,MAAM,eAAe,EAAE;QACvB,OAAO,EAAE,CAAC;IAEd,MAAM,SAAS,GAAG,MAAM,YAAY,EAAE,CAAC;IAEvC,IAAI,SAAS,IAAI,IAAI;QACjB,OAAO,EAAE,CAAC;IAEd,OAAO,CAAC,cAAc,EAAE,SAAS,CAAC,CAAC;AACvC,CAAC;AAED,KAAK,UAAU,uBAAuB,CAAC,UAAkB;IACrD,IAAI,OAAO,CAAC,IAAI,KAAK,UAAU;QAC3B,OAAO,IAAI,CAAC;IAEhB,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,CAAC;IAClC,MAAM,QAAQ,GAAG,OAAO,CAAC,IAAI,CAAC;IAE9B,MAAM,iBAAiB,GAAG,GAAG,QAAQ,SAAS,QAAQ,WAAW,UAAU,QAAQ,CAAC;IAEpF,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,wBAAwB,EAAE,iBAAiB,CAAC,CAAC;IAExE,IAAI,MAAM,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC;QAC7B,OAAO,QAAQ,CAAC;IAEpB,OAAO,IAAI,CAAC;AAChB,CAAC;AAED,KAAK,UAAU,mBAAmB,CAAC,aAAqB,EAAE,WAAmB;IACzE,MAAM,cAAc,GAAG,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,SAAS,CAAC,CAAC;IAE3D,IAAI,MAAM,EAAE,CAAC,UAAU,CAAC,cAAc,CAAC,EAAE;QACrC,MAAM,EAAE,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC;QAC7B,MAAM,EAAE,CAAC,IAAI,CAAC,cAAc,EAAE,WAAW,CAAC,CAAC;QAE3C,MAAM,SAAS,GAAG,MAAM,EAAE,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC;QAEhD,MAAM,OAAO,CAAC,GAAG,CACb,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CACxB,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,QAAQ,CAAC,EAAE,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,QAAQ,CAAC,EAAE;YAC1E,SAAS,EAAE,IAAI;SAClB,CAAC,CACL,CAAC,CACL,CAAC;QAEF,MAAM,EAAE,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC;KAChC;AACL,CAAC"}

package/llama/addon.cpp

@@ -448,3 +448,4 @@ Napi::Object registerCallback(Napi::Env env, Napi::Object exports) {
 }
 
 NODE_API_MODULE(NODE_GYP_MODULE_NAME, registerCallback)
+

package/llama/binariesGithubRelease.json

@@ -1,3 +1,3 @@
 {
-    "release": "b3465"
+    "release": "b3658"
 }

package/llamaBins/mac-arm64/ggml-common.h

@@ -19,7 +19,11 @@ typedef half2 ggml_half2;
 
 #define GGML_COMMON_DECL
 #elif defined(GGML_COMMON_DECL_CUDA)
+#if defined(GGML_COMMON_DECL_MUSA)
+#include <musa_fp16.h>
+#else
 #include <cuda_fp16.h>
+#endif
 #include <cstdint>
 
 typedef half  ggml_half;
@@ -415,7 +419,7 @@ static_assert(sizeof(block_iq4_xs) == sizeof(ggml_half) + sizeof(uint16_t) + QK_
 #define GGML_TABLE_END() };
 
 #define GGML_COMMON_IMPL
-#elif defined(GGML_COMMON_IMPL_CUDA) || defined(GGML_COMMON_IMPL_HIP)
+#elif defined(GGML_COMMON_IMPL_CUDA) || defined(GGML_COMMON_IMPL_HIP) || defined(GGML_COMMON_IMPL_MUSA)
 #include <cstdint>
 
 #define GGML_TABLE_BEGIN(type, name, size) static const __device__ type name[size] = {

package/llamaBins/mac-arm64/ggml-metal.metal

@@ -17,7 +17,7 @@ enum ggml_sort_order {
     GGML_SORT_ORDER_DESC,
 };
 
-// general-purpose kernel for addition, multiplication and division of two tensors
+// general-purpose kernel for addition, subtraction, multiplication and division of two tensors
 // pros: works for non-contiguous tensors, supports broadcast across all dims
 // cons: not very efficient
 kernel void kernel_add(
@@ -70,6 +70,56 @@ kernel void kernel_add(
     }
 }
 
+kernel void kernel_sub(
+        device const char * src0,
+        device const char * src1,
+        device       char * dst,
+        constant  int64_t & ne00,
+        constant  int64_t & ne01,
+        constant  int64_t & ne02,
+        constant  int64_t & ne03,
+        constant uint64_t & nb00,
+        constant uint64_t & nb01,
+        constant uint64_t & nb02,
+        constant uint64_t & nb03,
+        constant  int64_t & ne10,
+        constant  int64_t & ne11,
+        constant  int64_t & ne12,
+        constant  int64_t & ne13,
+        constant uint64_t & nb10,
+        constant uint64_t & nb11,
+        constant uint64_t & nb12,
+        constant uint64_t & nb13,
+        constant  int64_t & ne0,
+        constant  int64_t & ne1,
+        constant  int64_t & ne2,
+        constant  int64_t & ne3,
+        constant uint64_t & nb0,
+        constant uint64_t & nb1,
+        constant uint64_t & nb2,
+        constant uint64_t & nb3,
+        constant  int64_t & offs,
+        uint3 tgpig[[threadgroup_position_in_grid]],
+        uint3 tpitg[[thread_position_in_threadgroup]],
+        uint3   ntg[[threads_per_threadgroup]]) {
+    const int64_t i03 = tgpig.z;
+    const int64_t i02 = tgpig.y;
+    const int64_t i01 = tgpig.x;
+
+    const int64_t i13 = i03 % ne13;
+    const int64_t i12 = i02 % ne12;
+    const int64_t i11 = i01 % ne11;
+
+    device const char * src0_ptr = src0 + i03*nb03 + i02*nb02 + i01*nb01 + offs;
+    device const char * src1_ptr = src1 + i13*nb13 + i12*nb12 + i11*nb11;
+    device       char * dst_ptr  = dst  + i03*nb3  + i02*nb2  + i01*nb1  + offs;
+
+    for (int i0 = tpitg.x; i0 < ne0; i0 += ntg.x) {
+        const int i10 = i0 % ne10;
+        *((device float *)(dst_ptr + i0*nb0)) = *((device float *)(src0_ptr + i0*nb00)) - *((device float *)(src1_ptr + i10*nb10));
+    }
+}
+
 kernel void kernel_mul(
         device const char * src0,
         device const char * src1,
@@ -226,6 +276,15 @@ kernel void kernel_add_row(
     dst[tpig] = src0[tpig] + src1[tpig % nb];
 }
 
+kernel void kernel_sub_row(
+        device const float4 * src0,
+        device const float4 * src1,
+        device       float4 * dst,
+        constant   uint64_t & nb [[buffer(28)]],
+        uint tpig[[thread_position_in_grid]]) {
+    dst[tpig] = src0[tpig] - src1[tpig % nb];
+}
+
 kernel void kernel_mul_row(
         device const float4 * src0,
         device const float4 * src1,
@@ -358,6 +417,27 @@ kernel void kernel_sqr(
     dst[tpig] = src0[tpig] * src0[tpig];
 }
 
+kernel void kernel_sqrt(
+        device const float * src0,
+        device       float * dst,
+        uint tpig[[thread_position_in_grid]]) {
+    dst[tpig] = sqrt(src0[tpig]);
+}
+
+kernel void kernel_sin(
+        device const float * src0,
+        device       float * dst,
+        uint tpig[[thread_position_in_grid]]) {
+    dst[tpig] = sin(src0[tpig]);
+}
+
+kernel void kernel_cos(
+        device const float * src0,
+        device       float * dst,
+        uint tpig[[thread_position_in_grid]]) {
+    dst[tpig] = cos(src0[tpig]);
+}
+
 kernel void kernel_sum_rows(
         device const float * src0,
         device       float * dst,
@@ -667,6 +747,127 @@ kernel void kernel_diag_mask_inf_8(
     }
 }
 
+// ref: ggml.c:ggml_compute_forward_ssm_conv_f32
+// TODO: optimize
+kernel void kernel_ssm_conv_f32(
+        device const  void * src0,
+        device const  void * src1,
+        device       float * dst,
+        constant   int64_t & ne00,
+        constant   int64_t & ne01,
+        constant   int64_t & ne02,
+        constant  uint64_t & nb00,
+        constant  uint64_t & nb01,
+        constant  uint64_t & nb02,
+        constant   int64_t & ne10,
+        constant   int64_t & ne11,
+        constant  uint64_t & nb10,
+        constant  uint64_t & nb11,
+        constant   int64_t & ne0,
+        constant   int64_t & ne1,
+        constant   int64_t & ne2,
+        constant  uint64_t & nb0,
+        constant  uint64_t & nb1,
+        constant  uint64_t & nb2,
+        uint3 tgpig[[threadgroup_position_in_grid]],
+        uint3 tpitg[[thread_position_in_threadgroup]],
+        uint3   ntg[[threads_per_threadgroup]]) {
+    const int64_t ir = tgpig.x;
+    const int64_t i2 = tgpig.y;
+    const int64_t i3 = tgpig.z;
+
+    const int64_t nc  = ne10;
+    const int64_t ncs = ne00;
+    const int64_t nr  = ne01;
+    const int64_t n_t = ne1;
+    const int64_t n_s = ne2;
+
+    device const float * s = (device const float *) ((device const char *) src0 + ir*nb01 + i2*nb00 + i3*nb02);
+    device const float * c = (device const float *) ((device const char *) src1 + ir*nb11);
+    device       float * x = (device       float *) ((device       char *) dst  + ir*nb0  + i2*nb1  + i3*nb2);
+
+    float sumf = 0.0f;
+
+    for (int64_t i0 = 0; i0 < nc; ++i0) {
+        sumf += s[i0] * c[i0];
+    }
+
+    x[0] = sumf;
+}
+
+// ref: ggml.c:ggml_compute_forward_ssm_scan_f32
+// TODO: optimize
+kernel void kernel_ssm_scan_f32(
+        device const void * src0,
+        device const void * src1,
+        device const void * src2,
+        device const void * src3,
+        device const void * src4,
+        device const void * src5,
+        device      float * dst,
+        constant  int64_t & d_state,
+        constant  int64_t & d_inner,
+        constant  int64_t & n_seq_tokens,
+        constant  int64_t & n_seqs,
+        constant uint64_t & nb00,
+        constant uint64_t & nb01,
+        constant uint64_t & nb02,
+        constant uint64_t & nb10,
+        constant uint64_t & nb11,
+        constant uint64_t & nb12,
+        constant uint64_t & nb13,
+        constant uint64_t & nb20,
+        constant uint64_t & nb21,
+        constant uint64_t & nb22,
+        constant uint64_t & nb30,
+        constant uint64_t & nb31,
+        constant uint64_t & nb40,
+        constant uint64_t & nb41,
+        constant uint64_t & nb42,
+        constant uint64_t & nb50,
+        constant uint64_t & nb51,
+        constant uint64_t & nb52,
+        uint3 tgpig[[threadgroup_position_in_grid]],
+        uint3 tpitg[[thread_position_in_threadgroup]],
+        uint3   ntg[[threads_per_threadgroup]]) {
+    const int64_t ir = tgpig.x;
+    const int64_t i3 = tgpig.y;
+
+    const int64_t nc  = d_state;
+    const int64_t nr  = d_inner;
+    const int64_t n_t = n_seq_tokens;
+    const int64_t n_s = n_seqs;
+
+    for (int64_t i2 = 0; i2 < n_t; ++i2) {
+        device const float * s0 = (device const float *) ((device const char *) src0 + ir*nb01 + i3*nb02);
+        device const float * x  = (device const float *) ((device const char *) src1 + ir*nb10 + i2*nb11 + i3*nb12);
+        device const float * dt = (device const float *) ((device const char *) src2 + ir*nb20 + i2*nb21 + i3*nb22);
+        device const float * A  = (device const float *) ((device const char *) src3 + ir*nb31);
+        device const float * B  = (device const float *) ((device const char *) src4 + i2*nb41 + i3*nb42);
+        device const float * C  = (device const float *) ((device const char *) src5 + i2*nb51 + i3*nb52);
+        device       float * y  = (device       float *) ((device       char *) dst  + ir*nb10 + i2*nb11 + i3*nb12); // TODO: do not use src1 strides
+        device       float * s  = (device       float *) ((device       char *) dst  + ir*nb01 + i3*nb02 +    nb13);
+
+        if (i2 > 0) {
+            s0 = s;
+        }
+
+        // i1 == 0
+        float dt_soft_plus = dt[0] <= 20.0f ? log(1.0f + exp(dt[0])) : dt[0];
+        float x_dt = x[0] * dt_soft_plus;
+        float sumf = 0.0f;
+
+        for (int64_t i0 = 0; i0 < nc; ++i0) {
+            int64_t i = i0;
+            float state = (s0[i] * exp(dt_soft_plus * A[i])) + (B[i0] * x_dt);
+            sumf += state * C[i0];
+            s[i] = state;
+        }
+
+        y[0] = sumf;
+    }
+}
+
 kernel void kernel_norm(
         device const  void * src0,
         device       float * dst,
@@ -1976,6 +2177,7 @@ typedef void (flash_attn_ext_f16_t)(
         constant     float & m0,
         constant     float & m1,
         constant  uint32_t & n_head_log2,
+        constant     float & logit_softcap,
         threadgroup   half * shared,
         uint3  tgpig[[threadgroup_position_in_grid]],
         uint3  tpitg[[thread_position_in_threadgroup]],
@@ -2014,6 +2216,7 @@ kernel void kernel_flash_attn_ext_f16(
         constant     float & m0,
         constant     float & m1,
         constant  uint32_t & n_head_log2,
+        constant     float & logit_softcap,
         threadgroup   half * shared [[threadgroup(0)]],
         uint3  tgpig[[threadgroup_position_in_grid]],
         uint3  tpitg[[thread_position_in_threadgroup]],
@@ -2138,19 +2341,6 @@ kernel void kernel_flash_attn_ext_f16(
                     }
 
                     simdgroup_store(mqk, ss + 8*cc, TF, 0, false);
-
-                    const short tx = tiisg%4;
-                    const short ty = tiisg/4;
-
-                    if (mask != q) {
-                        // mqk = mqk*scale + mask*slope
-                        ss[8*cc + ty*TF + 2*tx + 0] = scale*ss[8*cc + ty*TF + 2*tx + 0] + slope*mp[ic + 8*cc + ty*nb31/sizeof(half) + 2*tx + 0];
-                        ss[8*cc + ty*TF + 2*tx + 1] = scale*ss[8*cc + ty*TF + 2*tx + 1] + slope*mp[ic + 8*cc + ty*nb31/sizeof(half) + 2*tx + 1];
-                    } else {
-                        // mqk = mqk*scale
-                        ss[8*cc + ty*TF + 2*tx + 0] *= scale;
-                        ss[8*cc + ty*TF + 2*tx + 1] *= scale;
-                    }
                 }
             }
 
@@ -2162,10 +2352,19 @@ kernel void kernel_flash_attn_ext_f16(
                 float ms[Q];
 
                 for (short j = 0; j < Q; ++j) {
-                    const short p = tiisg;
-
                     const float m = M[j];
-                    const float s = ss[j*TF + p];
+
+                    // scale and apply the logitcap / mask
+                    float s = ss[j*TF + tiisg]*scale;
+
+                    if (logit_softcap != 0.0f) {
+                        s = logit_softcap*precise::tanh(s);
+                    }
+
+                    if (mask != q) {
+                        // mqk = mqk + mask*slope
+                        s += slope*mp[ic + j*nb31/sizeof(half) + tiisg];
+                    }
 
                     smax = simd_max(max(smax, s));
                     M[j] = simd_max(max(M[j], s));
@@ -2176,7 +2375,7 @@ kernel void kernel_flash_attn_ext_f16(
                     S[j] = S[j]*ms[j] + simd_sum(vs);
 
                     // the P matrix from the paper (Q rows, C columns)
-                    ss[j*TF + p] = vs;
+                    ss[j*TF + tiisg] = vs;
                 }
 
                 // create a QxQ diagonal matrix for rescaling the output
@@ -2345,6 +2544,7 @@ kernel void kernel_flash_attn_ext_vec_f16(
         constant     float & m0,
         constant     float & m1,
         constant  uint32_t & n_head_log2,
+        constant     float & logit_softcap,
         threadgroup   half * shared [[threadgroup(0)]],
         uint3  tgpig[[threadgroup_position_in_grid]],
         uint3  tpitg[[thread_position_in_threadgroup]],
@@ -2479,7 +2679,13 @@ kernel void kernel_flash_attn_ext_vec_f16(
 
                     // mqk = mqk*scale + mask*slope
                     if (tiisg == 0) {
-                        mqk = mqk*scale + ((mask != q) ? ((float4) mp4[ic/4 + cc])*slope : (float4) 0.0f);
+                        mqk *= scale;
+
+                        if (logit_softcap != 0.0f) {
+                            mqk = logit_softcap*precise::tanh(mqk);
+                        }
+
+                        mqk += (mask != q) ? ((float4) mp4[ic/4 + cc])*slope : (float4) 0.0f;
 
                         ss4[cc] = mqk;
                     }

package/llamaBins/mac-x64/ggml-common.h

@@ -19,7 +19,11 @@ typedef half2 ggml_half2;
 
 #define GGML_COMMON_DECL
 #elif defined(GGML_COMMON_DECL_CUDA)
+#if defined(GGML_COMMON_DECL_MUSA)
+#include <musa_fp16.h>
+#else
 #include <cuda_fp16.h>
+#endif
 #include <cstdint>
 
 typedef half  ggml_half;
@@ -415,7 +419,7 @@ static_assert(sizeof(block_iq4_xs) == sizeof(ggml_half) + sizeof(uint16_t) + QK_
 #define GGML_TABLE_END() };
 
 #define GGML_COMMON_IMPL
-#elif defined(GGML_COMMON_IMPL_CUDA) || defined(GGML_COMMON_IMPL_HIP)
+#elif defined(GGML_COMMON_IMPL_CUDA) || defined(GGML_COMMON_IMPL_HIP) || defined(GGML_COMMON_IMPL_MUSA)
 #include <cstdint>
 
 #define GGML_TABLE_BEGIN(type, name, size) static const __device__ type name[size] = {

package/llamaBins/mac-x64/ggml-metal.metal

@@ -17,7 +17,7 @@ enum ggml_sort_order {
     GGML_SORT_ORDER_DESC,
 };
 
-// general-purpose kernel for addition, multiplication and division of two tensors
+// general-purpose kernel for addition, subtraction, multiplication and division of two tensors
 // pros: works for non-contiguous tensors, supports broadcast across all dims
 // cons: not very efficient
 kernel void kernel_add(
@@ -70,6 +70,56 @@ kernel void kernel_add(
     }
 }
 
+kernel void kernel_sub(
+        device const char * src0,
+        device const char * src1,
+        device       char * dst,
+        constant  int64_t & ne00,
+        constant  int64_t & ne01,
+        constant  int64_t & ne02,
+        constant  int64_t & ne03,
+        constant uint64_t & nb00,
+        constant uint64_t & nb01,
+        constant uint64_t & nb02,
+        constant uint64_t & nb03,
+        constant  int64_t & ne10,
+        constant  int64_t & ne11,
+        constant  int64_t & ne12,
+        constant  int64_t & ne13,
+        constant uint64_t & nb10,
+        constant uint64_t & nb11,
+        constant uint64_t & nb12,
+        constant uint64_t & nb13,
+        constant  int64_t & ne0,
+        constant  int64_t & ne1,
+        constant  int64_t & ne2,
+        constant  int64_t & ne3,
+        constant uint64_t & nb0,
+        constant uint64_t & nb1,
+        constant uint64_t & nb2,
+        constant uint64_t & nb3,
+        constant  int64_t & offs,
+        uint3 tgpig[[threadgroup_position_in_grid]],
+        uint3 tpitg[[thread_position_in_threadgroup]],
+        uint3   ntg[[threads_per_threadgroup]]) {
+    const int64_t i03 = tgpig.z;
+    const int64_t i02 = tgpig.y;
+    const int64_t i01 = tgpig.x;
+
+    const int64_t i13 = i03 % ne13;
+    const int64_t i12 = i02 % ne12;
+    const int64_t i11 = i01 % ne11;
+
+    device const char * src0_ptr = src0 + i03*nb03 + i02*nb02 + i01*nb01 + offs;
+    device const char * src1_ptr = src1 + i13*nb13 + i12*nb12 + i11*nb11;
+    device       char * dst_ptr  = dst  + i03*nb3  + i02*nb2  + i01*nb1  + offs;
+
+    for (int i0 = tpitg.x; i0 < ne0; i0 += ntg.x) {
+        const int i10 = i0 % ne10;
+        *((device float *)(dst_ptr + i0*nb0)) = *((device float *)(src0_ptr + i0*nb00)) - *((device float *)(src1_ptr + i10*nb10));
+    }
+}
+
 kernel void kernel_mul(
         device const char * src0,
         device const char * src1,
@@ -226,6 +276,15 @@ kernel void kernel_add_row(
     dst[tpig] = src0[tpig] + src1[tpig % nb];
 }
 
+kernel void kernel_sub_row(
+        device const float4 * src0,
+        device const float4 * src1,
+        device       float4 * dst,
+        constant   uint64_t & nb [[buffer(28)]],
+        uint tpig[[thread_position_in_grid]]) {
+    dst[tpig] = src0[tpig] - src1[tpig % nb];
+}
+
 kernel void kernel_mul_row(
         device const float4 * src0,
         device const float4 * src1,
@@ -358,6 +417,27 @@ kernel void kernel_sqr(
     dst[tpig] = src0[tpig] * src0[tpig];
 }
 
+kernel void kernel_sqrt(
+        device const float * src0,
+        device       float * dst,
+        uint tpig[[thread_position_in_grid]]) {
+    dst[tpig] = sqrt(src0[tpig]);
+}
+
+kernel void kernel_sin(
+        device const float * src0,
+        device       float * dst,
+        uint tpig[[thread_position_in_grid]]) {
+    dst[tpig] = sin(src0[tpig]);
+}
+
+kernel void kernel_cos(
+        device const float * src0,
+        device       float * dst,
+        uint tpig[[thread_position_in_grid]]) {
+    dst[tpig] = cos(src0[tpig]);
+}
+
 kernel void kernel_sum_rows(
         device const float * src0,
         device       float * dst,
@@ -667,6 +747,127 @@ kernel void kernel_diag_mask_inf_8(
     }
 }
 
+// ref: ggml.c:ggml_compute_forward_ssm_conv_f32
+// TODO: optimize
+kernel void kernel_ssm_conv_f32(
+        device const  void * src0,
+        device const  void * src1,
+        device       float * dst,
+        constant   int64_t & ne00,
+        constant   int64_t & ne01,
+        constant   int64_t & ne02,
+        constant  uint64_t & nb00,
+        constant  uint64_t & nb01,
+        constant  uint64_t & nb02,
+        constant   int64_t & ne10,
+        constant   int64_t & ne11,
+        constant  uint64_t & nb10,
+        constant  uint64_t & nb11,
+        constant   int64_t & ne0,
+        constant   int64_t & ne1,
+        constant   int64_t & ne2,
+        constant  uint64_t & nb0,
+        constant  uint64_t & nb1,
+        constant  uint64_t & nb2,
+        uint3 tgpig[[threadgroup_position_in_grid]],
+        uint3 tpitg[[thread_position_in_threadgroup]],
+        uint3   ntg[[threads_per_threadgroup]]) {
+    const int64_t ir = tgpig.x;
+    const int64_t i2 = tgpig.y;
+    const int64_t i3 = tgpig.z;
+
+    const int64_t nc  = ne10;
+    const int64_t ncs = ne00;
+    const int64_t nr  = ne01;
+    const int64_t n_t = ne1;
+    const int64_t n_s = ne2;
+
+    device const float * s = (device const float *) ((device const char *) src0 + ir*nb01 + i2*nb00 + i3*nb02);
+    device const float * c = (device const float *) ((device const char *) src1 + ir*nb11);
+    device       float * x = (device       float *) ((device       char *) dst  + ir*nb0  + i2*nb1  + i3*nb2);
+
+    float sumf = 0.0f;
+
+    for (int64_t i0 = 0; i0 < nc; ++i0) {
+        sumf += s[i0] * c[i0];
+    }
+
+    x[0] = sumf;
+}
+
+// ref: ggml.c:ggml_compute_forward_ssm_scan_f32
+// TODO: optimize
+kernel void kernel_ssm_scan_f32(
+        device const void * src0,
+        device const void * src1,
+        device const void * src2,
+        device const void * src3,
+        device const void * src4,
+        device const void * src5,
+        device      float * dst,
+        constant  int64_t & d_state,
+        constant  int64_t & d_inner,
+        constant  int64_t & n_seq_tokens,
+        constant  int64_t & n_seqs,
+        constant uint64_t & nb00,
+        constant uint64_t & nb01,
+        constant uint64_t & nb02,
+        constant uint64_t & nb10,
+        constant uint64_t & nb11,
+        constant uint64_t & nb12,
+        constant uint64_t & nb13,
+        constant uint64_t & nb20,
+        constant uint64_t & nb21,
+        constant uint64_t & nb22,
+        constant uint64_t & nb30,
+        constant uint64_t & nb31,
+        constant uint64_t & nb40,
+        constant uint64_t & nb41,
+        constant uint64_t & nb42,
+        constant uint64_t & nb50,
+        constant uint64_t & nb51,
+        constant uint64_t & nb52,
+        uint3 tgpig[[threadgroup_position_in_grid]],
+        uint3 tpitg[[thread_position_in_threadgroup]],
+        uint3   ntg[[threads_per_threadgroup]]) {
+    const int64_t ir = tgpig.x;
+    const int64_t i3 = tgpig.y;
+
+    const int64_t nc  = d_state;
+    const int64_t nr  = d_inner;
+    const int64_t n_t = n_seq_tokens;
+    const int64_t n_s = n_seqs;
+
+    for (int64_t i2 = 0; i2 < n_t; ++i2) {
+        device const float * s0 = (device const float *) ((device const char *) src0 + ir*nb01 + i3*nb02);
+        device const float * x  = (device const float *) ((device const char *) src1 + ir*nb10 + i2*nb11 + i3*nb12);
+        device const float * dt = (device const float *) ((device const char *) src2 + ir*nb20 + i2*nb21 + i3*nb22);
+        device const float * A  = (device const float *) ((device const char *) src3 + ir*nb31);
+        device const float * B  = (device const float *) ((device const char *) src4 + i2*nb41 + i3*nb42);
+        device const float * C  = (device const float *) ((device const char *) src5 + i2*nb51 + i3*nb52);
+        device       float * y  = (device       float *) ((device       char *) dst  + ir*nb10 + i2*nb11 + i3*nb12); // TODO: do not use src1 strides
+        device       float * s  = (device       float *) ((device       char *) dst  + ir*nb01 + i3*nb02 +    nb13);
+
+        if (i2 > 0) {
+            s0 = s;
+        }
+
+        // i1 == 0
+        float dt_soft_plus = dt[0] <= 20.0f ? log(1.0f + exp(dt[0])) : dt[0];
+        float x_dt = x[0] * dt_soft_plus;
+        float sumf = 0.0f;
+
+        for (int64_t i0 = 0; i0 < nc; ++i0) {
+            int64_t i = i0;
+            float state = (s0[i] * exp(dt_soft_plus * A[i])) + (B[i0] * x_dt);
+            sumf += state * C[i0];
+            s[i] = state;
+        }
+
+        y[0] = sumf;
+    }
+}
+
 kernel void kernel_norm(
         device const  void * src0,
         device       float * dst,
@@ -1976,6 +2177,7 @@ typedef void (flash_attn_ext_f16_t)(
         constant     float & m0,
         constant     float & m1,
         constant  uint32_t & n_head_log2,
+        constant     float & logit_softcap,
         threadgroup   half * shared,
         uint3  tgpig[[threadgroup_position_in_grid]],
         uint3  tpitg[[thread_position_in_threadgroup]],
@@ -2014,6 +2216,7 @@ kernel void kernel_flash_attn_ext_f16(
         constant     float & m0,
         constant     float & m1,
         constant  uint32_t & n_head_log2,
+        constant     float & logit_softcap,
         threadgroup   half * shared [[threadgroup(0)]],
         uint3  tgpig[[threadgroup_position_in_grid]],
         uint3  tpitg[[thread_position_in_threadgroup]],
@@ -2138,19 +2341,6 @@ kernel void kernel_flash_attn_ext_f16(
                     }
 
                     simdgroup_store(mqk, ss + 8*cc, TF, 0, false);
-
-                    const short tx = tiisg%4;
-                    const short ty = tiisg/4;
-
-                    if (mask != q) {
-                        // mqk = mqk*scale + mask*slope
-                        ss[8*cc + ty*TF + 2*tx + 0] = scale*ss[8*cc + ty*TF + 2*tx + 0] + slope*mp[ic + 8*cc + ty*nb31/sizeof(half) + 2*tx + 0];
-                        ss[8*cc + ty*TF + 2*tx + 1] = scale*ss[8*cc + ty*TF + 2*tx + 1] + slope*mp[ic + 8*cc + ty*nb31/sizeof(half) + 2*tx + 1];
-                    } else {
-                        // mqk = mqk*scale
-                        ss[8*cc + ty*TF + 2*tx + 0] *= scale;
-                        ss[8*cc + ty*TF + 2*tx + 1] *= scale;
-                    }
                 }
             }
 
@@ -2162,10 +2352,19 @@ kernel void kernel_flash_attn_ext_f16(
                 float ms[Q];
 
                 for (short j = 0; j < Q; ++j) {
-                    const short p = tiisg;
-
                     const float m = M[j];
-                    const float s = ss[j*TF + p];
+
+                    // scale and apply the logitcap / mask
+                    float s = ss[j*TF + tiisg]*scale;
+
+                    if (logit_softcap != 0.0f) {
+                        s = logit_softcap*precise::tanh(s);
+                    }
+
+                    if (mask != q) {
+                        // mqk = mqk + mask*slope
+                        s += slope*mp[ic + j*nb31/sizeof(half) + tiisg];
+                    }
 
                     smax = simd_max(max(smax, s));
                     M[j] = simd_max(max(M[j], s));
@@ -2176,7 +2375,7 @@ kernel void kernel_flash_attn_ext_f16(
                     S[j] = S[j]*ms[j] + simd_sum(vs);
 
                     // the P matrix from the paper (Q rows, C columns)
-                    ss[j*TF + p] = vs;
+                    ss[j*TF + tiisg] = vs;
                 }
 
                 // create a QxQ diagonal matrix for rescaling the output
@@ -2345,6 +2544,7 @@ kernel void kernel_flash_attn_ext_vec_f16(
         constant     float & m0,
         constant     float & m1,
         constant  uint32_t & n_head_log2,
+        constant     float & logit_softcap,
         threadgroup   half * shared [[threadgroup(0)]],
         uint3  tgpig[[threadgroup_position_in_grid]],
         uint3  tpitg[[thread_position_in_threadgroup]],
@@ -2479,7 +2679,13 @@ kernel void kernel_flash_attn_ext_vec_f16(
 
                     // mqk = mqk*scale + mask*slope
                     if (tiisg == 0) {
-                        mqk = mqk*scale + ((mask != q) ? ((float4) mp4[ic/4 + cc])*slope : (float4) 0.0f);
+                        mqk *= scale;
+
+                        if (logit_softcap != 0.0f) {
+                            mqk = logit_softcap*precise::tanh(mqk);
+                        }
+
+                        mqk += (mask != q) ? ((float4) mp4[ic/4 + cc])*slope : (float4) 0.0f;
 
                         ss4[cc] = mqk;
                     }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "node-llama-cpp",
-  "version": "2.8.14",
+  "version": "2.8.16",
   "description": "Run AI models locally on your machine with node.js bindings for llama.cpp. Force a JSON schema on the model output on the generation level",
   "main": "dist/index.js",
   "type": "module",
@@ -97,7 +97,7 @@
   "bugs": {
     "url": "https://github.com/withcatai/node-llama-cpp/issues"
   },
-  "homepage": "https://withcatai.github.io/node-llama-cpp/",
+  "homepage": "https://node-llama-cpp.withcat.ai",
   "devDependencies": {
     "@commitlint/cli": "^17.7.1",
     "@commitlint/config-conventional": "^17.7.0",