Try and fix build on mac.

gpt4all-backend/CMakeLists.txt

@@ -76,7 +76,7 @@ foreach(BUILD_VARIANT IN LISTS BUILD_VARIANTS)
 
     # Add each individual implementations
     add_library(llamamodel-mainline-${BUILD_VARIANT} SHARED
-        llamamodel.cpp)
+        llamamodel.cpp llmodel_shared.cpp)
     target_compile_definitions(llamamodel-mainline-${BUILD_VARIANT} PRIVATE
         LLAMA_VERSIONS=>=3 LLAMA_DATE=999999)
     prepare_target(llamamodel-mainline llama-mainline)
@@ -94,11 +94,11 @@ foreach(BUILD_VARIANT IN LISTS BUILD_VARIANTS)
     prepare_target(llamamodel-230511 llama-230511)
 
     add_library(gptj-${BUILD_VARIANT} SHARED
-        gptj.cpp utils.h utils.cpp)
+        gptj.cpp utils.h utils.cpp llmodel_shared.cpp)
     prepare_target(gptj ggml-230511)
 
     add_library(mpt-${BUILD_VARIANT} SHARED
-        mpt.cpp utils.h utils.cpp)
+        mpt.cpp utils.h utils.cpp llmodel_shared.cpp)
     prepare_target(mpt ggml-230511)
 endforeach()
 

gpt4all-backend/llmodel.cpp

@@ -96,28 +96,6 @@ const LLModel::Implementation* LLModel::implementation(std::ifstream& f, const s
     return nullptr;
 }
 
-void LLModel::recalculateContext(PromptContext &promptCtx, std::function<bool(bool)> recalculate) {
-    size_t i = 0;
-    promptCtx.n_past = 0;
-    while (i < promptCtx.tokens.size()) {
-        size_t batch_end = std::min(i + promptCtx.n_batch, promptCtx.tokens.size());
-        std::vector<int32_t> batch(promptCtx.tokens.begin() + i, promptCtx.tokens.begin() + batch_end);
-        assert(promptCtx.n_past + int32_t(batch.size()) <= promptCtx.n_ctx);
-        if (!evalTokens(promptCtx, batch)) {
-            std::cerr << "LLModel ERROR: Failed to process prompt\n";
-            goto stop_generating;
-        }
-        promptCtx.n_past += batch.size();
-        if (!recalculate(true))
-            goto stop_generating;
-        i = batch_end;
-    }
-    assert(promptCtx.n_past == int32_t(promptCtx.tokens.size()));
-
-stop_generating:
-    recalculate(false);
-}
-
 LLModel *LLModel::construct(const std::string &modelPath, std::string buildVariant) {
     //TODO: Auto-detect CUDA/OpenCL
     if (buildVariant == "auto") {

gpt4all-backend/llmodel_shared.cpp

@@ -0,0 +1,26 @@
+#include "llmodel.h"
+
+#include <cassert>
+#include <iostream>
+
+void LLModel::recalculateContext(PromptContext &promptCtx, std::function<bool(bool)> recalculate) {
+    size_t i = 0;
+    promptCtx.n_past = 0;
+    while (i < promptCtx.tokens.size()) {
+        size_t batch_end = std::min(i + promptCtx.n_batch, promptCtx.tokens.size());
+        std::vector<int32_t> batch(promptCtx.tokens.begin() + i, promptCtx.tokens.begin() + batch_end);
+        assert(promptCtx.n_past + int32_t(batch.size()) <= promptCtx.n_ctx);
+        if (!evalTokens(promptCtx, batch)) {
+            std::cerr << "LLModel ERROR: Failed to process prompt\n";
+            goto stop_generating;
+        }
+        promptCtx.n_past += batch.size();
+        if (!recalculate(true))
+            goto stop_generating;
+        i = batch_end;
+    }
+    assert(promptCtx.n_past == int32_t(promptCtx.tokens.size()));
+
+stop_generating:
+    recalculate(false);
+}