add Falcon 7B model

Tested with https://huggingface.co/TheBloke/falcon-7b-instruct-GGML/blob/main/falcon7b-instruct.ggmlv3.q4_0.bin

gpt4all-backend/CMakeLists.txt

@@ -117,6 +117,10 @@ foreach(BUILD_VARIANT IN LISTS BUILD_VARIANTS)
             gptj.cpp utils.h utils.cpp llmodel_shared.cpp)
         prepare_target(gptj ggml-230511)
 
+        add_library(falcon-${BUILD_VARIANT} SHARED
+            falcon.cpp utils.h utils.cpp llmodel_shared.cpp)
+        prepare_target(falcon llama-mainline)
+
         add_library(mpt-${BUILD_VARIANT} SHARED
             mpt.cpp utils.h utils.cpp llmodel_shared.cpp)
         prepare_target(mpt ggml-230511)

gpt4all-backend/falcon_impl.h

@@ -0,0 +1,40 @@
+#ifndef FALCON_H_I_KNOW_WHAT_I_AM_DOING_WHEN_INCLUDING_THIS_FILE
+#error This file is NOT meant to be included outside of falcon.cpp. Doing so is DANGEROUS. Be sure to know what you are doing before proceeding to #define FALCON_H_I_KNOW_WHAT_I_AM_DOING_WHEN_INCLUDING_THIS_FILE
+#endif
+#ifndef FALCON_H
+#define FALCON_H
+
+#include <string>
+#include <functional>
+#include <vector>
+#include <memory>
+#include "llmodel.h"
+
+struct FalconPrivate;
+class Falcon : public LLModel {
+public:
+    Falcon();
+    ~Falcon();
+
+    bool loadModel(const std::string &modelPath) override;
+    bool isModelLoaded() const override;
+    size_t requiredMem(const std::string &modelPath) override;
+    size_t stateSize() const override;
+    size_t saveState(uint8_t *dest) const override;
+    size_t restoreState(const uint8_t *src) override;
+    void setThreadCount(int32_t n_threads) override;
+    int32_t threadCount() const override;
+
+private:
+    std::unique_ptr<FalconPrivate> d_ptr;
+
+protected:
+    std::vector<Token> tokenize(PromptContext &, const std::string&) const override;
+    Token sampleToken(PromptContext &ctx) const override;
+    std::string tokenToString(Token) const override;
+    bool evalTokens(PromptContext &ctx, const std::vector<int32_t> &tokens) const override;
+    int32_t contextLength() const override;
+    const std::vector<Token>& endTokens() const override;
+};
+
+#endif // Falcon_H

gpt4all-backend/llamamodel.cpp

@@ -282,14 +282,15 @@ DLL_EXPORT bool magic_match(std::istream& f) {
     if (!(version LLAMA_VERSIONS)) {
         return false;
     }
+    llama_file_hparams hparams;
+    f.read(reinterpret_cast<char*>(&hparams), sizeof(hparams));
+    if (!(hparams.n_vocab >= 32000 && hparams.n_vocab <= 32100)) {
+        return false; // not a llama.
+    }
 #ifdef GGML_USE_METAL
     // Check quant supported on metal
     // skip fields
-    off_t offset = sizeof(uint32_t) * 6; // n_vocab, n_embd, n_mult, n_head, n_layer, n_rot
-    f.seekg(offset, std::ios_base::cur);
-    uint32_t ftype;
-    f.read(reinterpret_cast<char*>(&ftype), sizeof(ftype)); // ftype
-    switch((enum llama_ftype) ftype) {
+    switch(hparams.ftype) {
         // currently supported on Metal https://github.com/ggerganov/llama.cpp/blob/ae9663f1887513e152839e91f61c513075a19422/ggml-metal.m#L51-L55
         case LLAMA_FTYPE_MOSTLY_F16:
         case LLAMA_FTYPE_MOSTLY_Q2_K:

gpt4all-chat/CMakeLists.txt

@@ -178,6 +178,8 @@ install(TARGETS llamamodel-mainline-default DESTINATION lib COMPONENT ${COMPONEN
 if(APPLE)
 install(TARGETS llamamodel-mainline-metal DESTINATION lib COMPONENT ${COMPONENT_NAME_MAIN})
 endif()
+install(TARGETS falcon-avxonly DESTINATION lib COMPONENT ${COMPONENT_NAME_MAIN})
+install(TARGETS falcon-default DESTINATION lib COMPONENT ${COMPONENT_NAME_MAIN})
 install(TARGETS mpt-avxonly DESTINATION lib COMPONENT ${COMPONENT_NAME_MAIN})
 install(TARGETS mpt-default DESTINATION lib COMPONENT ${COMPONENT_NAME_MAIN})
 install(TARGETS replit-mainline-avxonly DESTINATION lib COMPONENT ${COMPONENT_NAME_MAIN})

gpt4all-chat/chatllm.cpp

@@ -224,6 +224,7 @@ bool ChatLLM::loadModel(const ModelInfo &modelInfo)
                     case 'G': m_llModelType = LLModelType::GPTJ_; break;
                     case 'M': m_llModelType = LLModelType::MPT_; break;
                     case 'R': m_llModelType = LLModelType::REPLIT_; break;
+                    case 'F': m_llModelType = LLModelType::FALCON_; break;
                     default:
                         {
                             delete std::exchange(m_llModelInfo.model, nullptr);

gpt4all-chat/chatllm.h

@@ -14,7 +14,8 @@ enum LLModelType {
     GPTJ_,
     LLAMA_,
     CHATGPT_,
-    REPLIT_
+    REPLIT_,
+    FALCON_
 };
 
 struct LLModelInfo {