@fugood/llama.node 0.3.16 → 0.4.0

package/src/llama.cpp/tools/mtmd/clip.h

@@ -0,0 +1,99 @@
+#pragma once
+
+#include "ggml.h"
+#include <stddef.h>
+#include <stdint.h>
+
+struct clip_ctx;
+
+struct clip_image_size {
+    int width;
+    int height;
+};
+
+struct clip_image_f32;
+struct clip_image_u8_batch;
+struct clip_image_f32_batch;
+
+struct clip_context_params {
+    bool use_gpu;
+    enum ggml_log_level verbosity;
+};
+
+struct clip_ctx * clip_init(const char * fname, struct clip_context_params ctx_params);
+
+void clip_free(struct clip_ctx * ctx);
+
+size_t clip_embd_nbytes(const struct clip_ctx * ctx);
+size_t clip_embd_nbytes_by_img(const struct clip_ctx * ctx, int img_w, int img_h);
+
+int32_t clip_get_image_size (const struct clip_ctx * ctx);
+int32_t clip_get_patch_size (const struct clip_ctx * ctx);
+int32_t clip_get_hidden_size(const struct clip_ctx * ctx);
+
+// TODO: should be enum, not string
+const char * clip_patch_merge_type(const struct clip_ctx * ctx);
+
+const int32_t * clip_image_grid(const struct clip_ctx * ctx);
+size_t get_clip_image_grid_size(const struct clip_ctx * ctx);
+
+int clip_n_output_tokens(const struct clip_ctx * ctx, struct clip_image_f32 * img);
+
+// for M-RoPE, this will be the number of token positions in X and Y directions
+// for other models, X will be the total number of tokens and Y will be 1
+int clip_n_output_tokens_x(const struct clip_ctx * ctx, struct clip_image_f32 * img);
+int clip_n_output_tokens_y(const struct clip_ctx * ctx, struct clip_image_f32 * img);
+
+// this should be equal to the embedding dimension of the text model
+int clip_n_mmproj_embd(const struct clip_ctx * ctx);
+
+int clip_uhd_num_image_embeds_col(struct clip_ctx * ctx_clip);
+void clip_add_load_image_size(struct clip_ctx * ctx_clip, struct clip_image_size * load_image_size);
+struct clip_image_size * clip_get_load_image_size(struct clip_ctx * ctx_clip);
+
+struct clip_image_size      * clip_image_size_init(void);
+struct clip_image_u8        * clip_image_u8_init (void);
+struct clip_image_f32       * clip_image_f32_init(void);
+struct clip_image_f32_batch * clip_image_f32_batch_init(void); // only used by libllava
+
+// nx, ny are the output image dimensions
+unsigned char * clip_image_u8_get_data(struct clip_image_u8 * img, uint32_t * nx, uint32_t * ny);
+
+void clip_image_size_free (struct clip_image_size * img_size);
+void clip_image_u8_free (struct clip_image_u8  * img);
+void clip_image_f32_free(struct clip_image_f32 * img);
+void clip_image_u8_batch_free (struct clip_image_u8_batch  * batch);
+void clip_image_f32_batch_free(struct clip_image_f32_batch * batch);
+
+// use for accessing underlay data of clip_image_f32_batch
+size_t clip_image_f32_batch_n_images(const struct clip_image_f32_batch * batch); // equivalent to batch->size()
+size_t clip_image_f32_batch_nx(const struct clip_image_f32_batch * batch, int idx); // equivalent to batch[idx]->nx
+size_t clip_image_f32_batch_ny(const struct clip_image_f32_batch * batch, int idx); // equivalent to batch[idx]->ny
+struct clip_image_f32 * clip_image_f32_get_img(const struct clip_image_f32_batch * batch, int idx); // equivalent to batch[idx]->data
+
+/**
+ * Build image from pixels decoded by other libraries instead of stb_image.h for better performance.
+ * The memory layout is RGBRGBRGB..., input buffer length must be 3*nx*ny bytes
+ */
+void clip_build_img_from_pixels(const unsigned char * rgb_pixels, int nx, int ny, struct clip_image_u8 * img);
+
+bool clip_image_load_from_file(const char * fname, struct clip_image_u8 * img);
+
+/** interpret bytes as an image file with length bytes_length, and use the result to populate img */
+bool clip_image_load_from_bytes(const unsigned char * bytes, size_t bytes_length, struct clip_image_u8 * img);
+
+/** preprocess img and store the result in res_imgs, pad_to_square may be overridden to false depending on model configuration */
+bool clip_image_preprocess(struct clip_ctx * ctx, const struct clip_image_u8 * img, struct clip_image_f32_batch * res_imgs );
+
+struct ggml_tensor * clip_get_newline_tensor(const struct clip_ctx * ctx);
+
+bool clip_image_encode      (struct clip_ctx * ctx, int n_threads, struct clip_image_f32 * img, float * vec);
+bool clip_image_batch_encode(struct clip_ctx * ctx, int n_threads, const struct clip_image_f32_batch * imgs, float * vec);
+
+int clip_is_minicpmv(const struct clip_ctx * ctx);
+bool clip_is_glm(const struct clip_ctx * ctx);
+bool clip_is_qwen2vl(const struct clip_ctx * ctx);
+bool clip_is_llava(const struct clip_ctx * ctx);
+bool clip_is_gemma3(const struct clip_ctx * ctx);
+
+bool clip_encode_float_image (struct clip_ctx * ctx, int n_threads, float * img, int h, int w, float * vec);

package/src/llama.cpp/tools/mtmd/deprecation-warning.cpp

@@ -0,0 +1,22 @@
+#include <cstdio>
+#include <string>
+
+int main(int argc, char** argv) {
+    std::string filename = "main";
+    if (argc >= 1) {
+        filename = argv[0];
+    }
+
+    // Get only the program name from the full path
+    size_t pos = filename.find_last_of("/\\");
+    if (pos != std::string::npos) {
+        filename = filename.substr(pos+1);
+    }
+
+    fprintf(stdout, "\n");
+    fprintf(stdout, "WARNING: The binary '%s' is deprecated.\n", filename.c_str());
+    fprintf(stdout, "Please use 'llama-mtmd-cli' instead.\n");
+    fprintf(stdout, "\n");
+
+    return EXIT_FAILURE;
+}

package/src/llama.cpp/tools/mtmd/mtmd-cli.cpp

@@ -0,0 +1,370 @@
+#include "arg.h"
+#include "log.h"
+#include "common.h"
+#include "sampling.h"
+#include "llama.h"
+#include "ggml.h"
+#include "console.h"
+#include "chat.h"
+#include "mtmd.h"
+
+#include <vector>
+#include <limits.h>
+#include <cinttypes>
+
+#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
+#include <signal.h>
+#include <unistd.h>
+#elif defined (_WIN32)
+#define WIN32_LEAN_AND_MEAN
+#ifndef NOMINMAX
+#define NOMINMAX
+#endif
+#include <windows.h>
+#include <signal.h>
+#endif
+
+// volatile, because of signal being an interrupt
+static volatile bool g_is_generating = false;
+static volatile bool g_is_interrupted = false;
+
+/**
+ * Please note that this is NOT a production-ready stuff.
+ * It is a playground for trying multimodal support in llama.cpp.
+ * For contributors: please keep this code simple and easy to understand.
+ */
+
+static void show_additional_info(int /*argc*/, char ** argv) {
+    LOG(
+        "Experimental CLI for multimodal\n\n"
+        "Usage: %s [options] -m <model> --mmproj <mmproj> --image <image> -p <prompt>\n\n"
+        "  -m and --mmproj are required\n"
+        "  -hf user/repo can replace both -m and --mmproj in most cases\n"
+        "  --image and -p are optional, if NOT provided, the CLI will run in chat mode\n"
+        "  to disable using GPU for mmproj model, add --no-mmproj-offload\n",
+        argv[0]
+    );
+}
+
+#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__)) || defined (_WIN32)
+static void sigint_handler(int signo) {
+    if (signo == SIGINT) {
+        if (g_is_generating) {
+            g_is_generating = false;
+        } else {
+            console::cleanup();
+            if (g_is_interrupted) {
+                _exit(1);
+            }
+            g_is_interrupted = true;
+        }
+    }
+}
+#endif
+
+struct mtmd_cli_context {
+    mtmd::context_ptr ctx_vision;
+    common_init_result llama_init;
+
+    llama_model       * model;
+    llama_context     * lctx;
+    const llama_vocab * vocab;
+    llama_batch         batch;
+    int                 n_batch;
+
+    mtmd::bitmaps bitmaps;
+
+    // note: we know that gemma3 template is "linear", meaning each turn is completely separated to another
+    // so here we don't need to keep track of chat history
+    common_chat_templates_ptr tmpls;
+
+    // support for legacy templates (models not having EOT token)
+    llama_tokens antiprompt_tokens;
+
+    int n_threads    = 1;
+    llama_pos n_past = 0;
+
+    mtmd_cli_context(common_params & params) : llama_init(common_init_from_params(params)) {
+        model = llama_init.model.get();
+        lctx = llama_init.context.get();
+        vocab = llama_model_get_vocab(model);
+        n_threads = params.cpuparams.n_threads;
+        batch = llama_batch_init(params.n_batch, 0, 1);
+        n_batch = params.n_batch;
+
+        if (!model || !lctx) {
+            exit(1);
+        }
+
+        if (!llama_model_chat_template(model, nullptr) && params.chat_template.empty()) {
+            LOG_ERR("Model does not have chat template.\n");
+            LOG_ERR("  For old llava models, you may need to use '--chat-template vicuna'\n");
+            LOG_ERR("  For MobileVLM models, use '--chat-template deepseek'\n");
+            LOG_ERR("  For Mistral Small 3.1, use '--chat-template mistral-v7'\n");
+            exit(1);
+        }
+
+        tmpls = common_chat_templates_init(model, params.chat_template);
+        LOG_INF("%s: chat template example:\n%s\n", __func__, common_chat_format_example(tmpls.get(), params.use_jinja).c_str());
+
+        init_vision_context(params);
+
+        // load antiprompt tokens for legacy templates
+        if (params.chat_template == "vicuna") {
+            antiprompt_tokens = common_tokenize(lctx, "ASSISTANT:", false, true);
+        } else if (params.chat_template == "deepseek") {
+            antiprompt_tokens = common_tokenize(lctx, "###", false, true);
+        }
+    }
+
+    void init_vision_context(common_params & params) {
+        const char * clip_path = params.mmproj.path.c_str();
+        mtmd_context_params mparams = mtmd_context_params_default();
+        mparams.use_gpu = params.mmproj_use_gpu;
+        mparams.print_timings = true;
+        mparams.n_threads = params.cpuparams.n_threads;
+        mparams.verbosity = params.verbosity > 0 ? GGML_LOG_LEVEL_DEBUG : GGML_LOG_LEVEL_INFO;
+        ctx_vision.reset(mtmd_init_from_file(clip_path, model, mparams));
+        if (!ctx_vision.get()) {
+            LOG_ERR("Failed to load vision model from %s\n", clip_path);
+            exit(1);
+        }
+    }
+
+    bool check_antiprompt(const llama_tokens & generated_tokens) {
+        if (antiprompt_tokens.empty() || generated_tokens.size() < antiprompt_tokens.size()) {
+            return false;
+        }
+        return std::equal(
+            generated_tokens.end() - antiprompt_tokens.size(),
+            generated_tokens.end(),
+            antiprompt_tokens.begin()
+        );
+    }
+
+    bool load_image(const std::string & fname) {
+        mtmd::bitmap bmp(mtmd_helper_bitmap_init_from_file(fname.c_str()));
+        if (!bmp.ptr) {
+            return false;
+        }
+        bitmaps.entries.push_back(std::move(bmp));
+        return true;
+    }
+};
+
+static int generate_response(mtmd_cli_context & ctx, common_sampler * smpl, int n_predict) {
+    llama_tokens generated_tokens;
+    for (int i = 0; i < n_predict; i++) {
+        if (i > n_predict || !g_is_generating || g_is_interrupted) {
+            LOG("\n");
+            break;
+        }
+
+        llama_token token_id = common_sampler_sample(smpl, ctx.lctx, -1);
+        generated_tokens.push_back(token_id);
+        common_sampler_accept(smpl, token_id, true);
+
+        if (llama_vocab_is_eog(ctx.vocab, token_id) || ctx.check_antiprompt(generated_tokens)) {
+            LOG("\n");
+            break; // end of generation
+        }
+
+        LOG("%s", common_token_to_piece(ctx.lctx, token_id).c_str());
+        fflush(stdout);
+
+        if (g_is_interrupted) {
+            LOG("\n");
+            break;
+        }
+
+        // eval the token
+        common_batch_clear(ctx.batch);
+        common_batch_add(ctx.batch, token_id, ctx.n_past++, {0}, true);
+        if (llama_decode(ctx.lctx, ctx.batch)) {
+            LOG_ERR("failed to decode token\n");
+            return 1;
+        }
+    }
+    return 0;
+}
+
+static int eval_message(mtmd_cli_context & ctx, common_chat_msg & msg, bool add_bos = false) {
+    common_chat_templates_inputs tmpl_inputs;
+    tmpl_inputs.messages = {msg};
+    tmpl_inputs.add_generation_prompt = true;
+    tmpl_inputs.use_jinja = false; // jinja is buggy here
+    auto formatted_chat = common_chat_templates_apply(ctx.tmpls.get(), tmpl_inputs);
+    LOG_DBG("formatted_chat.prompt: %s\n", formatted_chat.prompt.c_str());
+
+    mtmd_input_text text;
+    text.text          = formatted_chat.prompt.c_str();
+    text.add_special   = add_bos;
+    text.parse_special = true;
+
+    if (g_is_interrupted) return 0;
+
+    mtmd::input_chunks chunks(mtmd_input_chunks_init());
+    auto bitmaps_c_ptr = ctx.bitmaps.c_ptr();
+    int32_t res = mtmd_tokenize(ctx.ctx_vision.get(),
+                        chunks.ptr.get(), // output
+                        &text, // text
+                        bitmaps_c_ptr.data(),
+                        bitmaps_c_ptr.size());
+    if (res != 0) {
+        LOG_ERR("Unable to tokenize prompt, res = %d\n", res);
+        return 1;
+    }
+
+    ctx.bitmaps.entries.clear();
+
+    llama_pos new_n_past;
+    if (mtmd_helper_eval_chunks(ctx.ctx_vision.get(),
+                ctx.lctx, // lctx
+                chunks.ptr.get(), // chunks
+                ctx.n_past, // n_past
+                0, // seq_id
+                ctx.n_batch, // n_batch
+                true, // logits_last
+                &new_n_past)) {
+        LOG_ERR("Unable to eval prompt\n");
+        return 1;
+    }
+
+    ctx.n_past = new_n_past;
+
+    LOG("\n");
+
+    return 0;
+}
+
+int main(int argc, char ** argv) {
+    ggml_time_init();
+
+    common_params params;
+    params.sampling.temp = 0.2; // lower temp by default for better quality
+
+    if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_LLAVA, show_additional_info)) {
+        return 1;
+    }
+
+    common_init();
+
+    if (params.mmproj.path.empty()) {
+        show_additional_info(argc, argv);
+        LOG_ERR("ERR: Missing --mmproj argument\n");
+        return 1;
+    }
+
+    mtmd_cli_context ctx(params);
+    LOG("%s: loading model: %s\n", __func__, params.model.path.c_str());
+
+    bool is_single_turn = !params.prompt.empty() && !params.image.empty();
+
+    struct common_sampler * smpl = common_sampler_init(ctx.model, params.sampling);
+    int n_predict = params.n_predict < 0 ? INT_MAX : params.n_predict;
+
+    // Ctrl+C handling
+    {
+#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
+        struct sigaction sigint_action;
+        sigint_action.sa_handler = sigint_handler;
+        sigemptyset (&sigint_action.sa_mask);
+        sigint_action.sa_flags = 0;
+        sigaction(SIGINT, &sigint_action, NULL);
+#elif defined (_WIN32)
+        auto console_ctrl_handler = +[](DWORD ctrl_type) -> BOOL {
+            return (ctrl_type == CTRL_C_EVENT) ? (sigint_handler(SIGINT), true) : false;
+        };
+        SetConsoleCtrlHandler(reinterpret_cast<PHANDLER_ROUTINE>(console_ctrl_handler), true);
+#endif
+    }
+
+    if (g_is_interrupted) return 130;
+
+    if (is_single_turn) {
+        g_is_generating = true;
+        if (params.prompt.find("<__image__>") == std::string::npos) {
+            params.prompt += " <__image__>";
+        }
+        common_chat_msg msg;
+        msg.role = "user";
+        msg.content = params.prompt;
+        for (const auto & image : params.image) {
+            if (!ctx.load_image(image)) {
+                return 1; // error is already printed by libmtmd
+            }
+        }
+        if (eval_message(ctx, msg, true)) {
+            return 1;
+        }
+        if (!g_is_interrupted && generate_response(ctx, smpl, n_predict)) {
+            return 1;
+        }
+
+    } else {
+        LOG("\n Running in chat mode, available commands:");
+        LOG("\n   /image <path>    load an image");
+        LOG("\n   /clear           clear the chat history");
+        LOG("\n   /quit or /exit   exit the program");
+        LOG("\n");
+
+        bool is_first_msg = true;
+        std::string content;
+
+        while (!g_is_interrupted) {
+            g_is_generating = false;
+            LOG("\n> ");
+            console::set_display(console::user_input);
+            std::string line;
+            console::readline(line, false);
+            if (g_is_interrupted) break;
+            console::set_display(console::reset);
+            line = string_strip(line);
+            if (line.empty()) {
+                continue;
+            }
+            if (line == "/quit" || line == "/exit") {
+                break;
+            }
+            if (line == "/clear") {
+                ctx.n_past = 0;
+                llama_kv_self_seq_rm(ctx.lctx, 0, 1, -1); // keep BOS
+                LOG("Chat history cleared\n\n");
+                continue;
+            }
+            g_is_generating = true;
+            if (line == "/image" || line.find("/image ") == 0) {
+                if (line.size() < 8) {
+                    LOG_ERR("ERR: Missing image filename\n");
+                    continue;
+                }
+                std::string image = line.substr(7);
+                if (ctx.load_image(image)) {
+                    LOG("Image %s loaded\n", image.c_str());
+                    content += "<__image__>";
+                }
+                // else, error is already printed by libmtmd
+                continue;
+            } else {
+                content += line;
+            }
+            common_chat_msg msg;
+            msg.role = "user";
+            msg.content = content;
+            int ret = eval_message(ctx, msg, is_first_msg);
+            if (ret) {
+                return 1;
+            }
+            if (g_is_interrupted) break;
+            if (generate_response(ctx, smpl, n_predict)) {
+                return 1;
+            }
+            content.clear();
+            is_first_msg = false;
+        }
+    }
+    if (g_is_interrupted) LOG("\nInterrupted by user\n");
+    LOG("\n\n");
+    llama_perf_context_print(ctx.lctx);
+    return g_is_interrupted ? 130 : 0;
+}