#include "common.h"
#include "log.h"
#include "ggml-backend.h"
#include "ggml.h"
#include "gguf.h"
#include "ggml-cpp.h"
#include "llama.h"
#include "llama-cpp.h"
#include "../src/llama-arch.h"
#include "../src/llama-model-saver.h"

#include <cinttypes>
#include <cstdio>
#include <cstring>
#include <cstdint>
#include <random>
#include <stdexcept>
#include <string>
#include <utility>
#include <vector>

// normalized mean squared error = mse(a, b) / mse(a, 0)
static double nmse(const std::vector<float> & a, const std::vector<float> & b) {
    GGML_ASSERT(a.size() == b.size());
    double mse_a_b = 0.0;
    double mse_a_0 = 0.0;

    for (size_t i = 0; i < a.size(); i++) {
        float a_i = a[i];
        float b_i = b[i];

        mse_a_b += (a_i - b_i) * (a_i - b_i);
        mse_a_0 += a_i * a_i;
    }

    return mse_a_b / mse_a_0;
}

static void set_tensor_data(struct ggml_tensor * tensor, void * userdata) {
    std::hash<std::string> hasher;
    std::mt19937 gen(hasher(tensor->name) + *(const size_t *) userdata);
    std::normal_distribution<float> dis(0.0f, 1.0e-2f);

    const int64_t ne = ggml_nelements(tensor);
    if (tensor->type == GGML_TYPE_F32) {
        std::vector<float> tmp(ne);
        for (int64_t i = 0; i < ne; i++) {
            tmp[i] = dis(gen);
        }
        ggml_backend_tensor_set(tensor, tmp.data(), 0, ggml_nbytes(tensor));
    } else if (tensor->type == GGML_TYPE_F16) {
        std::vector<ggml_fp16_t> tmp(ne);
        for (int64_t i = 0; i < ne; i++) {
            tmp[i] = ggml_fp32_to_fp16(dis(gen));
        }
        ggml_backend_tensor_set(tensor, tmp.data(), 0, ggml_nbytes(tensor));
    } else {
        GGML_ABORT("fatal error");
    }
}

static void usage(char ** argv) {
    printf("Usage: %s [-a/--arch arch] [-s/--seed seed] [-v/--verbose]\n", argv[0]);
}

static std::vector<llama_token> get_tokens(const uint32_t n_tokens, const uint32_t n_vocab, const size_t seed){
    std::mt19937 gen(seed);
    std::uniform_int_distribution<> dis(0, n_vocab - 1);
    std::vector<llama_token> ret;
    ret.reserve(n_tokens);
    for (uint32_t i = 0; i < n_tokens; i++) {
        ret.push_back(dis(gen));
    }
    return ret;
}

static gguf_context_ptr get_gguf_ctx(const llm_arch arch, const bool moe) {
    gguf_context_ptr ret(gguf_init_empty());
    llama_model_saver ms(arch, ret.get());
    const uint32_t n_ctx = 128;

    uint32_t n_vocab = 128;
    uint32_t n_embd  = 256;
    uint32_t n_head  = 2;
    uint32_t n_ff    = 384;
    uint32_t n_layer = 2;
    if (arch == LLM_ARCH_LLAMA4) {
        n_layer = 4; // hparams.n_no_rope_layer_step is hard-coded to 4
    } else if (arch == LLM_ARCH_GEMMA3N) {
        n_embd = 64;
        n_head = 1;
        n_ff   = 96;
        n_layer = 22; // hparams.n_layer_kv_from_start = 20 is hardcoded
    } else if (arch == LLM_ARCH_DEEPSEEK2
            || arch == LLM_ARCH_GLM_DSA
            || arch == LLM_ARCH_KIMI_LINEAR
            || arch == LLM_ARCH_MISTRAL4) {
        n_embd = 128;
        n_head = 1;
        n_ff   = 192;
    } else if (arch == LLM_ARCH_NEMOTRON_H || arch == LLM_ARCH_NEMOTRON_H_MOE) {
        n_layer = 3;
    } else if (arch == LLM_ARCH_CHAMELEON) {
        n_vocab = 10240;
    }

    const uint32_t n_embd_head = n_embd / n_head;

    ms.add_kv(LLM_KV_GENERAL_ARCHITECTURE,      llm_arch_name(arch));
    ms.add_kv(LLM_KV_VOCAB_SIZE,                n_vocab);
    ms.add_kv(LLM_KV_CONTEXT_LENGTH,            n_ctx);
    ms.add_kv(LLM_KV_EMBEDDING_LENGTH,          n_embd);
    ms.add_kv(LLM_KV_FEATURES_LENGTH,           n_embd);
    ms.add_kv(LLM_KV_BLOCK_COUNT,               n_layer);
    ms.add_kv(LLM_KV_LEADING_DENSE_BLOCK_COUNT, uint32_t(1));

    if (arch == LLM_ARCH_NEMOTRON_H || arch == LLM_ARCH_NEMOTRON_H_MOE) {
        std::vector<uint32_t> n_ff_per_layer;
        n_ff_per_layer.reserve(n_layer);
        for (uint32_t il = 0; il < n_layer; il++) {
            n_ff_per_layer.push_back(il <= 1 ? 0 : n_ff);
        }
        ms.add_kv(LLM_KV_FEED_FORWARD_LENGTH, n_ff_per_layer);
    } else {
        ms.add_kv(LLM_KV_FEED_FORWARD_LENGTH, n_ff);
    }

    ms.add_kv(LLM_KV_USE_PARALLEL_RESIDUAL,   false);
    ms.add_kv(LLM_KV_LOGIT_SCALE,             1.0f);
    ms.add_kv(LLM_KV_TIME_MIX_EXTRA_DIM,      uint32_t(64));
    ms.add_kv(LLM_KV_TIME_DECAY_EXTRA_DIM,    uint32_t(128));
    ms.add_kv(LLM_KV_FULL_ATTENTION_INTERVAL, uint32_t(2));

    if (arch == LLM_ARCH_PLAMO2 || arch == LLM_ARCH_JAMBA || arch == LLM_ARCH_NEMOTRON_H || arch == LLM_ARCH_NEMOTRON_H_MOE ||
            arch == LLM_ARCH_GRANITE_HYBRID || arch == LLM_ARCH_LFM2 || arch == LLM_ARCH_LFM2MOE || arch == LLM_ARCH_KIMI_LINEAR) {
        GGML_ASSERT(n_layer >= 2);
        std::vector<uint32_t> n_head_per_layer;
        n_head_per_layer.reserve(n_layer);
        for (uint32_t il = 0; il < n_layer; il++) {
            n_head_per_layer.push_back(il == 1 ? 0 : n_head);
        }
        ms.add_kv(LLM_KV_ATTENTION_HEAD_COUNT, n_head_per_layer);
        ms.add_kv(LLM_KV_ATTENTION_HEAD_COUNT_KV, n_head_per_layer);
    } else {
        ms.add_kv(LLM_KV_ATTENTION_HEAD_COUNT, n_head);
        ms.add_kv(LLM_KV_ATTENTION_HEAD_COUNT_KV, n_head);
    }

    ms.add_kv(LLM_KV_ATTENTION_MAX_ALIBI_BIAS, 8.0f);
    if (arch == LLM_ARCH_DEEPSEEK2
            || arch == LLM_ARCH_GLM_DSA
            || arch == LLM_ARCH_KIMI_LINEAR
            || arch == LLM_ARCH_MISTRAL4) {
        ms.add_kv(LLM_KV_ATTENTION_KEY_LENGTH,       uint32_t(576));
        ms.add_kv(LLM_KV_ATTENTION_VALUE_LENGTH,     uint32_t(512));
        ms.add_kv(LLM_KV_ROPE_DIMENSION_COUNT,       uint32_t(64));
        ms.add_kv(LLM_KV_ATTENTION_KEY_LENGTH_MLA,   uint32_t(192));
        ms.add_kv(LLM_KV_ATTENTION_VALUE_LENGTH_MLA, uint32_t(128));
    }
    ms.add_kv(LLM_KV_ATTENTION_CLAMP_KQV,              1.0f);
    ms.add_kv(LLM_KV_ATTENTION_LAYERNORM_EPS,          1e-5f);
    ms.add_kv(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS,      1e-5f);
    ms.add_kv(LLM_KV_ATTENTION_GROUPNORM_EPS,          1e-5f);
    ms.add_kv(LLM_KV_ATTENTION_GROUPNORM_GROUPS,       uint32_t(8));
    ms.add_kv(LLM_KV_ATTENTION_Q_LORA_RANK,            uint32_t(512));
    ms.add_kv(LLM_KV_ATTENTION_KV_LORA_RANK,           uint32_t(512));
    ms.add_kv(LLM_KV_ATTENTION_RELATIVE_BUCKETS_COUNT, uint32_t(8));
    ms.add_kv(LLM_KV_ATTENTION_SLIDING_WINDOW,         n_ctx/8);

    if (arch == LLM_ARCH_MIMO2 || arch == LLM_ARCH_STEP35) {
        std::vector<uint32_t> pattern;
        pattern.reserve(n_layer);
        for (uint32_t il = 0; il < n_layer; il++) {
            pattern.push_back(il % 2);
        }
        ms.add_kv(LLM_KV_ATTENTION_SLIDING_WINDOW_PATTERN, pattern);
    } else {
        ms.add_kv(LLM_KV_ATTENTION_SLIDING_WINDOW_PATTERN, uint32_t(2));
    }

    ms.add_kv(LLM_KV_ATTENTION_INDEXER_HEAD_COUNT, uint32_t(1));
    ms.add_kv(LLM_KV_ATTENTION_INDEXER_KEY_LENGTH, uint32_t(64));
    ms.add_kv(LLM_KV_ATTENTION_INDEXER_TOP_K,      uint32_t(8));
    ms.add_kv(LLM_KV_ROPE_DIMENSION_SECTIONS, std::vector<uint32_t>({n_embd_head/4, n_embd_head/4, n_embd_head/4, n_embd_head/4}));
    ms.add_kv(LLM_KV_TOKENIZER_MODEL,         "no_vocab");
    // ms.add_kv(LLM_KV_DENSE_2_FEAT_OUT,     n_embd);
    // ms.add_kv(LLM_KV_DENSE_3_FEAT_IN,      n_embd);

    if (moe) {
        ms.add_kv(LLM_KV_EXPERT_FEED_FORWARD_LENGTH, n_ff);
        ms.add_kv(LLM_KV_INTERLEAVE_MOE_LAYER_STEP,  uint32_t(2));
        ms.add_kv(LLM_KV_EXPERT_COUNT,               uint32_t(2));
        ms.add_kv(LLM_KV_EXPERT_USED_COUNT,          uint32_t(1));
        ms.add_kv(LLM_KV_EXPERT_SHARED_COUNT,        uint32_t(1));
        ms.add_kv(LLM_KV_EXPERT_GATING_FUNC,         uint32_t(2)); // sigmoid
        ms.add_kv(LLM_KV_EXPERT_GROUP_SCALE,         1.0f);
        ms.add_kv(LLM_KV_EXPERTS_PER_GROUP,          uint32_t(1));
    }

    ms.add_kv(LLM_KV_POSNET_EMBEDDING_LENGTH,   n_embd);
    ms.add_kv(LLM_KV_POSNET_BLOCK_COUNT,        n_layer);
    ms.add_kv(LLM_KV_CONVNEXT_EMBEDDING_LENGTH, n_embd);
    ms.add_kv(LLM_KV_CONVNEXT_BLOCK_COUNT,      n_layer);
    ms.add_kv(LLM_KV_XIELU_ALPHA_N,             1.0f);
    ms.add_kv(LLM_KV_XIELU_ALPHA_P,             1.0f);
    ms.add_kv(LLM_KV_XIELU_BETA,                1.0f);
    ms.add_kv(LLM_KV_XIELU_EPS,                 1.0e-7f);
    ms.add_kv(LLM_KV_SSM_INNER_SIZE,            arch == LLM_ARCH_QWEN3NEXT || arch == LLM_ARCH_QWEN35 || arch == LLM_ARCH_QWEN35MOE ? 64 : 2*n_embd);
    ms.add_kv(LLM_KV_SSM_CONV_KERNEL,           uint32_t(4));
    ms.add_kv(LLM_KV_SSM_STATE_SIZE,            uint32_t(32));
    ms.add_kv(LLM_KV_SSM_TIME_STEP_RANK,        n_head);
    ms.add_kv(LLM_KV_SSM_GROUP_COUNT,           arch == LLM_ARCH_PLAMO2 ? 0 : uint32_t(2));
    ms.add_kv(LLM_KV_KDA_HEAD_DIM,              uint32_t(128));
    ms.add_kv(LLM_KV_WKV_HEAD_SIZE,             n_embd/n_head);
    ms.add_kv(LLM_KV_SHORTCONV_L_CACHE,         uint32_t(3));

    for (uint32_t il = 0; il < n_layer; il++) {
        ggml_tensor t;
        memset(&t, 0, sizeof(ggml_tensor));
        t.type = GGML_TYPE_F16;
        ggml_format_name(&t, "conv%" PRIu32 "d.weight", il);
        gguf_add_tensor(ms.gguf_ctx, &t);
        ggml_format_name(&t, "posnet.%" PRIu32 ".conv1.weight", il);
        gguf_add_tensor(ms.gguf_ctx, &t);
        ggml_format_name(&t, "posnet.%" PRIu32 ".conv2.weight", il);
        gguf_add_tensor(ms.gguf_ctx, &t);
        ggml_format_name(&t, "convnext.%" PRIu32 ".dw.weight", il);
        gguf_add_tensor(ms.gguf_ctx, &t);
    }
    return ret;
}

static bool silent_model_load_progress(float /*progress*/, void * /*user_data*/) {
    return true;
}

static std::pair<llama_model_ptr, llama_context_ptr> get_model_and_ctx(
        struct gguf_context * gguf_ctx, FILE * file, const size_t seed, const std::vector<ggml_backend_dev_t> & devs) {
    GGML_ASSERT((gguf_ctx == nullptr) != (file == nullptr));
    llama_model_params model_params = llama_model_default_params();
    model_params.progress_callback = silent_model_load_progress;
    std::vector<ggml_backend_dev_t> devs_copy = devs;
    devs_copy.push_back(nullptr);
    model_params.devices = devs_copy.data();

    llama_context_params ctx_params = llama_context_default_params();
    ctx_params.n_ctx = 0;
    ctx_params.n_threads = 4;
    ctx_params.n_threads_batch = 4;

    size_t tmp = seed;
    llama_model_ptr model(gguf_ctx != nullptr ?
        llama_model_init_from_user(gguf_ctx, set_tensor_data, &tmp, model_params) :
        llama_model_load_from_file_ptr(file, model_params));
    if (!model) {
        throw std::runtime_error("failed to create llama model");
    }
    llama_context_ptr lctx(llama_init_from_model(model.get(), ctx_params));
    if (!lctx) {
        throw std::runtime_error("failed to create llama context");
    }
    return std::make_pair(std::move(model), std::move(lctx));
}

static std::vector<float> get_logits(
        llama_model * model, llama_context * lctx, const std::vector<llama_token> & tokens, bool encode = false) {
    const uint32_t n_vocab  = llama_vocab_n_tokens(llama_model_get_vocab(model));
    const uint32_t n_ctx    = llama_n_ctx(lctx);
    const uint32_t n_tokens = tokens.size();
    llama_batch batch = llama_batch_init(n_ctx, 0, 1);
    GGML_ASSERT(n_tokens <= n_ctx);
    for (uint32_t pos = 0; pos < n_tokens; pos++) {
        common_batch_add(batch, tokens[pos], pos, {0}, true);
    }
    batch.n_tokens = n_tokens;
    if (encode) {
        if (llama_encode(lctx, batch)) {
            llama_batch_free(batch);
            throw std::runtime_error("failed to encode batch");
        }
    }
    if (llama_decode(lctx, batch)) {
        llama_batch_free(batch);
        throw std::runtime_error("failed to decode batch");
    }

    std::vector<float> ret;
    ret.reserve(n_tokens*n_vocab);
    for (uint32_t i = 0; i < n_tokens; i++) {
        const float * logits_ith = llama_get_logits_ith(lctx, i);
        for (uint32_t j = 0; j < n_vocab; j++) {
            ret.push_back(logits_ith[j]);
        }
    }
    llama_batch_free(batch);
    return ret;
}

static bool moe_mandatory(const llm_arch arch) {
    switch (arch) {
        case LLM_ARCH_LLAMA4:
        case LLM_ARCH_GROK:
        case LLM_ARCH_QWEN2MOE:
        case LLM_ARCH_QWEN3MOE:
        case LLM_ARCH_QWEN3NEXT:
        case LLM_ARCH_QWEN3VLMOE:
        case LLM_ARCH_QWEN35MOE:
        case LLM_ARCH_PHIMOE:
        case LLM_ARCH_DBRX:
        case LLM_ARCH_OLMOE:
        case LLM_ARCH_ARCTIC:
        case LLM_ARCH_DEEPSEEK:
        case LLM_ARCH_DEEPSEEK2:
        case LLM_ARCH_GLM4_MOE:
        case LLM_ARCH_GLM_DSA:
        case LLM_ARCH_EXAONE_MOE:
        case LLM_ARCH_BAILINGMOE:
        case LLM_ARCH_BAILINGMOE2:
        case LLM_ARCH_DOTS1:
        case LLM_ARCH_AFMOE:
        case LLM_ARCH_ERNIE4_5:
        case LLM_ARCH_ERNIE4_5_MOE:
        case LLM_ARCH_HUNYUAN_MOE:
        case LLM_ARCH_OPENAI_MOE:
        case LLM_ARCH_LFM2MOE:
        case LLM_ARCH_SMALLTHINKER:
        case LLM_ARCH_LLADA_MOE:
        case LLM_ARCH_GROVEMOE:
        case LLM_ARCH_MINIMAX_M2:
        case LLM_ARCH_RND1:
        case LLM_ARCH_PADDLEOCR:
        case LLM_ARCH_MIMO2:
        case LLM_ARCH_KIMI_LINEAR:
        case LLM_ARCH_STEP35:
        case LLM_ARCH_MISTRAL4:
            return true;
        default:
            return false;
    }
}

static bool moe_implemented(const llm_arch arch) {
    if (moe_mandatory(arch)) {
        return true;
    }
    switch (arch) {
        case LLM_ARCH_LLAMA:
        case LLM_ARCH_REFACT:
        case LLM_ARCH_MINICPM:
        case LLM_ARCH_GRANITE:
        case LLM_ARCH_GRANITE_MOE:
        case LLM_ARCH_MISTRAL3:
        case LLM_ARCH_LLAMA_EMBED:
            return true;
        default:
            return false;
    }
}

static int save_models(const llm_arch target_arch, const size_t seed, const ggml_log_level log_level, const std::string & dir) {
    struct user_data_t {
        struct {
            ggml_log_callback callback;
            void * user_data;
        } original_logger;
        ggml_log_level min_level; // prints below this log level go to debug log
    };
    user_data_t ud;
    llama_log_get(&ud.original_logger.callback, &ud.original_logger.user_data);
    ud.min_level = log_level;

    llama_log_set([](ggml_log_level level, const char * text, void * user_data) {
        const user_data_t * ud = (const user_data_t *) user_data;
        const ggml_log_level level_eff = level >= ud->min_level ? level : GGML_LOG_LEVEL_DEBUG;
        ud->original_logger.callback(level_eff, text, ud->original_logger.user_data);
    }, &ud);

    for (const llm_arch & arch : llm_arch_all()) {
        if (target_arch != LLM_ARCH_UNKNOWN && arch != target_arch) {
            continue;
        }
        if (arch == LLM_ARCH_CLIP || arch == LLM_ARCH_GPTJ || arch == LLM_ARCH_UNKNOWN) {
            continue; // These models don't have usable implementations.
        }
        if (arch == LLM_ARCH_CHAMELEON) {
            continue; // Only half-implemented and to be removed in the future.
        }
        if (arch == LLM_ARCH_GEMMA4) {
            continue; // FIXME @ngxson
        }
        if (arch == LLM_ARCH_RWKV6 || arch == LLM_ARCH_RWKV6QWEN2 || arch == LLM_ARCH_RWKV7 || arch == LLM_ARCH_ARWKV7) {
            continue; // FIXME
        }
        if (arch == LLM_ARCH_BERT || arch == LLM_ARCH_MODERN_BERT || arch == LLM_ARCH_NOMIC_BERT || arch == LLM_ARCH_NOMIC_BERT_MOE ||
                arch == LLM_ARCH_NEO_BERT || arch == LLM_ARCH_JINA_BERT_V2 || arch == LLM_ARCH_JINA_BERT_V3 || arch == LLM_ARCH_EUROBERT) {
            continue; // TODO vocab
        }
        if (arch == LLM_ARCH_PLM) {
            continue; // TODO tensor shapes
        }
        for (bool moe : {false, true}) {
            if (moe && !moe_implemented(arch)) {
                continue;
            }
            if (!moe && moe_mandatory(arch)) {
                continue;
            }
            if (!llama_model_saver_supports_arch(arch)) {
                LOG_INF("%s: %s model (%s) is unsupported, skipping\n", __func__, llm_arch_name(arch), moe ? "MoE" : "dense");
                continue;
            }
            gguf_context_ptr gguf_ctx = get_gguf_ctx(arch, moe);
            auto model_and_ctx = get_model_and_ctx(gguf_ctx.get(), nullptr, seed, {});
            const std::string path = dir + "/" + llm_arch_name(arch) + (moe ? "-moe.gguf" : "-dense.gguf");
            LOG_INF("%s: Saving %s model (%s) to %s...\n", __func__, llm_arch_name(arch), moe ? "MoE" : "dense", path.c_str());
            llama_model_save_to_file(model_and_ctx.first.get(), path.c_str());
        }
    }
    llama_log_set(ud.original_logger.callback, ud.original_logger.user_data);
    return 0;
}

static int test_backends(const llm_arch target_arch, const size_t seed, const ggml_log_level log_level) {
    struct user_data_t {
        struct {
            ggml_log_callback callback;
            void * user_data;
        } original_logger;
        ggml_log_level min_level; // prints below this log level go to debug log
    };
    user_data_t ud;
    llama_log_get(&ud.original_logger.callback, &ud.original_logger.user_data);
    ud.min_level = log_level;

    llama_log_set([](ggml_log_level level, const char * text, void * user_data) {
        const user_data_t * ud = (const user_data_t *) user_data;
        const ggml_log_level level_eff = level >= ud->min_level ? level : GGML_LOG_LEVEL_DEBUG;
        ud->original_logger.callback(level_eff, text, ud->original_logger.user_data);
    }, &ud);

    const std::vector<llama_token> tokens = get_tokens(128, 128, seed);

    bool all_ok = true;
    common_log_flush(common_log_main());
    printf("|%15s|%30s|%6s|%15s|%9s|\n", "Model arch.", "Device", "Config", "NMSE vs. CPU", "Roundtrip");
    printf("|---------------|------------------------------|------|---------------|---------|\n");
    for (const llm_arch & arch : llm_arch_all()) {
        if (target_arch != LLM_ARCH_UNKNOWN && arch != target_arch) {
            continue;
        }
        if (arch == LLM_ARCH_CLIP || arch == LLM_ARCH_GPTJ || arch == LLM_ARCH_UNKNOWN) {
            continue; // These models don't have usable implementations.
        }
        if (arch == LLM_ARCH_CHAMELEON) {
            continue; // Only half-implemented and to be removed in the future.
        }
        if (arch == LLM_ARCH_GEMMA4) {
            continue; // FIXME @ngxson
        }
        if (arch == LLM_ARCH_WAVTOKENIZER_DEC) {
            continue; // FIXME CUDA backend crashes.
        }
        if (arch == LLM_ARCH_LLAMA_EMBED || arch == LLM_ARCH_GEMMA_EMBEDDING || arch == LLM_ARCH_T5ENCODER) {
            continue; // FIXME Embedding (?) models produce inconsistent results.
        }
        if (arch == LLM_ARCH_RWKV6 || arch == LLM_ARCH_RWKV6QWEN2 || arch == LLM_ARCH_RWKV7 || arch == LLM_ARCH_ARWKV7) {
            continue; // FIXME RWKV models hang indefinitely.
        }
        if (arch == LLM_ARCH_BERT || arch == LLM_ARCH_MODERN_BERT || arch == LLM_ARCH_NOMIC_BERT || arch == LLM_ARCH_NOMIC_BERT_MOE ||
                arch == LLM_ARCH_NEO_BERT || arch == LLM_ARCH_JINA_BERT_V2 || arch == LLM_ARCH_JINA_BERT_V3 || arch == LLM_ARCH_EUROBERT) {
            continue; // TODO vocab
        }
        if (arch == LLM_ARCH_PLM) {
            continue; // TODO tensor shapes
        }
        if (arch == LLM_ARCH_DEEPSEEK2OCR) {
            continue; // TODO tensor shapes
        }

        // FIXME some models are segfaulting with WebGPU:
#ifdef GGML_USE_WEBGPU
        if (arch == LLM_ARCH_QWEN3NEXT || arch == LLM_ARCH_QWEN35 || arch == LLM_ARCH_QWEN35MOE || arch == LLM_ARCH_KIMI_LINEAR) {
            continue;
        }
#endif // GGML_USE_WEBGPU

        const bool encode = arch == LLM_ARCH_T5;
        for (bool moe : {false, true}) {
            if (moe && !moe_implemented(arch)) {
                continue;
            }
            if (!moe && moe_mandatory(arch)) {
                continue;
            }
            gguf_context_ptr gguf_ctx = get_gguf_ctx(arch, moe);
            auto model_and_ctx_cpu = get_model_and_ctx(gguf_ctx.get(), nullptr, seed, {});
            const std::vector<float> logits_cpu = get_logits(model_and_ctx_cpu.first.get(), model_and_ctx_cpu.second.get(), tokens, encode);
            for (size_t i = 0; i < ggml_backend_dev_count(); i++) {
                ggml_backend_dev_t dev = ggml_backend_dev_get(i);
                if (ggml_backend_dev_type(dev) == GGML_BACKEND_DEVICE_TYPE_CPU) {
                    continue;
                }
                auto model_and_ctx_dev = get_model_and_ctx(gguf_ctx.get(), nullptr, seed, {dev});
                std::string config_name = moe ? "MoE" : "Dense";
                const std::vector<float> logits_dev = get_logits(model_and_ctx_dev.first.get(), model_and_ctx_dev.second.get(), tokens, encode);
                const double nmse_val = nmse(logits_cpu, logits_dev);
                char nmse_str[10];
                snprintf(nmse_str, sizeof(nmse_str), "%.2e", nmse_val);
                std::string status_nmse = "\033[1;32mOK\033[0m";
                if (nmse_val > 1e-4) {
                    all_ok = false;
                    status_nmse = "\033[1;31mFAIL\033[0m";
                }

                std::string status_roundtrip = "\033[1;33mSKIP\033[0m";
                FILE * file = tmpfile(); // Can be null on Windows without administrator privileges.
                if (file != nullptr && llama_model_saver_supports_arch(arch)) {
                    llama_model_saver ms = llama_model_saver(model_and_ctx_dev.first.get());
                    ms.add_kv_from_model();
                    ms.add_tensors_from_model();
                    ms.save(file);
                    rewind(file);

                    auto model_and_ctx_roundtrip = get_model_and_ctx(nullptr, file, seed, {dev});
                    const std::vector<float> logits_roundtrip = get_logits(
                        model_and_ctx_roundtrip.first.get(), model_and_ctx_roundtrip.second.get(), tokens, encode);
                    status_roundtrip = "\033[1;32mOK\033[0m";
                    GGML_ASSERT(logits_roundtrip.size() == logits_dev.size());
                    for (size_t i = 0; i < logits_roundtrip.size(); i++) {
                        if (logits_roundtrip[i] != logits_dev[i]) {
                            all_ok = false;
                            status_roundtrip = "\033[1;31mFAIL\033[0m";
                            break;
                        }
                    }
                }

                printf("|%15s|%30s|%6s|%15s (%8s)|%20s|\n", llm_arch_name(arch), ggml_backend_dev_description(dev),
                    config_name.c_str(), status_nmse.c_str(), nmse_str, status_roundtrip.c_str());
            }
        }
    }
    llama_log_set(ud.original_logger.callback, ud.original_logger.user_data);
    return all_ok ? 0 : 1;
}

int main(int argc, char ** argv) {
    // FIXME these tests are disabled in the CI for macOS-latest-cmake-arm64 because they are segfaulting
    common_init();
    std::random_device rd;

    llm_arch arch = LLM_ARCH_UNKNOWN;
    size_t seed = rd();
    ggml_log_level log_level = GGML_LOG_LEVEL_ERROR;
    std::string out;

    for (int i = 1; i < argc; i++) {
        if (strcmp(argv[i], "-a") == 0 || strcmp(argv[i], "--arch") == 0) {
            if (i + 1 < argc) {
                const std::string arch_name = argv[++i];
                arch = llm_arch_from_string(arch_name);
                if (arch == LLM_ARCH_UNKNOWN) {
                    LOG_ERR("%s: unkown LLM architecture: %s\n", __func__, arch_name.c_str());
                    return 1;
                }
            } else {
                usage(argv);
                return 1;
            }
        }
        if (strcmp(argv[i], "-s") == 0 || strcmp(argv[i], "--seed") == 0) {
            if (i + 1 < argc) {
                seed = std::stoull(argv[++i]);
            } else {
                usage(argv);
                return 1;
            }
        }
        if (strcmp(argv[i], "-v") == 0 || strcmp(argv[i], "--verbose") == 0) {
            log_level = GGML_LOG_LEVEL_INFO;
            continue;
        }
        if (strcmp(argv[i], "-o") == 0 || strcmp(argv[i], "--out") == 0) {
            if (i + 1 < argc) {
                out = argv[++i];
            } else {
                usage(argv);
                return 1;
            }
        }
    }
    printf("%s: using seed %zu\n", __func__, seed);

    try {
        if (!out.empty()) {
            return save_models(arch, seed, log_level, out);
        }
        return test_backends(arch, seed, log_level);
    } catch (const std::exception & err) {
        fprintf(stderr, "encountered runtime error: %s\n", err.what());
        return -1;
    }
}