@fugood/llama.node 1.1.10 → 1.2.0-rc.0

package/src/common.hpp

@@ -4,16 +4,9 @@
 #include "common/common.h"
 #include "common/sampling.h"
 #include "llama.h"
-#include "tools/mtmd/clip.h"
-#include "tools/mtmd/mtmd.h"
-#include "tools/mtmd/mtmd-helper.h"
 #include <memory>
-#include <mutex>
 #include <napi.h>
 #include <string>
-#include <thread>
-#include <tuple>
-#include <vector>
 
 typedef std::unique_ptr<common_sampler, decltype(&common_sampler_free)>
     LlamaCppSampling;
@@ -64,455 +57,4 @@ constexpr T get_option(const Napi::Object &options, const std::string &name,
   } else {
     return default_value;
   }
-}
-
-class LlamaSession {
-public:
-  LlamaSession(common_params params) : params_(params) {
-    llama_init_ = common_init_from_params(params);
-    tokens_.reserve(params.n_ctx);
-  }
-
-  ~LlamaSession() { dispose(); }
-
-  inline llama_context *context() { return llama_init_.context.get(); }
-
-  inline llama_model *model() { return llama_init_.model.get(); }
-
-  inline std::vector<llama_token> *tokens_ptr() { return &tokens_; }
-
-  inline void set_tokens(std::vector<llama_token> tokens) {
-    tokens_ = std::move(tokens);
-  }
-
-  inline std::vector<std::string> *mtmd_bitmap_past_hashes_ptr() {
-    return &mtmd_bitmap_past_hashes_;
-  }
-
-  inline void set_mtmd_bitmap_past_hashes(std::vector<std::string> hashes) {
-    mtmd_bitmap_past_hashes_ = std::move(hashes);
-  }
-
-  inline const common_params &params() const { return params_; }
-
-  inline std::mutex &get_mutex() { return mutex; }
-
-  // Getter for the multimodal context
-  inline mtmd_context *get_mtmd_ctx() { return _mtmd_ctx; }
-
-  // Setter for the multimodal context
-  inline void set_mtmd_ctx(mtmd_context *ctx) { _mtmd_ctx = ctx; }
-
-  void dispose() {
-    std::lock_guard<std::mutex> lock(mutex);
-    tokens_.clear();
-
-    // mtmd_ctx is owned by LlamaContext, so we don't free it here
-    _mtmd_ctx = nullptr;
-  }
-
-private:
-  common_init_result llama_init_;
-  const common_params params_;
-  std::vector<llama_token> tokens_{};
-  std::vector<std::string> mtmd_bitmap_past_hashes_{};
-  std::mutex mutex;
-  mtmd_context *_mtmd_ctx = nullptr;
-};
-
-typedef std::shared_ptr<LlamaSession> LlamaSessionPtr;
-
-static size_t common_tokens_part(const std::vector<llama_token> &a,
-                                 const std::vector<llama_token> &b) {
-  size_t i = 0;
-  while (i < a.size() && i < b.size() && a[i] == b[i]) {
-    i++;
-  }
-  return i;
-}
-
-// Computes FNV-1a hash of the data
-static std::string fnv_hash(const uint8_t *data, size_t len) {
-  const uint64_t fnv_prime = 0x100000001b3ULL;
-  uint64_t hash = 0xcbf29ce484222325ULL;
-
-  for (size_t i = 0; i < len; ++i) {
-    hash ^= data[i];
-    hash *= fnv_prime;
-  }
-  return std::to_string(hash);
-}
-
-static const std::string base64_chars = "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
-                                        "abcdefghijklmnopqrstuvwxyz"
-                                        "0123456789+/";
-
-// Base64 decoding function
-static std::vector<uint8_t> base64_decode(const std::string &encoded_string) {
-  std::vector<uint8_t> decoded;
-  int in_len = encoded_string.size();
-  int i = 0;
-  int j = 0;
-  int in_ = 0;
-  unsigned char char_array_4[4], char_array_3[3];
-
-  while (in_len-- && (encoded_string[in_] != '=')) {
-    if (isspace(encoded_string[in_])) {
-      in_++;
-      continue;
-    }
-
-    if (encoded_string[in_] == '=' ||
-        base64_chars.find(encoded_string[in_]) == std::string::npos) {
-      break;
-    }
-
-    char_array_4[i++] = encoded_string[in_];
-    in_++;
-    if (i == 4) {
-      for (i = 0; i < 4; i++) {
-        char_array_4[i] = base64_chars.find(char_array_4[i]);
-      }
-
-      char_array_3[0] =
-          (char_array_4[0] << 2) + ((char_array_4[1] & 0x30) >> 4);
-      char_array_3[1] =
-          ((char_array_4[1] & 0xf) << 4) + ((char_array_4[2] & 0x3c) >> 2);
-      char_array_3[2] = ((char_array_4[2] & 0x3) << 6) + char_array_4[3];
-
-      for (i = 0; i < 3; i++) {
-        decoded.push_back(char_array_3[i]);
-      }
-      i = 0;
-    }
-  }
-
-  if (i) {
-    for (j = i; j < 4; j++) {
-      char_array_4[j] = 0;
-    }
-
-    for (j = 0; j < 4; j++) {
-      char_array_4[j] = base64_chars.find(char_array_4[j]);
-    }
-
-    char_array_3[0] = (char_array_4[0] << 2) + ((char_array_4[1] & 0x30) >> 4);
-    char_array_3[1] =
-        ((char_array_4[1] & 0xf) << 4) + ((char_array_4[2] & 0x3c) >> 2);
-    char_array_3[2] = ((char_array_4[2] & 0x3) << 6) + char_array_4[3];
-
-    for (j = 0; j < i - 1; j++) {
-      decoded.push_back(char_array_3[j]);
-    }
-  }
-
-  return decoded;
-}
-
-struct TokenizeResult {
-  std::vector<llama_token> tokens;
-
-  bool has_media = false;
-  std::vector<std::string> bitmap_hashes;
-  std::vector<size_t> chunk_pos;       // both text and media
-  std::vector<size_t> chunk_pos_media; // media only
-  mtmd_input_chunks *chunks = nullptr;
-};
-
-static TokenizeResult
-tokenizeWithMedia(mtmd_context *mtmd_ctx, const std::string &prompt,
-                  const std::vector<std::string> &media_paths) {
-  if (mtmd_ctx == nullptr) {
-    throw std::runtime_error("Multimodal context is not initialized");
-  }
-
-  TokenizeResult result;
-  result.has_media = !media_paths.empty();
-
-  mtmd::bitmaps bitmaps;
-
-  // Load all media paths
-  for (const auto &media_path : media_paths) {
-    fprintf(
-        stdout, "[DEBUG] Loading media: %s\n",
-        media_path.substr(0, 50).c_str()); // Only log part of path for base64
-
-    // Check if it's a base64 media
-    if (media_path.compare(0, 11, "data:image/") == 0 ||
-        media_path.compare(0, 11, "data:audio/") == 0) {
-
-      // Parse base64 data
-      std::vector<std::string> parts;
-      size_t comma_pos = media_path.find(',');
-      if (comma_pos == std::string::npos) {
-        result.bitmap_hashes.clear();
-        throw std::runtime_error(
-            "Invalid base64 media format, missing comma separator");
-      }
-
-      std::string header = media_path.substr(0, comma_pos);
-      std::string base64_data = media_path.substr(comma_pos + 1);
-
-      if (header.find("base64") == std::string::npos) {
-        result.bitmap_hashes.clear();
-        throw std::runtime_error("Invalid base64 media");
-      }
-
-      // Decode base64
-      try {
-        // Decode base64 to binary
-        std::vector<uint8_t> media_data = base64_decode(base64_data);
-
-        // Load bitmap from memory buffer using direct initialization
-        mtmd::bitmap bmp(mtmd_helper_bitmap_init_from_buf(mtmd_ctx, media_data.data(),
-                                                          media_data.size()));
-        if (!bmp.ptr) {
-          bitmaps.entries.clear();
-          throw std::runtime_error("Failed to load base64 media");
-        }
-
-        // Calculate bitmap hash (for KV caching)
-        std::string hash = fnv_hash(bmp.data(), bmp.n_bytes());
-        bmp.set_id(hash.c_str());
-        bitmaps.entries.push_back(std::move(bmp));
-        result.bitmap_hashes.push_back(hash.c_str());
-      } catch (const std::exception &e) {
-        bitmaps.entries.clear();
-        throw std::runtime_error("Failed to decode base64 media");
-      }
-    } else if (media_path.compare(0, 7, "http://") == 0 ||
-               media_path.compare(0, 8, "https://") == 0) {
-      // HTTP URLs are not supported yet
-      bitmaps.entries.clear();
-      throw std::runtime_error("HTTP/HTTPS URLs are not supported yet");
-    } else {
-      // Regular file path
-      // Check if file exists
-      FILE *file = fopen(media_path.c_str(), "rb");
-      if (file == nullptr) {
-        bitmaps.entries.clear();
-        throw std::runtime_error("File does not exist or cannot be opened");
-      }
-
-      // Get file size
-      fseek(file, 0, SEEK_END);
-      long file_size = ftell(file);
-      fseek(file, 0, SEEK_SET);
-      fclose(file);
-
-      // Create bitmap directly
-      mtmd::bitmap bmp(mtmd_helper_bitmap_init_from_file(mtmd_ctx, media_path.c_str()));
-      if (!bmp.ptr) {
-        bitmaps.entries.clear();
-        throw std::runtime_error("Failed to load media");
-      }
-
-      // Calculate bitmap hash (for KV caching)
-      std::string hash = fnv_hash(bmp.data(), bmp.nx() * bmp.ny() * 3);
-      bmp.set_id(hash.c_str());
-      bitmaps.entries.push_back(std::move(bmp));
-      result.bitmap_hashes.push_back(hash.c_str());
-    }
-  }
-
-  result.chunks = mtmd_input_chunks_init();
-  if (result.chunks == nullptr) {
-    bitmaps.entries.clear();
-    throw std::runtime_error("Failed to initialize input chunks");
-  }
-
-  // Create input text
-  mtmd_input_text input_text;
-  input_text.text = prompt.c_str(); // Use the full prompt with media marker
-  input_text.add_special = true;   // Add BOS token if this is the first message
-  input_text.parse_special = true; // Parse special tokens like <__media__>
-
-  // Tokenize the text and media
-  fprintf(stdout, "[DEBUG] Tokenizing text and %zu media\n",
-          bitmaps.entries.size());
-  auto bitmaps_c_ptr = bitmaps.c_ptr();
-
-  // Cast away const for mtmd_tokenize
-  int32_t res =
-      mtmd_tokenize(const_cast<mtmd_context *>(mtmd_ctx), result.chunks,
-                    &input_text, bitmaps_c_ptr.data(), bitmaps_c_ptr.size());
-
-  if (res != 0) {
-    mtmd_input_chunks_free(result.chunks);
-    bitmaps.entries.clear();
-    throw std::runtime_error("Failed to tokenize text and media");
-  }
-
-  // Log chunk information
-  size_t num_chunks = mtmd_input_chunks_size(result.chunks);
-  fprintf(stdout, "[DEBUG] Tokenization successful: num_chunks=%zu\n",
-          num_chunks);
-
-  // Track the total number of tokens (both text and media)
-  size_t total_token_count = 0;
-
-  // chunk pos
-  for (size_t i = 0; i < num_chunks; i++) {
-    result.chunk_pos.push_back(total_token_count);
-
-    const mtmd_input_chunk *chunk = mtmd_input_chunks_get(result.chunks, i);
-    mtmd_input_chunk_type chunk_type = mtmd_input_chunk_get_type(chunk);
-
-    if (chunk_type == MTMD_INPUT_CHUNK_TYPE_TEXT) {
-      size_t n_tokens;
-      const llama_token *tokens =
-          mtmd_input_chunk_get_tokens_text(chunk, &n_tokens);
-
-      result.tokens.insert(result.tokens.end(), tokens, tokens + n_tokens);
-      total_token_count += n_tokens;
-    } else if (chunk_type == MTMD_INPUT_CHUNK_TYPE_IMAGE ||
-               chunk_type == MTMD_INPUT_CHUNK_TYPE_AUDIO) {
-      result.chunk_pos_media.push_back(total_token_count);
-
-      size_t n_tokens = mtmd_input_chunk_get_n_tokens(chunk);
-      size_t n_pos = mtmd_input_chunk_get_n_pos(chunk);
-      fprintf(stdout, "[DEBUG] Chunk %zu: type=%s, n_tokens=%zu, n_pos=%zu\n",
-              i, chunk_type == MTMD_INPUT_CHUNK_TYPE_IMAGE ? "IMAGE" : "AUDIO",
-              n_tokens, n_pos);
-
-      for (size_t j = 0; j < n_pos; j++) {
-        result.tokens.push_back(LLAMA_TOKEN_NULL);
-      }
-      total_token_count += n_pos;
-    }
-  }
-
-  bitmaps.entries.clear();
-
-  return result;
-}
-
-// Process media and add them to the tokenized input
-static llama_pos
-processMediaPrompt(llama_context *ctx, mtmd_context *mtmd_ctx,
-                   LlamaSessionPtr sess, const common_params &params,
-                   const std::vector<std::string> &media_paths) {
-  if (mtmd_ctx == nullptr) {
-    throw std::runtime_error("Multimodal context is not initialized");
-  }
-
-  // Multimodal path
-  std::string full_prompt = params.prompt;
-  auto default_media_marker = mtmd_default_marker();
-  // Add media marker if it doesn't already exist
-  if (full_prompt.find(default_media_marker) == std::string::npos) {
-    full_prompt += " ";
-    full_prompt += default_media_marker;
-  }
-
-  auto result = tokenizeWithMedia(mtmd_ctx, full_prompt, media_paths);
-
-  auto all_tokens = result.tokens;
-  auto chunks = result.chunks;
-  auto chunk_pos = result.chunk_pos;
-  auto chunk_pos_media = result.chunk_pos_media;
-  auto bitmap_hashes = result.bitmap_hashes;
-
-  llama_pos n_past = common_tokens_part(*sess->tokens_ptr(), all_tokens);
-
-  llama_pos new_n_past = n_past;
-
-  // Adjust n_past to position of the text chunk
-  // TODO: Edit the text chunk to remove the tokens before n_past to speed up
-  // need to update the mtmd api
-  auto adjusted_n_past = -1;
-  for (size_t i = 0; i < chunk_pos.size(); i++) {
-    if (n_past < chunk_pos[i]) {
-      break;
-    }
-    bool is_end = i + 1 == chunk_pos.size();
-    if (chunk_pos[i] < n_past && (!is_end && chunk_pos[i + 1] > n_past)
-        // is_end & n_past < total_token_count:
-        // don't need to adjust and it will skip eval_chunk_single, let
-        // nextToken() to finish the job
-    ) {
-      adjusted_n_past = chunk_pos[i];
-    }
-  }
-  if (adjusted_n_past != -1) {
-    n_past = adjusted_n_past;
-    new_n_past = n_past;
-    fprintf(stdout, "[DEBUG] Adjusted n_past to %d\n", n_past);
-  }
-
-  // Compare bitmap hashes, if they are not the same, backtrack n_past to the
-  // position of the first mismatch
-  auto mtmd_bitmap_past_hashes = sess->mtmd_bitmap_past_hashes_ptr();
-  if (mtmd_bitmap_past_hashes->size() > 0) {
-    for (size_t i = 0; i < bitmap_hashes.size(); i++) {
-      auto pos = chunk_pos_media[i];
-      if (n_past < pos) {
-        break;
-      }
-      if (i >= mtmd_bitmap_past_hashes->size()) {
-        break;
-      }
-      if (bitmap_hashes[i] != (*mtmd_bitmap_past_hashes)[i]) {
-        n_past = chunk_pos_media[i];
-        new_n_past = n_past;
-        break;
-      }
-    }
-  }
-
-  // Clear all KV cache entries after position n_past
-  auto * kv = llama_get_memory(ctx);
-  bool clear_result = llama_memory_seq_rm(kv, 0, n_past, -1);
-  if (!clear_result) {
-    fprintf(stdout, "[DEBUG] llama_memory_seq_rm failed (likely using a non-Transformer model)! Trying full clear...");
-    llama_memory_clear(kv, false);
-    n_past = 0;
-    new_n_past = n_past;
-  }
-
-  size_t num_chunks = mtmd_input_chunks_size(chunks);
-
-  for (size_t i = 0; i < chunk_pos.size(); i++) {
-    fprintf(stdout, "[DEBUG] Evaluating chunk %zu: n_past=%d, chunk_pos=%zu\n",
-            i, n_past, chunk_pos[i]);
-
-    // Process chunk only if it's after the current n_past
-    if (chunk_pos[i] >= new_n_past) {
-      bool chunk_logits_last = (i == num_chunks - 1);
-      auto chunk = mtmd_input_chunks_get(chunks, i);
-
-      // Cast away const for mtmd_helper_eval_chunk_single
-      int32_t res = mtmd_helper_eval_chunk_single(
-          const_cast<mtmd_context *>(mtmd_ctx), ctx, chunk, n_past, 0,
-          params.n_batch, // batch size
-          chunk_logits_last, &new_n_past);
-
-      if (res != 0) {
-        mtmd_input_chunks_free(chunks);
-        throw std::runtime_error("Failed to process chunk");
-      }
-      n_past = new_n_past;
-    }
-  }
-
-  if (n_past == all_tokens.size() && n_past > 0 &&
-      all_tokens[n_past - 1] != LLAMA_TOKEN_NULL) {
-    // we have to evaluate at least 1 token to generate logits.
-    n_past--;
-  }
-
-  // Update sampling context to process token sequences
-  for (auto &token : all_tokens) {
-    if (token == LLAMA_TOKEN_NULL) {
-      continue;
-    }
-  }
-  // Set the tokens
-  sess->set_tokens(std::move(all_tokens));
-
-  sess->set_mtmd_bitmap_past_hashes(bitmap_hashes);
-
-  // Clean up media resources
-  mtmd_input_chunks_free(chunks);
-  return n_past;
 }

package/src/llama.cpp/common/arg.cpp

@@ -1106,7 +1106,7 @@ static void common_params_print_completion(common_params_context & ctx_arg) {
     printf("\"\n\n");
 
     printf("    case \"$prev\" in\n");
-    printf("        --model)\n");
+    printf("        --model|-m)\n");
     printf("            COMPREPLY=( $(compgen -f -X '!*.gguf' -- \"$cur\") $(compgen -d -- \"$cur\") )\n");
     printf("            return 0\n");
     printf("            ;;\n");
@@ -1263,6 +1263,18 @@ static std::string list_builtin_chat_templates() {
     return msg.str();
 }
 
+static bool is_truthy(const std::string & value) {
+    return value == "on" || value == "enabled" || value == "1";
+}
+
+static bool is_falsey(const std::string & value) {
+    return value == "off" || value == "disabled" || value == "0";
+}
+
+static bool is_autoy(const std::string & value) {
+    return value == "auto" || value == "-1";
+}
+
 common_params_context common_params_parser_init(common_params & params, llama_example ex, void(*print_usage)(int, char **)) {
     // load dynamic backends
     ggml_backend_load_all();
@@ -1544,13 +1556,21 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.n_chunks = value;
         }
     ).set_examples({LLAMA_EXAMPLE_IMATRIX, LLAMA_EXAMPLE_PERPLEXITY, LLAMA_EXAMPLE_RETRIEVAL}));
-    add_opt(common_arg(
-        {"-fa", "--flash-attn"},
-        string_format("enable Flash Attention (default: %s)", params.flash_attn ? "enabled" : "disabled"),
-        [](common_params & params) {
-            params.flash_attn = true;
-        }
-    ).set_env("LLAMA_ARG_FLASH_ATTN"));
+    add_opt(common_arg({ "-fa", "--flash-attn" }, "[on|off|auto]",
+                       string_format("set Flash Attention use ('on', 'off', or 'auto', default: '%s')",
+                                     llama_flash_attn_type_name(params.flash_attn_type)),
+                       [](common_params & params, const std::string & value) {
+                           if (is_truthy(value)) {
+                               params.flash_attn_type = LLAMA_FLASH_ATTN_TYPE_ENABLED;
+                           } else if (is_falsey(value)) {
+                               params.flash_attn_type = LLAMA_FLASH_ATTN_TYPE_DISABLED;
+                           } else if (is_autoy(value)) {
+                               params.flash_attn_type = LLAMA_FLASH_ATTN_TYPE_AUTO;
+                           } else {
+                               throw std::runtime_error(
+                                   string_format("error: unkown value for --flash-attn: '%s'\n", value.c_str()));
+                           }
+                       }).set_env("LLAMA_ARG_FLASH_ATTN"));
     add_opt(common_arg(
         {"-p", "--prompt"}, "PROMPT",
         "prompt to start generation with; for system message, use -sys",
@@ -2458,7 +2478,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
     ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_N_CPU_MOE_DRAFT"));
     add_opt(common_arg(
         {"-ngl", "--gpu-layers", "--n-gpu-layers"}, "N",
-        "number of layers to store in VRAM",
+        string_format("max. number of layers to store in VRAM (default: %d)", params.n_gpu_layers),
         [](common_params & params, int value) {
             params.n_gpu_layers = value;
             if (!llama_supports_gpu_offload()) {
@@ -2555,7 +2575,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         {"--lora"}, "FNAME",
         "path to LoRA adapter (can be repeated to use multiple adapters)",
         [](common_params & params, const std::string & value) {
-            params.lora_adapters.push_back({ std::string(value), 1.0, nullptr });
+            params.lora_adapters.push_back({ std::string(value), 1.0, "", "", nullptr });
         }
         // we define this arg on both COMMON and EXPORT_LORA, so when showing help message of export-lora, it will be categorized as "example-specific" arg
     ).set_examples({LLAMA_EXAMPLE_COMMON, LLAMA_EXAMPLE_EXPORT_LORA}));
@@ -2563,7 +2583,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         {"--lora-scaled"}, "FNAME", "SCALE",
         "path to LoRA adapter with user defined scaling (can be repeated to use multiple adapters)",
         [](common_params & params, const std::string & fname, const std::string & scale) {
-            params.lora_adapters.push_back({ fname, std::stof(scale), nullptr });
+            params.lora_adapters.push_back({ fname, std::stof(scale), "", "", nullptr });
         }
         // we define this arg on both COMMON and EXPORT_LORA, so when showing help message of export-lora, it will be categorized as "example-specific" arg
     ).set_examples({LLAMA_EXAMPLE_COMMON, LLAMA_EXAMPLE_EXPORT_LORA}));
@@ -2954,13 +2974,6 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.endpoint_metrics = true;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_ENDPOINT_METRICS"));
-    add_opt(common_arg(
-        {"--slots"},
-        string_format("enable slots monitoring endpoint (default: %s)", params.endpoint_slots ? "enabled" : "disabled"),
-        [](common_params & params) {
-            params.endpoint_slots = true;
-        }
-    ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_ENDPOINT_SLOTS"));
     add_opt(common_arg(
         {"--props"},
         string_format("enable changing global properties via POST /props (default: %s)", params.endpoint_props ? "enabled" : "disabled"),
@@ -2968,6 +2981,13 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.endpoint_props = true;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_ENDPOINT_PROPS"));
+    add_opt(common_arg(
+        {"--slots"},
+        string_format("enable slots monitoring endpoint (default: %s)", params.endpoint_slots ? "enabled" : "disabled"),
+        [](common_params & params) {
+            params.endpoint_slots = true;
+        }
+    ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_ENDPOINT_SLOTS"));
     add_opt(common_arg(
         {"--no-slots"},
         "disables slots monitoring endpoint",
@@ -3126,13 +3146,21 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             common_log_set_file(common_log_main(), value.c_str());
         }
     ));
-    add_opt(common_arg(
-        {"--log-colors"},
-        "Enable colored logging",
-        [](common_params &) {
-            common_log_set_colors(common_log_main(), true);
-        }
-    ).set_env("LLAMA_LOG_COLORS"));
+    add_opt(common_arg({ "--log-colors" }, "[on|off|auto]",
+                       "Set colored logging ('on', 'off', or 'auto', default: 'auto')\n"
+                       "'auto' enables colors when output is to a terminal",
+                       [](common_params &, const std::string & value) {
+                           if (is_truthy(value)) {
+                               common_log_set_colors(common_log_main(), LOG_COLORS_ENABLED);
+                           } else if (is_falsey(value)) {
+                               common_log_set_colors(common_log_main(), LOG_COLORS_DISABLED);
+                           } else if (is_autoy(value)) {
+                               common_log_set_colors(common_log_main(), LOG_COLORS_AUTO);
+                           } else {
+                               throw std::invalid_argument(
+                                   string_format("error: unkown value for --log-colors: '%s'\n", value.c_str()));
+                           }
+                       }).set_env("LLAMA_LOG_COLORS"));
     add_opt(common_arg(
         {"-v", "--verbose", "--log-verbose"},
         "Set verbosity level to infinity (i.e. log all messages, useful for debugging)",
@@ -3459,8 +3487,6 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.model.hf_repo = "ggml-org/Qwen2.5-Coder-1.5B-Q8_0-GGUF";
             params.model.hf_file = "qwen2.5-coder-1.5b-q8_0.gguf";
             params.port = 8012;
-            params.n_gpu_layers = 99;
-            params.flash_attn = true;
             params.n_ubatch = 1024;
             params.n_batch = 1024;
             params.n_ctx = 0;
@@ -3475,8 +3501,6 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.model.hf_repo = "ggml-org/Qwen2.5-Coder-3B-Q8_0-GGUF";
             params.model.hf_file = "qwen2.5-coder-3b-q8_0.gguf";
             params.port = 8012;
-            params.n_gpu_layers = 99;
-            params.flash_attn = true;
             params.n_ubatch = 1024;
             params.n_batch = 1024;
             params.n_ctx = 0;
@@ -3491,8 +3515,6 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.model.hf_repo = "ggml-org/Qwen2.5-Coder-7B-Q8_0-GGUF";
             params.model.hf_file = "qwen2.5-coder-7b-q8_0.gguf";
             params.port = 8012;
-            params.n_gpu_layers = 99;
-            params.flash_attn = true;
             params.n_ubatch = 1024;
             params.n_batch = 1024;
             params.n_ctx = 0;
@@ -3508,10 +3530,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.model.hf_file = "qwen2.5-coder-7b-q8_0.gguf";
             params.speculative.model.hf_repo = "ggml-org/Qwen2.5-Coder-0.5B-Q8_0-GGUF";
             params.speculative.model.hf_file = "qwen2.5-coder-0.5b-q8_0.gguf";
-            params.speculative.n_gpu_layers = 99;
             params.port = 8012;
-            params.n_gpu_layers = 99;
-            params.flash_attn = true;
             params.n_ubatch = 1024;
             params.n_batch = 1024;
             params.n_ctx = 0;
@@ -3527,10 +3546,21 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.model.hf_file = "qwen2.5-coder-14b-q8_0.gguf";
             params.speculative.model.hf_repo = "ggml-org/Qwen2.5-Coder-0.5B-Q8_0-GGUF";
             params.speculative.model.hf_file = "qwen2.5-coder-0.5b-q8_0.gguf";
-            params.speculative.n_gpu_layers = 99;
             params.port = 8012;
-            params.n_gpu_layers = 99;
-            params.flash_attn = true;
+            params.n_ubatch = 1024;
+            params.n_batch = 1024;
+            params.n_ctx = 0;
+            params.n_cache_reuse = 256;
+        }
+    ).set_examples({LLAMA_EXAMPLE_SERVER}));
+
+    add_opt(common_arg(
+        {"--fim-qwen-30b-default"},
+        string_format("use default Qwen 3 Coder 30B A3B Instruct (note: can download weights from the internet)"),
+        [](common_params & params) {
+            params.model.hf_repo = "ggml-org/Qwen3-Coder-30B-A3B-Instruct-Q8_0-GGUF";
+            params.model.hf_file = "qwen3-coder-30b-a3b-instruct-q8_0.gguf";
+            params.port = 8012;
             params.n_ubatch = 1024;
             params.n_batch = 1024;
             params.n_ctx = 0;