(本来要发到远景的，但是快写完的时候才想起没有验证手机号它不让发帖，干脆发到小众，反正这也是软件相关的话题。)
最近 llama.cpp 的服务器命令行加入了对图像的支持（目前仅能图生文），借这个机会讲一下怎么让 llm 跑在 AMD GPU 上(仅限黑苹果)。下面的方法需要你会使用终端执行命令行。
安装步骤
brew install libomp vulkan-headers glslang molten-vk shaderc vulkan-loader. 解释：使用 homebrew 安装如命令行所示的几个包，后面编译需要。如果你之前安装过这几个包，仍然还是需要执行这条命令，将它们更新到最新版本。git clone https://github.com/KhronosGroup/MoltenVK.git. 请自行克隆到想要的目录。cd MoltenVK && ./fetchDependencies --macos && make macoscp Package/Release/MoltenVK/dynamic/dylib/macOS/libMoltenVK.dylib /usr/local/Cellar/molten-vk/1.3.0/lib/libMoltenVK.dylib- 说明：步骤 2-4 是编译安装MoltenVK.
git clone https://github.com/ggml-org/llama.cpp.git请自行克隆到想要的目录。cd llama.cpp.gitgit reset --hard 33eff402408. 说明：最新的llama.cpp编译出来会跑不起来(写本文时最新的commit是62d4250e52)，用33eff402408则没有问题。cmake -B build -DGGML_METAL=OFF -DGGML_VULKAN=ON -DBUILD_SHARED_LIBS=OFFcmake --build build --config Release --target install -j8- 说明：步骤 6-10 是安装 llama.cpp.
如果以上步骤没有出错的话，会在系统上安装以 llama- 开头的一众命令，如下：
$ ls /usr/local/bin/llama-*
/usr/local/bin/llama-batched
/usr/local/bin/llama-batched-bench
/usr/local/bin/llama-bench
/usr/local/bin/llama-cli
/usr/local/bin/llama-convert-llama2c-to-ggml
/usr/local/bin/llama-cvector-generator
/usr/local/bin/llama-embedding
/usr/local/bin/llama-eval-callback
/usr/local/bin/llama-export-lora
/usr/local/bin/llama-gen-docs
/usr/local/bin/llama-gguf
/usr/local/bin/llama-gguf-hash
/usr/local/bin/llama-gguf-split
/usr/local/bin/llama-gritlm
/usr/local/bin/llama-imatrix
/usr/local/bin/llama-infill
/usr/local/bin/llama-llava-clip-quantize-cli
/usr/local/bin/llama-lookahead
/usr/local/bin/llama-lookup
/usr/local/bin/llama-lookup-create
/usr/local/bin/llama-lookup-merge
/usr/local/bin/llama-lookup-stats
/usr/local/bin/llama-mtmd-cli
/usr/local/bin/llama-parallel
/usr/local/bin/llama-passkey
/usr/local/bin/llama-perplexity
/usr/local/bin/llama-quantize
/usr/local/bin/llama-retrieval
/usr/local/bin/llama-run
/usr/local/bin/llama-save-load-state
/usr/local/bin/llama-server
/usr/local/bin/llama-simple
/usr/local/bin/llama-simple-chat
/usr/local/bin/llama-speculative
/usr/local/bin/llama-speculative-simple
/usr/local/bin/llama-tokenize
/usr/local/bin/llama-tts
演示
下面是以 gemma3 演示。执行如下命令：
llama-server -m gemma-3-12b-it-Q8_0.gguf --mmproj mmproj-model-f16.gguf --n-gpu-layers 100 -fa -n 1024
这两个文件 gemma-3-12b-it-Q8_0.gguf, mmproj-model-f16.gguf 可在 hugging-face 下载，也可在 hf-mirror.com 下载，关键字 ggml-org/gemma3.
成功之后，会显示以下信息：
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = AMD Radeon RX 6800 XT (MoltenVK) | uma: 0 | fp16: 1 | warp size: 64 | shared memory: 65536 | int dot: 0 | matrix cores: none
build: 5331 (33eff402) with Apple clang version 16.0.0 (clang-1600.0.26.6) for x86_64-apple-darwin24.4.0
system info: n_threads = 8, n_threads_batch = 8, total_threads = 16
system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | LLAMAFILE = 1 | ACCELERATE = 1 | AARCH64_REPACK = 1 |
main: binding port with default address family
main: HTTP server is listening, hostname: 127.0.0.1, port: 8080, http threads: 15
main: loading model
srv load_model: loading model '/Volumes/D/ai/models/gemma-3/gemma-3-12b-it-Q8_0.gguf'
llama_model_load_from_file_impl: using device Vulkan0 (AMD Radeon RX 6800 XT) - 16368 MiB free
llama_model_loader: loaded meta data with 40 key-value pairs and 626 tensors from /Volumes/D/ai/models/gemma-3/gemma-3-12b-it-Q8_0.gguf (version GGUF V3 (latest))
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
llama_model_loader: - kv 0: general.architecture str = gemma3
llama_model_loader: - kv 1: general.type str = model
llama_model_loader: - kv 2: general.name str = Gemma 3 12b It
llama_model_loader: - kv 3: general.finetune str = it
llama_model_loader: - kv 4: general.basename str = gemma-3
llama_model_loader: - kv 5: general.size_label str = 12B
llama_model_loader: - kv 6: general.license str = gemma
llama_model_loader: - kv 7: general.base_model.count u32 = 1
llama_model_loader: - kv 8: general.base_model.0.name str = Gemma 3 12b Pt
llama_model_loader: - kv 9: general.base_model.0.organization str = Google
llama_model_loader: - kv 10: general.base_model.0.repo_url str = https://huggingface.co/google/gemma-3...
llama_model_loader: - kv 11: general.tags arr[str,1] = ["image-text-to-text"]
llama_model_loader: - kv 12: gemma3.context_length u32 = 131072
llama_model_loader: - kv 13: gemma3.embedding_length u32 = 3840
llama_model_loader: - kv 14: gemma3.block_count u32 = 48
llama_model_loader: - kv 15: gemma3.feed_forward_length u32 = 15360
llama_model_loader: - kv 16: gemma3.attention.head_count u32 = 16
llama_model_loader: - kv 17: gemma3.attention.layer_norm_rms_epsilon f32 = 0.000001
llama_model_loader: - kv 18: gemma3.attention.key_length u32 = 256
llama_model_loader: - kv 19: gemma3.attention.value_length u32 = 256
llama_model_loader: - kv 20: general.file_type u32 = 7
llama_model_loader: - kv 21: gemma3.rope.freq_base f32 = 1000000.000000
llama_model_loader: - kv 22: gemma3.attention.sliding_window u32 = 1024
llama_model_loader: - kv 23: gemma3.attention.head_count_kv u32 = 8
llama_model_loader: - kv 24: gemma3.rope.scaling.type str = linear
llama_model_loader: - kv 25: gemma3.rope.scaling.factor f32 = 8.000000
llama_model_loader: - kv 26: tokenizer.ggml.model str = llama
llama_model_loader: - kv 27: tokenizer.ggml.pre str = default
llama_model_loader: - kv 28: tokenizer.ggml.tokens arr[str,262144] = ["<pad>", "<eos>", "<bos>", "<unk>", ...
llama_model_loader: - kv 29: tokenizer.ggml.scores arr[f32,262144] = [-1000.000000, -1000.000000, -1000.00...
llama_model_loader: - kv 30: tokenizer.ggml.token_type arr[i32,262144] = [3, 3, 3, 3, 3, 4, 3, 3, 3, 3, 3, 3, ...
llama_model_loader: - kv 31: tokenizer.ggml.bos_token_id u32 = 2
llama_model_loader: - kv 32: tokenizer.ggml.eos_token_id u32 = 1
llama_model_loader: - kv 33: tokenizer.ggml.unknown_token_id u32 = 3
llama_model_loader: - kv 34: tokenizer.ggml.padding_token_id u32 = 0
llama_model_loader: - kv 35: tokenizer.ggml.add_bos_token bool = true
llama_model_loader: - kv 36: tokenizer.ggml.add_eos_token bool = false
llama_model_loader: - kv 37: tokenizer.chat_template str = {{ bos_token }}\n{%- if messages[0]['r...
llama_model_loader: - kv 38: tokenizer.ggml.add_space_prefix bool = false
llama_model_loader: - kv 39: general.quantization_version u32 = 2
llama_model_loader: - type f32: 289 tensors
llama_model_loader: - type q8_0: 337 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type = Q8_0
print_info: file size = 11.64 GiB (8.50 BPW)
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
load: special tokens cache size = 6414
load: token to piece cache size = 1.9446 MB
print_info: arch = gemma3
print_info: vocab_only = 0
print_info: n_ctx_train = 131072
print_info: n_embd = 3840
print_info: n_layer = 48
print_info: n_head = 16
print_info: n_head_kv = 8
print_info: n_rot = 256
print_info: n_swa = 1024
print_info: n_swa_pattern = 6
print_info: n_embd_head_k = 256
print_info: n_embd_head_v = 256
print_info: n_gqa = 2
print_info: n_embd_k_gqa = 2048
print_info: n_embd_v_gqa = 2048
print_info: f_norm_eps = 0.0e+00
print_info: f_norm_rms_eps = 1.0e-06
print_info: f_clamp_kqv = 0.0e+00
print_info: f_max_alibi_bias = 0.0e+00
print_info: f_logit_scale = 0.0e+00
print_info: f_attn_scale = 6.2e-02
print_info: n_ff = 15360
print_info: n_expert = 0
print_info: n_expert_used = 0
print_info: causal attn = 1
print_info: pooling type = 0
print_info: rope type = 2
print_info: rope scaling = linear
print_info: freq_base_train = 1000000.0
print_info: freq_scale_train = 0.125
print_info: n_ctx_orig_yarn = 131072
print_info: rope_finetuned = unknown
print_info: ssm_d_conv = 0
print_info: ssm_d_inner = 0
print_info: ssm_d_state = 0
print_info: ssm_dt_rank = 0
print_info: ssm_dt_b_c_rms = 0
print_info: model type = 12B
print_info: model params = 11.77 B
print_info: general.name = Gemma 3 12b It
print_info: vocab type = SPM
print_info: n_vocab = 262144
print_info: n_merges = 0
print_info: BOS token = 2 '<bos>'
print_info: EOS token = 1 '<eos>'
print_info: EOT token = 106 '<end_of_turn>'
print_info: UNK token = 3 '<unk>'
print_info: PAD token = 0 '<pad>'
print_info: LF token = 248 '<0x0A>'
print_info: EOG token = 1 '<eos>'
print_info: EOG token = 106 '<end_of_turn>'
print_info: max token length = 48
load_tensors: loading model tensors, this can take a while... (mmap = true)
load_tensors: offloading 48 repeating layers to GPU
load_tensors: offloading output layer to GPU
load_tensors: offloaded 49/49 layers to GPU
load_tensors: CPU_Mapped model buffer size = 1020.00 MiB
load_tensors: Vulkan0 model buffer size = 11924.17 MiB
.......................................................................................
llama_context: constructing llama_context
llama_context: n_seq_max = 1
llama_context: n_ctx = 4096
llama_context: n_ctx_per_seq = 4096
llama_context: n_batch = 2048
llama_context: n_ubatch = 512
llama_context: causal_attn = 1
llama_context: flash_attn = 1
llama_context: freq_base = 1000000.0
llama_context: freq_scale = 0.125
llama_context: n_ctx_per_seq (4096) < n_ctx_train (131072) -- the full capacity of the model will not be utilized
llama_context: Vulkan_Host output buffer size = 1.00 MiB
llama_kv_cache_unified: kv_size = 4096, type_k = 'f16', type_v = 'f16', n_layer = 48, can_shift = 1, padding = 256
llama_kv_cache_unified: Vulkan0 KV buffer size = 1536.00 MiB
llama_kv_cache_unified: KV self size = 1536.00 MiB, K (f16): 768.00 MiB, V (f16): 768.00 MiB
llama_context: Vulkan0 compute buffer size = 519.50 MiB
llama_context: Vulkan_Host compute buffer size = 60.01 MiB
llama_context: graph nodes = 1833
llama_context: graph splits = 98
common_init_from_params: setting dry_penalty_last_n to ctx_size = 4096
common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
clip_ctx: CLIP using Vulkan0 backend
clip_model_loader: model name:
clip_model_loader: description:
clip_model_loader: GGUF version: 3
clip_model_loader: alignment: 32
clip_model_loader: n_tensors: 439
clip_model_loader: n_kv: 16
load_hparams: projector: gemma3
load_hparams: n_embd: 1152
load_hparams: n_head: 16
load_hparams: n_ff: 4304
load_hparams: n_layer: 27
load_hparams: projection_dim: 3840
load_hparams: image_size: 896
load_hparams: patch_size: 14
load_hparams: has_llava_proj: 0
load_hparams: minicpmv_version: 0
load_hparams: proj_scale_factor: 4
load_hparams: n_wa_pattern: 0
load_hparams: ffn_op: gelu
load_hparams: model size: 814.60 MiB
load_hparams: metadata size: 0.15 MiB
alloc_compute_meta: Vulkan0 compute buffer size = 1132.00 MiB
alloc_compute_meta: CPU compute buffer size = 9.19 MiB
srv load_model: loaded multimodal model, '/Volumes/D/ai/models/gemma-3/mmproj-model-f16.gguf'
srv load_model: ctx_shift is not supported by multimodal, it will be disabled
srv init: initializing slots, n_slots = 1
slot init: id 0 | task -1 | new slot n_ctx_slot = 4096
main: model loaded
main: chat template, chat_template: {{ bos_token }}
{%- if messages[0]['role'] == 'system' -%}
{%- if messages[0]['content'] is string -%}
{%- set first_user_prefix = messages[0]['content'] + '
' -%}
{%- else -%}
{%- set first_user_prefix = messages[0]['content'][0]['text'] + '
' -%}
{%- endif -%}
{%- set loop_messages = messages[1:] -%}
{%- else -%}
{%- set first_user_prefix = "" -%}
{%- set loop_messages = messages -%}
{%- endif -%}
{%- for message in loop_messages -%}
{%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) -%}
{{ raise_exception("Conversation roles must alternate user/assistant/user/assistant/...") }}
{%- endif -%}
{%- if (message['role'] == 'assistant') -%}
{%- set role = "model" -%}
{%- else -%}
{%- set role = message['role'] -%}
{%- endif -%}
{{ '<start_of_turn>' + role + '
' + (first_user_prefix if loop.first else "") }}
{%- if message['content'] is string -%}
{{ message['content'] | trim }}
{%- elif message['content'] is iterable -%}
{%- for item in message['content'] -%}
{%- if item['type'] == 'image' -%}
{{ '<start_of_image>' }}
{%- elif item['type'] == 'text' -%}
{{ item['text'] | trim }}
{%- endif -%}
{%- endfor -%}
{%- else -%}
{{ raise_exception("Invalid content type") }}
{%- endif -%}
{{ '<end_of_turn>
' }}
{%- endfor -%}
{%- if add_generation_prompt -%}
{{'<start_of_turn>model
'}}
{%- endif -%}
, example_format: '<start_of_turn>user
You are a helpful assistant
Hello<end_of_turn>
<start_of_turn>model
Hi there<end_of_turn>
<start_of_turn>user
How are you?<end_of_turn>
<start_of_turn>model
'
main: server is listening on http://127.0.0.1:8080 - starting the main loop
srv update_slots: all slots are idle
最后倒数第二行可以看到，服务器监听在 http://127.0.0.1:8080. 在浏览器中打开这个地址，可以看到在文本输入框的右边有个按钮可以上传图片。
演示图片就不上传了，太麻烦。