5. 手动加载模型文件
在上一章我们已经找到了下载到本地的大模型文件，现在我们尝试以手工命令行方式加载。
5.1 runner命令行加载
之前的例子我们是通过Ollama应用来使用大模型，其底层是通过一个"runner"程序启动的推理引擎。“runner”代码在Ollama主仓库的cmd/runner目录下，是Go语言和C/C++混合编写的程序。
现在我们尝试从命令行启动，进入Ollama仓库的cmd/runner子目录，执行以下命令：
$ go run cmd/runner/main.go -h
Runner usage
-batch-size int
Batch size (default 512)
-ctx-size int
Context (or KV cache) size (default 2048)
-flash-attn
Enable flash attention
-kv-cache-type string
quantization type for KV cache (default: f16)
-lora value
Path to lora layer file (can be specified multiple times)
-main-gpu int
Main GPU
-mlock
force system to keep model in RAM rather than swapping or compressing
-mmproj string
Path to projector binary file
-model string
Path to model binary file
-multiuser-cache
optimize input cache algorithm for multiple users
-n-gpu-layers int
Number of layers to offload to GPU
-no-mmap
do not memory-map model (slower load but may reduce pageouts if not using mlock)
-parallel int
Number of sequences to handle simultaneously (default 1)
-port int
Port to expose the server on (default 8080)
-tensor-split string
fraction of the model to offload to each GPU, comma-separated list of proportions
-threads int
Number of threads to use during generation (default 4)
-verbose
verbose output (default: disabled)
$
其中最重要是-model参数指定大模型文件。通过以下命令运行deepseek-r1:1.5b大模型：
$ go run cmd/runner/main.go -model=$HOME/.ollama/models/blobs/sha256-aabd4debf0c8f08881923f2c25fc0fdeed24435271c2b3e92c4af36704040dbc
time=2025-03-05T09:58:06.228+08:00 level=INFO source=runner.go:932 msg="starting go runner"
time=2025-03-05T09:58:06.231+08:00 level=INFO source=runner.go:935 msg=system info="CPU : SSE3 = 1 | SSSE3 = 1 | LLAMAFILE = 1 | cgo(clang)" threads=4
time=2025-03-05T09:58:06.232+08:00 level=INFO source=.:0 msg="Server listening on 127.0.0.1:8080"
llama_model_loader: loaded meta data with 26 key-value pairs and 339 tensors from /Users/chai/.ollama/models/blobs/sha256-aabd4debf0c8f08881923f2c25fc0fdeed24435271c2b3e92c4af36704040dbc (version GGUF V3 (latest))
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
llama_model_loader: - kv 0: general.architecture str = qwen2
llama_model_loader: - kv 1: general.type str = model
llama_model_loader: - kv 2: general.name str = DeepSeek R1 Distill Qwen 1.5B
llama_model_loader: - kv 3: general.basename str = DeepSeek-R1-Distill-Qwen
llama_model_loader: - kv 4: general.size_label str = 1.5B
llama_model_loader: - kv 5: qwen2.block_count u32 = 28
llama_model_loader: - kv 6: qwen2.context_length u32 = 131072
llama_model_loader: - kv 7: qwen2.embedding_length u32 = 1536
llama_model_loader: - kv 8: qwen2.feed_forward_length u32 = 8960
llama_model_loader: - kv 9: qwen2.attention.head_count u32 = 12
llama_model_loader: - kv 10: qwen2.attention.head_count_kv u32 = 2
llama_model_loader: - kv 11: qwen2.rope.freq_base f32 = 10000.000000
llama_model_loader: - kv 12: qwen2.attention.layer_norm_rms_epsilon f32 = 0.000001
llama_model_loader: - kv 13: general.file_type u32 = 15
llama_model_loader: - kv 14: tokenizer.ggml.model str = gpt2
llama_model_loader: - kv 15: tokenizer.ggml.pre str = qwen2
llama_model_loader: - kv 16: tokenizer.ggml.tokens arr[str,151936] = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv 17: tokenizer.ggml.token_type arr[i32,151936] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv 18: tokenizer.ggml.merges arr[str,151387] = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv 19: tokenizer.ggml.bos_token_id u32 = 151646
llama_model_loader: - kv 20: tokenizer.ggml.eos_token_id u32 = 151643
llama_model_loader: - kv 21: tokenizer.ggml.padding_token_id u32 = 151643
llama_model_loader: - kv 22: tokenizer.ggml.add_bos_token bool = true
llama_model_loader: - kv 23: tokenizer.ggml.add_eos_token bool = false
llama_model_loader: - kv 24: tokenizer.chat_template str = {% if not add_generation_prompt is de...
llama_model_loader: - kv 25: general.quantization_version u32 = 2
llama_model_loader: - type f32: 141 tensors
llama_model_loader: - type q4_K: 169 tensors
llama_model_loader: - type q6_K: 29 tensors
llm_load_vocab: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
llm_load_vocab: special tokens cache size = 22
llm_load_vocab: token to piece cache size = 0.9310 MB
llm_load_print_meta: format = GGUF V3 (latest)
llm_load_print_meta: arch = qwen2
llm_load_print_meta: vocab type = BPE
llm_load_print_meta: n_vocab = 151936
llm_load_print_meta: n_merges = 151387
llm_load_print_meta: vocab_only = 0
llm_load_print_meta: n_ctx_train = 131072
llm_load_print_meta: n_embd = 1536
llm_load_print_meta: n_layer = 28
llm_load_print_meta: n_head = 12
llm_load_print_meta: n_head_kv = 2
llm_load_print_meta: n_rot = 128
llm_load_print_meta: n_swa = 0
llm_load_print_meta: n_embd_head_k = 128
llm_load_print_meta: n_embd_head_v = 128
llm_load_print_meta: n_gqa = 6
llm_load_print_meta: n_embd_k_gqa = 256
llm_load_print_meta: n_embd_v_gqa = 256
llm_load_print_meta: f_norm_eps = 0.0e+00
llm_load_print_meta: f_norm_rms_eps = 1.0e-06
llm_load_print_meta: f_clamp_kqv = 0.0e+00
llm_load_print_meta: f_max_alibi_bias = 0.0e+00
llm_load_print_meta: f_logit_scale = 0.0e+00
llm_load_print_meta: n_ff = 8960
llm_load_print_meta: n_expert = 0
llm_load_print_meta: n_expert_used = 0
llm_load_print_meta: causal attn = 1
llm_load_print_meta: pooling type = 0
llm_load_print_meta: rope type = 2
llm_load_print_meta: rope scaling = linear
llm_load_print_meta: freq_base_train = 10000.0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_ctx_orig_yarn = 131072
llm_load_print_meta: rope_finetuned = unknown
llm_load_print_meta: ssm_d_conv = 0
llm_load_print_meta: ssm_d_inner = 0
llm_load_print_meta: ssm_d_state = 0
llm_load_print_meta: ssm_dt_rank = 0
llm_load_print_meta: ssm_dt_b_c_rms = 0
llm_load_print_meta: model type = 1.5B
llm_load_print_meta: model ftype = Q4_K - Medium
llm_load_print_meta: model params = 1.78 B
llm_load_print_meta: model size = 1.04 GiB (5.00 BPW)
llm_load_print_meta: general.name = DeepSeek R1 Distill Qwen 1.5B
llm_load_print_meta: BOS token = 151646 '<｜begin▁of▁sentence｜>'
llm_load_print_meta: EOS token = 151643 '<｜end▁of▁sentence｜>'
llm_load_print_meta: EOT token = 151643 '<｜end▁of▁sentence｜>'
llm_load_print_meta: PAD token = 151643 '<｜end▁of▁sentence｜>'
llm_load_print_meta: LF token = 148848 'ÄĬ'
llm_load_print_meta: FIM PRE token = 151659 '<|fim_prefix|>'
llm_load_print_meta: FIM SUF token = 151661 '<|fim_suffix|>'
llm_load_print_meta: FIM MID token = 151660 '<|fim_middle|>'
llm_load_print_meta: FIM PAD token = 151662 '<|fim_pad|>'
llm_load_print_meta: FIM REP token = 151663 '<|repo_name|>'
llm_load_print_meta: FIM SEP token = 151664 '<|file_sep|>'
llm_load_print_meta: EOG token = 151643 '<｜end▁of▁sentence｜>'
llm_load_print_meta: EOG token = 151662 '<|fim_pad|>'
llm_load_print_meta: EOG token = 151663 '<|repo_name|>'
llm_load_print_meta: EOG token = 151664 '<|file_sep|>'
llm_load_print_meta: max token length = 256
llm_load_tensors: CPU_Mapped model buffer size = 1059.89 MiB
llama_new_context_with_model: n_seq_max = 1
llama_new_context_with_model: n_ctx = 2048
llama_new_context_with_model: n_ctx_per_seq = 2048
llama_new_context_with_model: n_batch = 512
llama_new_context_with_model: n_ubatch = 512
llama_new_context_with_model: flash_attn = 0
llama_new_context_with_model: freq_base = 10000.0
llama_new_context_with_model: freq_scale = 1
llama_new_context_with_model: n_ctx_per_seq (2048) < n_ctx_train (131072) -- the full capacity of the model will not be utilized
llama_kv_cache_init: kv_size = 2048, offload = 1, type_k = 'f16', type_v = 'f16', n_layer = 28, can_shift = 1
llama_kv_cache_init: CPU KV buffer size = 56.00 MiB
llama_new_context_with_model: KV self size = 56.00 MiB, K (f16): 28.00 MiB, V (f16): 28.00 MiB
llama_new_context_with_model: CPU output buffer size = 0.59 MiB
llama_new_context_with_model: CPU compute buffer size = 299.75 MiB
llama_new_context_with_model: graph nodes = 986
llama_new_context_with_model: graph splits = 1
启动后可以看到一些和命令行参数帮助信息对应的参数，比较重要的是端口信息。现在我们可以通过8080端口来测试服务是否正常：
$ curl http://localhost:8080/health
{"status":"ok","progress":1}
也可以访问/completion补全接口：
$ curl -X POST -H "Content-Type: application/json" -d '{"prompt": "hi"}' http://localhost:8080/completion
{"content":",","stop":false,"timings":{"predicted_n":0,"predicted_ms":0,"prompt_n":0,"prompt_ms":0}}
{"content":" I","stop":false,"timings":{"predicted_n":0,"predicted_ms":0,"prompt_n":0,"prompt_ms":0}}
{"content":"'m","stop":false,"timings":{"predicted_n":0,"predicted_ms":0,"prompt_n":0,"prompt_ms":0}}
{"content":" trying","stop":false,"timings":{"predicted_n":0,"predicted_ms":0,"prompt_n":0,"prompt_ms":0}}
...
补全会产生很多结果，用户需要根据实际情况决定何时结束推理。
此外还有/embedding接口：
$ curl -X POST -H "Content-Type: application/json" -d '{"prompt": "turn me into an embedding"}' http://localhost:8080/embedding
{"embedding":[1.6663613,-2.8487935,1.8253331,-1.123977,...,0.95781755]}
5.2 通过程序加载模型
TODO