LlamaCPP通过LlamaIndex推理时卡死问题排查求助

使用LlamaIndex库调用本地GGUF格式的Llama-2-13B-chat模型，因显存不足采用CPU推理，程序输出初始化信息后无限卡死，无任何响应。当前内存占用约9.5GB/16GB，CPU占用约50%。

运行代码 #

import json

from llama_index.llms.llama_cpp import LlamaCPP

MODEL_URL = "https://huggingface.co/TheBloke/Llama-2-13B-chat-GGUF/resolve/main/llama-2-13b-chat.Q4_0.gguf"
MODEL_PATH = None

with open("./paths.json", "r") as f:
    paths = json.load(f)
    if "llama-2-13b-chat" in paths:
        MODEL_URL = None
        MODEL_PATH = paths["llama-2-13b-chat"]

llm = LlamaCPP(
    model_url=MODEL_URL,
    model_path=MODEL_PATH,
    temperature=0.1,
    max_new_tokens=256,
    context_window=3900,
    model_kwargs={"n_gpu_layers": 0}, # Use CPU for inference
    verbose=True,
)

response = llm.complete("Hello, how are you?")
print(str(response))

初始化输出信息 #

llama_model_loader: loaded meta data with 19 key-value pairs and 363 tensors from ../models/llama-2-13b-chat.Q4_0.gguf (version GGUF V2)
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
llama_model_loader: - kv   0:                       general.architecture str              = llama
llama_model_loader: - kv   1:                               general.name str              = LLaMA v2
llama_model_loader: - kv   2:                       llama.context_length u32              = 4096
llama_model_loader: - kv   3:                     llama.embedding_length u32              = 5120
llama_model_loader: - kv   4:                          llama.block_count u32              = 40
llama_model_loader: - kv   5:                  llama.feed_forward_length u32              = 13824
llama_model_loader: - kv   6:                 llama.rope.dimension_count u32              = 128
llama_model_loader: - kv   7:                 llama.attention.head_count u32              = 40
llama_model_loader: - kv   8:              llama.attention.head_count_kv u32              = 40
llama_model_loader: - kv   9:     llama.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  10:                          general.file_type u32              = 2
llama_model_loader: - kv  11:                       tokenizer.ggml.model str              = llama
llama_model_loader: - kv  12:                      tokenizer.ggml.tokens arr[str,32000]   = ["<unk>", "<s>", "</s>", "<0x00>", "<...
llama_model_loader: - kv  13:                      tokenizer.ggml.scores arr[f32,32000]   = [0.000000, 0.000000, 0.000000, 0.0000...
llama_model_loader: - kv  14:                  tokenizer.ggml.token_type arr[i32,32000]   = [2, 3, 3, 6, 6, 6, 6, 6, 6, 6, 6, 6, ...
llama_model_loader: - kv  15:                tokenizer.ggml.bos_token_id u32              = 1
llama_model_loader: - kv  16:                tokenizer.ggml.eos_token_id u32              = 2
llama_model_loader: - kv  17:            tokenizer.ggml.unknown_token_id u32              = 0
llama_model_loader: - kv  18:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:   81 tensors
llama_model_loader: - type q4_0:  281 tensors
llama_model_loader: - type q6_K:    1 tensors
llm_load_vocab: special tokens definition check successful ( 259/32000 ).
llm_load_print_meta: format           = GGUF V2
llm_load_print_meta: arch             = llama
llm_load_print_meta: vocab type       = SPM
llm_load_print_meta: n_vocab          = 32000
llm_load_print_meta: n_merges         = 0
llm_load_print_meta: n_ctx_train      = 4096
llm_load_print_meta: n_embd           = 5120
llm_load_print_meta: n_head           = 40
llm_load_print_meta: n_head_kv        = 40
llm_load_print_meta: n_layer          = 40
llm_load_print_meta: n_rot            = 128
llm_load_print_meta: n_embd_head_k    = 128
llm_load_print_meta: n_embd_head_v    = 128
llm_load_print_meta: n_gqa            = 1
llm_load_print_meta: n_embd_k_gqa     = 5120
llm_load_print_meta: n_embd_v_gqa     = 5120
llm_load_print_meta: f_norm_eps       = 0.0e+00
llm_load_print_meta: f_norm_rms_eps   = 1.0e-05
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             = 13824
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        = 0
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_yarn_orig_ctx  = 4096
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: model type       = 13B
llm_load_print_meta: model ftype      = Q4_0
llm_load_print_meta: model params     = 13.02 B
llm_load_print_meta: model size       = 6.86 GiB (4.53 BPW) 
llm_load_print_meta: general.name     = LLaMA v2
llm_load_print_meta: BOS token        = 1 '<s>'
llm_load_print_meta: EOS token        = 2 '</s>'
llm_load_print_meta: UNK token        = 0 '<unk>'
llm_load_print_meta: LF token         = 13 '<0x0A>'
llm_load_tensors: ggml ctx size =    0.18 MiB
llm_load_tensors:        CPU buffer size =  7023.90 MiB
...................................................................................................
llama_new_context_with_model: n_ctx      = 4096
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_kv_cache_init:        CPU KV buffer size =  3200.00 MiB
llama_new_context_with_model: KV self size  = 3200.00 MiB, K (f16): 1600.00 MiB, V (f16): 1600.00 MiB
llama_new_context_with_model:        CPU  output buffer size =     0.12 MiB
llama_new_context_with_model:        CPU compute buffer size =   368.01 MiB
llama_new_context_with_model: graph nodes  = 1286
llama_new_context_with_model: graph splits = 1
AVX = 1 | AVX_VNNI = 0 | AVX2 = 1 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 0 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 | MATMUL_INT8 = 0 | LLAMAFILE = 1 | 
Model metadata: {'tokenizer.ggml.unknown_token_id': '0', 'tokenizer.ggml.eos_token_id': '2', 'general.architecture': 'llama', 'llama.context_length': '4096', 'general.name': 'LLaMA v2', 'llama.embedding_length': '5120', 'llama.feed_forward_length': '13824', 'llama.attention.layer_norm_rms_epsilon': '0.000010', 'llama.rope.dimension_count': '128', 'llama.attention.head_count': '40', 'tokenizer.ggml.bos_token_id': '1', 'llama.block_count': '40', 'llama.attention.head_count_kv': '40', 'general.quantization_version': '2', 'tokenizer.ggml.model': 'llama', 'general.file_type': '2'}
Using fallback chat format: llama-2

排查与解决方法 #

1. 内存资源瓶颈 #

从日志看，模型本身占6.86GiB，加上KV缓存3.12GiB，总内存需求接近10GiB，16GB内存剩余空间有限，系统可能因内存交换（swap）导致进程阻塞：

• 降低context_window参数至2048，减少KV缓存占用；
• 改用更低量化版本的模型（如Q2_K、Q3_K_M），进一步压缩模型内存占用。

2. CPU批次与指令集优化 #

当前CPU占用50%可能是批次设置过高或指令集优化未启用：

• 在model_kwargs中添加n_batch=256，降低推理批次；
• 重新编译llama-cpp-python时启用CPU优化：设置环境变量CMAKE_ARGS="-DLLAMA_AVX2=ON -DLLAMA_FMA=ON"后再安装。

3. 版本兼容性检查 #

确认llama-index-llms-llama-cpp和llama-cpp-python版本匹配，建议升级到最新稳定版，避免版本兼容导致的死锁。

4. 显式指定聊天格式 #

虽然日志显示使用llama-2格式，但显式指定可避免格式错误：

llm = LlamaCPP(
    # 其他参数不变
    chat_format="llama-2",
)

内容的提问来源于stack exchange，提问作者Calder Johnson