在《大语言模型之十二 SentencePiece扩充LLama2中文词汇》一文中已经扩充好了中文词汇表，接下来就是使用整理的中文语料对模型进行预训练了。这里先跳过预训练环节。先试用已经训练好的模型，看看如何推理。
合并模型
这一步骤会合并LoRA权重，生成全量模型权重。此处可以选择输出PyTorch版本权重（.pth文件）或者输出HuggingFace版本权重（.bin文件）。执行以下命令：
$ python scripts/merge_llama2_with_chinese_lora_low_mem.py \--base_model path_to_original_llama2_hf_dir \--lora_model path_to_chinese_llama2_or_alpaca2_lora \--output_type huggingface \--output_dir path_to_output_dir
参数说明：
- –base_model：存放HF格式的Llama-2模型权重和配置文件的目录，这可以在《大语言模型之十二 SentencePiece扩充LLama2中文词汇》的1.下载原版LLama-2模型小节找到如何将原始meta的LlaMA-2模型转为Huggingface的格式。
- –lora_model：中文LLaMA-2/Alpaca-2 LoRA解压后文件所在目录，也可使用🤗Model Hub模型调用名称（会自动下载），这里使用Chinese-LLaMA-Alpaca-2给出的预训练好的7B模型。
- –output_type：指定输出格式，可为pth或huggingface。若不指定，默认为huggingface
- –output_dir：指定保存全量模型权重的目录，默认为./
- （可选）–verbose：显示合并过程中的详细信息
转换好格式之后，内容如下（时间戳为11：28的即为转换生成文件)：
其中的ggml开头的事量化文件是用于模型推理。
推理
在attn_and_long_ctx_patches.py实现了基于NTK的自适应上下文适配方法，其中基于transformers的推理脚本。
- 当上下文小于4K时，默认关闭，因为原生的效果更好
- 大于4K时开启NTK，AUTO_COEFF默认为1.0
以下是不同AUTO_COEFF下，在不同上下文长度上的PPL变化（越低越好），供使用参考。
对NTK方法熟悉的用户可直接修改代码中的ALPHA取值。 - 12K以下：几乎和原生4K的PPL没有显著差异
- 12K-16K：开始存在一定损失，大约是3比特量化级别的效果
- 18K+：存在较大损失，大约是2比特量化级别效果，20K+不可用
以上结果仅供参考，应在实际场景中测试调整AUTO_COEFF或者ALPHA取值。
使用llama.cpp推理
Step 1: 克隆和编译llama.cpp
- （可选）如果已下载旧版仓库，建议git pull拉取最新代码，并执行make clean进行清理
- 拉取最新版llama.cpp仓库代码
$ git clone https://github.com/ggerganov/llama.cpp
- 对llama.cpp项目进行编译，生成./main（用于推理）和./quantize（用于量化）二进制文件。
$ make
Step 2: 生成量化版本模型
目前llama.cpp已支持.pth文件以及huggingface格式.bin的转换。将完整模型权重转换为GGML的FP16格式，生成文件路径为zh-models/7B/ggml-model-f16.gguf。进一步对FP16模型进行4-bit量化，生成量化模型文件路径为zh-models/7B/ggml-model-q4_0.gguf。不同量化方法的性能对比见本Wiki最后部分。
python3 convert.py ../merged_chinese_llama_7b
$ ./quantize ../merged_chinese_llama_7b/ggml-model-f16.gguf ../merged_chinese_llama_7b/ggml-model-q4_0.gguf q4_0
Step 3: 加载并启动模型
llama.cpp git:(master) ✗ ./main -s 1 -m ../merged_chinese_llama_7b/ggml-model-q4_0.gguf -p "中国的首都是" --ignore-eos -c 64 -n 128 -t 3 -ngl 10
-
GPU推理：通过Metal编译则只需在./main中指定-ngl 1；cuBLAS编译需要指定offload层数，例如-ngl 40表示offload 40层模型参数到GPU
-
加载长上下文模型（16K）：
- 启动模型（./main）后debug信息中显示llm_load_print_meta: freq_scale = 0.25，则表示模型转换时已载入相应超参，无需其他特殊设置
- 如果上述debug信息显示为llm_load_print_meta: freq_scale = 1.0，则需在./main中额外指定–rope-scale 4
-
默认的量化方法为q4_0，虽然速度最快但损失也较大，推荐使用Q4_K作为替代
-
机器资源够用且对速度要求不是那么苛刻的情况下可以使用q8_0或Q6_K，非常接近F16模型的效果
如果使用的是Mac Intel可能报如下错：
ggml_metal_init: load pipeline error: Error Domain=CompilerError Code=2 "SC compilation failure
There is a call to an undefined label" UserInfo={NSLocalizedDescription=SC compilation failure
There is a call to an undefined label}
llama_new_context_with_model: ggml_metal_init() failed
llama_init_from_gpt_params: error: failed to create context with model '../merged_chinese_llama_7b/ggml-model-q4_0.gguf'
main: error: unable to load model
可以按这里的修改
$ make clean
$ brew update && brew install clblast
#disable metal and enable clblast
$ make LLAMA_CLBLAST=1 LLAMA_NO_METAL=1
#这时可以用main进行推理
$./main -s 1 -m ../merged_chinese_llama_7b/ggml-model-q4_0.gguf -p "中国的首都是" --ignore-eos -c 64 -n 128 -t 3 -ngl 10
对应的终端输出为：
(venv) ➜ llama.cpp git:(master) ✗ ./main -s 1 -m ../merged_chinese_llama_7b/ggml-model-q4_0.gguf -p "中国的首都是" --ignore-eos -c 64 -n 128 -t 3 -ngl 10
Log start
main: warning: changing RoPE frequency base to 0 (default 10000.0)
main: warning: scaling RoPE frequency by 0 (default 1.0)
main: build = 1273 (99115f3)
main: built with Apple clang version 14.0.3 (clang-1403.0.22.14.1) for x86_64-apple-darwin22.5.0
main: seed = 1
ggml_opencl: selecting platform: 'Apple'
ggml_opencl: selecting device: 'Intel(R) UHD Graphics 630'
ggml_opencl: device FP16 support: false
llama_model_loader: loaded meta data with 19 key-value pairs and 291 tensors from ../merged_chinese_llama_7b/ggml-model-q4_0.gguf (version GGUF V2 (latest))
llama_model_loader: - tensor 0: token_embd.weight q4_0 [ 4096, 55296, 1, 1 ]
llama_model_loader: - tensor 1: blk.0.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 2: blk.0.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 3: blk.0.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 4: blk.0.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 5: blk.0.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 6: blk.0.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 7: blk.0.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 8: blk.0.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 9: blk.0.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 10: blk.1.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 11: blk.1.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 12: blk.1.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 13: blk.1.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 14: blk.1.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 15: blk.1.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 16: blk.1.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 17: blk.1.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 18: blk.1.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 19: blk.2.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 20: blk.2.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 21: blk.2.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 22: blk.2.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 23: blk.2.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 24: blk.2.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 25: blk.2.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 26: blk.2.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 27: blk.2.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 28: blk.3.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 29: blk.3.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 30: blk.3.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 31: blk.3.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 32: blk.3.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 33: blk.3.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 34: blk.3.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 35: blk.3.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 36: blk.3.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 37: blk.4.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 38: blk.4.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 39: blk.4.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 40: blk.4.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 41: blk.4.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 42: blk.4.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 43: blk.4.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 44: blk.4.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 45: blk.4.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 46: blk.5.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 47: blk.5.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 48: blk.5.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 49: blk.5.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 50: blk.5.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 51: blk.5.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 52: blk.5.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 53: blk.5.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 54: blk.5.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 55: blk.6.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 56: blk.6.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 57: blk.6.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 58: blk.6.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 59: blk.6.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 60: blk.6.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 61: blk.6.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 62: blk.6.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 63: blk.6.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 64: blk.7.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 65: blk.7.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 66: blk.7.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 67: blk.7.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 68: blk.7.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 69: blk.7.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 70: blk.7.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 71: blk.7.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 72: blk.7.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 73: blk.8.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 74: blk.8.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 75: blk.8.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 76: blk.8.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 77: blk.8.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 78: blk.8.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 79: blk.8.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 80: blk.8.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 81: blk.8.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 82: blk.9.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 83: blk.9.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 84: blk.9.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 85: blk.9.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 86: blk.9.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 87: blk.9.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 88: blk.9.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 89: blk.9.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 90: blk.9.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 91: blk.10.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 92: blk.10.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 93: blk.10.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 94: blk.10.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 95: blk.10.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 96: blk.10.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 97: blk.10.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 98: blk.10.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 99: blk.10.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 100: blk.11.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 101: blk.11.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 102: blk.11.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 103: blk.11.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 104: blk.11.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 105: blk.11.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 106: blk.11.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 107: blk.11.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 108: blk.11.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 109: blk.12.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 110: blk.12.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 111: blk.12.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 112: blk.12.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 113: blk.12.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 114: blk.12.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 115: blk.12.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 116: blk.12.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 117: blk.12.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 118: blk.13.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 119: blk.13.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 120: blk.13.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 121: blk.13.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 122: blk.13.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 123: blk.13.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 124: blk.13.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 125: blk.13.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 126: blk.13.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 127: blk.14.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 128: blk.14.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 129: blk.14.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 130: blk.14.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 131: blk.14.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 132: blk.14.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 133: blk.14.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 134: blk.14.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 135: blk.14.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 136: blk.15.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 137: blk.15.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 138: blk.15.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 139: blk.15.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 140: blk.15.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 141: blk.15.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 142: blk.15.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 143: blk.15.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 144: blk.15.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 145: blk.16.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 146: blk.16.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 147: blk.16.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 148: blk.16.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 149: blk.16.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 150: blk.16.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 151: blk.16.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 152: blk.16.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 153: blk.16.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 154: blk.17.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 155: blk.17.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 156: blk.17.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 157: blk.17.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 158: blk.17.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 159: blk.17.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 160: blk.17.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 161: blk.17.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 162: blk.17.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 163: blk.18.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 164: blk.18.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 165: blk.18.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 166: blk.18.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 167: blk.18.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 168: blk.18.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 169: blk.18.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 170: blk.18.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 171: blk.18.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 172: blk.19.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 173: blk.19.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 174: blk.19.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 175: blk.19.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 176: blk.19.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 177: blk.19.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 178: blk.19.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 179: blk.19.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 180: blk.19.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 181: blk.20.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 182: blk.20.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 183: blk.20.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 184: blk.20.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 185: blk.20.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 186: blk.20.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 187: blk.20.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 188: blk.20.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 189: blk.20.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 190: blk.21.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 191: blk.21.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 192: blk.21.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 193: blk.21.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 194: blk.21.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 195: blk.21.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 196: blk.21.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 197: blk.21.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 198: blk.21.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 199: blk.22.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 200: blk.22.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 201: blk.22.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 202: blk.22.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 203: blk.22.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 204: blk.22.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 205: blk.22.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 206: blk.22.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 207: blk.22.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 208: blk.23.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 209: blk.23.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 210: blk.23.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 211: blk.23.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 212: blk.23.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 213: blk.23.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 214: blk.23.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 215: blk.23.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 216: blk.23.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 217: blk.24.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 218: blk.24.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 219: blk.24.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 220: blk.24.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 221: blk.24.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 222: blk.24.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 223: blk.24.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 224: blk.24.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 225: blk.24.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 226: blk.25.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 227: blk.25.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 228: blk.25.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 229: blk.25.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 230: blk.25.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 231: blk.25.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 232: blk.25.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 233: blk.25.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 234: blk.25.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 235: blk.26.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 236: blk.26.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 237: blk.26.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 238: blk.26.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 239: blk.26.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 240: blk.26.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 241: blk.26.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 242: blk.26.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 243: blk.26.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 244: blk.27.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 245: blk.27.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 246: blk.27.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 247: blk.27.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 248: blk.27.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 249: blk.27.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 250: blk.27.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 251: blk.27.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 252: blk.27.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 253: blk.28.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 254: blk.28.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 255: blk.28.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 256: blk.28.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 257: blk.28.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 258: blk.28.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 259: blk.28.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 260: blk.28.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 261: blk.28.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 262: blk.29.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 263: blk.29.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 264: blk.29.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 265: blk.29.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 266: blk.29.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 267: blk.29.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 268: blk.29.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 269: blk.29.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 270: blk.29.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 271: blk.30.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 272: blk.30.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 273: blk.30.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 274: blk.30.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 275: blk.30.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 276: blk.30.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 277: blk.30.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 278: blk.30.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 279: blk.30.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 280: blk.31.attn_q.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 281: blk.31.attn_k.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 282: blk.31.attn_v.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 283: blk.31.attn_output.weight q4_0 [ 4096, 4096, 1, 1 ]
llama_model_loader: - tensor 284: blk.31.ffn_gate.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 285: blk.31.ffn_up.weight q4_0 [ 4096, 11008, 1, 1 ]
llama_model_loader: - tensor 286: blk.31.ffn_down.weight q4_0 [ 11008, 4096, 1, 1 ]
llama_model_loader: - tensor 287: blk.31.attn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 288: blk.31.ffn_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 289: output_norm.weight f32 [ 4096, 1, 1, 1 ]
llama_model_loader: - tensor 290: output.weight q6_K [ 4096, 55296, 1, 1 ]
llama_model_loader: - kv 0: general.architecture str
llama_model_loader: - kv 1: general.name str
llama_model_loader: - kv 2: llama.context_length u32
llama_model_loader: - kv 3: llama.embedding_length u32
llama_model_loader: - kv 4: llama.block_count u32
llama_model_loader: - kv 5: llama.feed_forward_length u32
llama_model_loader: - kv 6: llama.rope.dimension_count u32
llama_model_loader: - kv 7: llama.attention.head_count u32
llama_model_loader: - kv 8: llama.attention.head_count_kv u32
llama_model_loader: - kv 9: llama.attention.layer_norm_rms_epsilon f32
llama_model_loader: - kv 10: llama.rope.freq_base f32
llama_model_loader: - kv 11: general.file_type u32
llama_model_loader: - kv 12: tokenizer.ggml.model str
llama_model_loader: - kv 13: tokenizer.ggml.tokens arr
llama_model_loader: - kv 14: tokenizer.ggml.scores arr
llama_model_loader: - kv 15: tokenizer.ggml.token_type arr
llama_model_loader: - kv 16: tokenizer.ggml.bos_token_id u32
llama_model_loader: - kv 17: tokenizer.ggml.eos_token_id u32
llama_model_loader: - kv 18: general.quantization_version u32
llama_model_loader: - type f32: 65 tensors
llama_model_loader: - type q4_0: 225 tensors
llama_model_loader: - type q6_K: 1 tensors
llm_load_print_meta: format = GGUF V2 (latest)
llm_load_print_meta: arch = llama
llm_load_print_meta: vocab type = SPM
llm_load_print_meta: n_vocab = 55296
llm_load_print_meta: n_merges = 0
llm_load_print_meta: n_ctx_train = 2048
llm_load_print_meta: n_ctx = 64
llm_load_print_meta: n_embd = 4096
llm_load_print_meta: n_head = 32
llm_load_print_meta: n_head_kv = 32
llm_load_print_meta: n_layer = 32
llm_load_print_meta: n_rot = 128
llm_load_print_meta: n_gqa = 1
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: n_ff = 11008
llm_load_print_meta: freq_base = 10000.0
llm_load_print_meta: freq_scale = 1
llm_load_print_meta: model type = 7B
llm_load_print_meta: model ftype = mostly Q4_0
llm_load_print_meta: model params = 6.93 B
llm_load_print_meta: model size = 3.69 GiB (4.57 BPW)
llm_load_print_meta: general.name = ..
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 = 0.09 MB
llm_load_tensors: using OpenCL for GPU acceleration
llm_load_tensors: mem required = 2687.86 MB (+ 32.00 MB per state)
llm_load_tensors: offloading 10 repeating layers to GPU
llm_load_tensors: offloaded 10/33 layers to GPU
llm_load_tensors: VRAM used: 1086 MB
..............................................................................................
llama_new_context_with_model: kv self size = 32.00 MB
llama_new_context_with_model: compute buffer total size = 15.97 MBsystem_info: n_threads = 3 / 12 | AVX = 1 | 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 = 1 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 |
sampling: repeat_last_n = 64, repeat_penalty = 1.100000, presence_penalty = 0.000000, frequency_penalty = 0.000000, top_k = 40, tfs_z = 1.000000, top_p = 0.950000, typical_p = 1.000000, temp = 0.800000, mirostat = 0, mirostat_lr = 0.100000, mirostat_ent = 5.000000
generate: n_ctx = 64, n_batch = 512, n_predict = 128, n_keep = 0中国的首都是世界上政治、军事和文化中心。长安古称"京师",后为北京；北宋时期，东京开封府一度升格为"中都"或"大都"。《长安志》记载："自建都以来,因得名曰'长安'者有…
一些说明
这里将两个基座模型和LORA fine tune模型merge的原因在于扩充词汇表之后，Embedding也进行了扩充，词汇表比原始的LlaMA-2 32k大，因而要将Embedding层merge（实际是替换)，此外Attention（q,k,v)以及MLP（feedforward，w1,w2,w3)基本都进行了merge操作。由于改动如此之大，以至于《大语言模型之七- Llama-2单GPU微调SFT》博客里微调方法是一样的，但是改动量和训练的资源需求是不一样的，这也导致了扩充中文的微调训练在colab免费的12G GPU内存上是无法完成训练的。
PEFT是 Hugging Face提供的模型训练的高效库，LORA是其提供的方法之一，LORA方式是2021年论文 LoRA: Low-rank adaptation of Large Language Models.首先引入的方法。
其核心思想是可以在仅调整一小部分权重的同时实现出色的性能，进而无需在多台机器上调整数十亿个参数，使整个微调过程更加实用且经济可行。使用PEFT和量化允许在单个GPU上微调具有数十亿个参数的大型模型。比如Embedding是词向量的编码，虽然任务不同，如问答、摘要、协作类的大模型，虽然应用不同，但是词向量编码是可以复用的，不需要改，因而在微调的时候，就不改词向量了，这样就节省存储和运算资源。
