This model is converted from decapoda-research/llama-7b-hf to ziqingyang/chinese-alpaca-plus-lora-7b and quantized for use with ggerganov/llama.cpp.
The convertion and quantization is done on Google Colab following Wiki article of ymcui/Chinese-LLaMA-Alpaca.
The quantization methods have been updated for llama.cpp, so please cloning the latest repo and re-compile before loading the model.
The q8_0 and q5_1 indicate for different quantization method, the former one occupies larger space and theoratically produces better response.
Example of q8_0 model on llama.cpp (acceptable responses but very short):
> ./main -m ./models/chinese-Alpaca-7b-plus-ggml-q8_0.bin \
-t 8 \
-c 2048 \
-n 2048 \
--color \
--interactive-first \
--reverse-prompt '## 人类：' \
-f ./prompts/chat-with-vicuna-chs.txt
main: build = 0 (unknown)
main: seed = 1683883289
llama.cpp: loading model from ./models/chinese-Alpaca-7b-plus-ggml-q8_0.bin
llama_model_load_internal: format = ggjt v1 (latest)
llama_model_load_internal: n_vocab = 49954
llama_model_load_internal: n_ctx = 2048
llama_model_load_internal: n_embd = 4096
llama_model_load_internal: n_mult = 256
llama_model_load_internal: n_head = 32
llama_model_load_internal: n_layer = 32
llama_model_load_internal: n_rot = 128
llama_model_load_internal: ftype = 7 (mostly Q8_0)
llama_model_load_internal: n_ff = 11008
llama_model_load_internal: n_parts = 1
llama_model_load_internal: model size = 7B
llama_model_load_internal: ggml ctx size = 68.20 KB
llama_model_load_internal: mem required = 9180.13 MB (+ 1026.00 MB per state)
llama_init_from_file: kv self size = 1024.00 MB
system_info: n_threads = 8 / 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 = 0 | SSE3 = 1 | VSX = 0 |
main: interactive mode on.
Reverse prompt: '## 人类：'
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 = 2048, n_batch = 512, n_predict = 2048, n_keep = 0
== Running in interactive mode. ==
- Press Ctrl+C to interject at any time.
- Press Return to return control to LLaMa.
- To return control without starting a new line, end your input with '/'.
- If you want to submit another line, end your input with '\'.
这是一段人类和人工智能助理的对话。人工智能助理会对人类的问题给出详细、准确、礼貌的回答。
## 人类：你好，助理。
## 助理：你好，有什么需要帮助的吗？
## 人类：天空为什么是蓝色的？
助理：天空之所以呈现蓝色，是因为太阳光在大气层中发生散射作用。
## 人类：请再详细地解释一下。
助理：太阳光穿过大气层时，一部分被散射到不同波长的角度上，其中蓝色光的波长比其他颜色短。
## 人类：你真是惜字如金啊。
助理：谢谢你的夸奖，但我只是根据我学到的知识和算法来回答你的问题。
## 人类：
llama_print_timings: load time = 9418.31 ms
llama_print_timings: sample time = 107.95 ms / 73 runs ( 1.48 ms per run)
llama_print_timings: prompt eval time = 8645.76 ms / 85 tokens ( 101.71 ms per token)
llama_print_timings: eval time = 16303.43 ms / 73 runs ( 223.33 ms per run)
llama_print_timings: total time = 987546.29 ms