Z80-\u03bCLM\uff1a\u8ba9 1976 \u5e74的处理器开口说话

\u4e00台 1976 \u5e74的电脑\uff0c\u5982何拥\u6709\u7075\u9b42\uff1f

\u60f3\u8c61\u4e00\u4e0b\uff1a\u4f60\u6253\u5f00\u4e00\u53f0\u843d\u6ee1\u7070\u5c18的 ZX Spectrum\uff0c\u8f93\u5165\u4e00\u884c LOAD ""\uff0c\u5c4f\u5e55\u4e0a\u7f13\u7e43\u51fa\u73b0\u4e00\u4e2a > \u63d0\u793a\u7b26\u3002\u4f60\u6572\u4e0b HELLO\uff0c\u8fd9\u53f0\u53ea\u6709 4MHz \u4e3b\u9891\u300164KB \u5185\u5b58\u7684\u8001\u7237\u673a\u505c\u7559\u4e86\u96f6\u70b9\u51e0\u79d2\uff0c\u7136\u540e\u8ba4\u771f\u5730\u56de\u590d\u4f60\uff1a

> HELLO
HI
> ARE YOU A ROBOT
YES
> DO YOU DREAM
MAYBE

\u8fd9\u4e0d\u662f\u79d1\u5e7b\u7535\u5f71\u91cc\u7684\u65f6\u5149\u脱\u73a9\uff0c\u800c\u662f\u771f\u5b9e\u5b58\u5728\u7684\u5f00\u6e90\u9879\u76ee\u2014\u2014Z80-\u03bCLM\u3002\u5b83\u5c06\u4e00\u4e2a\u5927\u578b\u8bed\u8a00\u7cfb\u7edf\u538b\u7f29\u5230\u6781\u81f4\uff0c\u6700\u7ec8\u4ee5\u4e00\u4e2a\u7ea6 40KB \u7684 .COM \u4e8c\u8fdb\u5236\u6587\u4ef6\u5f62\u5f0f\uff0c\u5728 40 \u591a\u5e74\u524d\u7684\u786c\u4ef6\uff08\u6216\u5176\u6a21\u62df\u5668\uff09\u4e0a\u8fd0\u884c\uff0c\u6bcf\u4e00\u6b21\u5b57\u7b26\u8f93\u51fa\u90fd\u5e26\u968f\u7740 Z80 \u5904\u7406\u5668\u7cbe\u51c6\u7684\u6c47\u7f16\u6307\u4ee4\u6267\u884c\u3002

\u56fe1\uff1aGUESS \u6e38\u620f\u8fd0\u884c\u622a\u56fe\uff0c\u663e\u793a Z80 \u4e8c\u8fdb\u5236\u6587\u4ef6\u5728\u590d\u53e4\u7ec8\u7aef\u4e0a\u7684\u4ea4\u4e92\u6548\u679c

\u80cc\u666f\uff1a\u6781\u5ba2\u7684\u6d6a\u6f2b\u4e0e AI \u7684\u8fb9\u754c\u63a2\u7d22

\u8fd9\u4e2a\u9879\u76ee\u7684\u53d1\u8d77\u4eba HarryR \u63d0\u51fa\u4e86\u4e00\u4e2a\u6781\u5177\u6311\u6218\u6027\u7684\u95ee\u9898\uff1a**\u8bed\u8a00\u6a21\u578b\u80fd\u591f\u505a\u5230\u591a\u5c0f\uff0c\u540c\u65f6\u8fd8\u80fd\u4fdd\u7559\u201c\u7075\u9b42\u201d\uff1f**

\u8fd9\u4e0d\u662f\u4e00\u4e2a\u8ffd\u6c42 SOTA\uff08State-of-the-Art\uff09\u7684\u9879\u76ee\u3002\u5b83\u7684\u76ee\u6807\u4e0d\u662f\u6253\u8d25 GPT-4\uff0c\u800c\u662f\u56de\u7b54\u4e00\u4e2a\u57fa\u7840\u95ee\u9898\uff1a\u5728\u6781\u4f4e\u6bd4\u7279\u5ea6\u3001\u6781\u4f4e\u7b97\u529b\u7684\u7ea6\u675f\u4e0b\uff0c\u8bed\u8a00\u6a21\u578b\u7684\u201c\u7075\u9b42\u201d\u2014\u2014\u90a3\u4e9b\u8ba9\u5b83\u80fd\u6839\u636e\u4e0a\u4e0b\u6587\u751f\u6210\u6709\u610f\u4e49\u7684\u3001\u7b26\u5408\u7279\u5b9a\u98ce\u683c\u7684\u56de\u590d\u7684\u80fd\u529b\u2014\u2014\u7ec8\u7a76\u6765\u81ea\u4e8e\u54ea\u91cc\uff1f

\u4ece\u6280\u672f\u6f14\u8fdb\u89c2\u5bdf\uff0cZ80-\u03bCLM \u8e0a\u5728\u4e86 TinyML\uff08\u5fae\u578b\u673a\u5668\u5b66\u4e60\uff09\u4e0e\u590d\u53e4\u8ba1\u7b97\uff08Retrocomputing\uff09\u4e24\u4e2a\u9886\u57df\u7684\u4ea4\u6c47\u70b9\u3002\u8fd9\u5e74\u6765\uff0c\u91cf\u5316\u6280\u672f\u9a81\u98de\u53d1\u5c55\uff1a\u4ece FP32 \u5230 INT8\uff0c\u518d\u5230 INT4\uff0c\u751a\u81f3\u66f4\u4f4e\u3002\u4f46\u5f88\u5c11\u6709\u4eba\u771f\u6b63\u63a2\u7d22 2-bit \u91cf\u5316\u7684\u6781\u9650\u2014\u2014\u5728 {-2, -1, 0, +1, +2} \u4e94\u4e2a\u79bb\u6563\u503c\u4e0a\u8bad\u7ec3\u4e00\u4e2a\u5b8c\u6574\u7684\u8bed\u8a00\u6a21\u578b\uff0c\u5e76\u8ba9\u5b83\u5728\u771f\u5b9e\u7684 8-bit CPU \u4e0a\u8fd0\u884c\u3002Z80-\u03bCLM \u5c31\u505a\u4e86\u8fd9\u4ef6\u4e8b\u3002

\u9879\u76ee\u88ab\u6807\u8bb0\u4e3a code-golf\uff08\u4ee3\u7801\u9ad8\u5c14\u592b\uff09\u3001tinyml\u3001quantization \u7b49 topics\uff0c\u6e05\u6670\u5730\u8fdb\u660e\u4e86\u5b83\u7684\u6781\u5ba2\u5c5e\u6027\uff1a\u8ffd\u6c42\u5728\u6781\u5c0f\u7a7a\u95f4\u4e2d\u5b9e\u73b0\u6700\u5927\u53ef\u80fd\u3002

\u7c7b\u6bd4\u89e3\u91ca\uff1a\u7ed9\u65e7\u65f6\u4ee3\u88c5\u4e00\u4e2a\u5fae\u578b\u5927\u8111

\u5982\u679c\u8bf4\u4f20\u7edf\u7684 LLM \u662f\u5728\u8d85\u7ea7\u8ba1\u7b97\u673a\u4e0a\u8fd0\u884c\u7684\u201c\u8d85\u7ea7\u5927\u8111\u201d\uff0c\u90a3\u4e48 Z80-\u03bCLM \u5c31\u50cf\u662f\u4e00\u4e2a\u88c5\u5728\u73a9\u5177\u673a\u5668\u4eba\u91cc\u7684\u201c\u5fae\u578b\u82af\u7247\u201d\u2014\u2014\u4e24\u8005\u90fd\u80fd\u5bf9\u8f93\u5165\u505a\u51fa\u53cd\u5e94\uff0c\u4f46\u80fd\u529b\u8303\u56f4\u5b8c\u5168\u4e0d\u540c\u3002

\u5177\u4f53\u6765\u8bf4\uff1a

• \u4f20\u7edf\u5927\u6a21\u578b\uff1a\u6570\u767e\u4ebf\u53c2\u6570 \u2192 \u9700\u8981 GPU \u2192 \u6d88\u8017\u6570\u767e\u74e6\u7535\u529b \u2192 \u54cd\u5e94\u65f6\u95f4\u4ee5\u79d2\u8ba1
• Z80-\u03bCLM\uff1a\u7ea6 256\u2192192\u2192128\u219264 \u7684\u5c0f\u578b MLP\uff08\u591a\u5c42\u611f\u77e5\u673a\uff09 \u2192 \u4ec5\u9700 4MHz Z80 \u5904\u7406\u5668 \u2192 \u6d88\u8017\u6beb\u74e6\u7ea7\u7535\u529b \u2192 \u5b57\u7b26\u9010\u4e2a\u8f93\u51fa\uff0c\u901f\u7387\u4e0e\u6253\u5b57\u901f\u5ea6\u76f8\u5f53

\u6574\u4e2a\u63a8\u7406\u8fc7\u7a0b\u5c31\u662f\u4e00\u4e2a\u7eaf\u7cb9\u7684 Z80 \u6c47\u7f16\u5faa\u73af\uff1a\u63a5\u6536\u8f93\u5165 \u2192 trigram \u54c8\u5e0c\u7f16\u7801 \u2192 \u795e\u7ecf\u7f51\u7edc\u524d\u5411\u4f20\u64ad \u2192 2-bit \u6743\u91cd\u77e9\u9635\u4e58\u6cd5 \u2192 softmax \u91c7\u6837 \u2192 \u5b57\u7b26\u8f93\u51fa \u2192 \u66f4\u65b0\u4e0a\u4e0b\u6587 \u2192 \u5faa\u73af\u3002

\u6838\u5fc3\u529f\u80fd\uff1a\u4ece\u8bad\u7ec3\u5230\u8fd0\u884c\u7684\u5b8c\u6574\u7ba1\u9053

\u91cf\u5316\u611f\u77e5\u8bad\u7ec3\uff08QAT\uff09

\u8fd9\u662f\u9879\u76ee\u6700\u6838\u5fc3\u7684\u6280\u672f\u521b\u65b0\u3002libqat.py \u5b9e\u73b0\u4e86\u4e00\u5957\u5b8c\u6574\u7684 Quantization-Aware Training \u6846\u67b6\uff1a

quant_temp = 0.3 + 0.7 * min(1.0, epoch / (epochs * 0.8))

\u8bad\u7ec3\u8fc7\u7a0b\u4ece 30% \u91cf\u5316\u9012\u6e10\u8fc1\u8fc1\u5230 100% \u91cf\u5316\uff08epoch 80% \u5904\uff09\uff0c\u6700\u540e 20% \u505a\u7eaf\u91cf\u5316\u7cbe\u8c03\u3002\u5173\u952e\u8bbe\u8ba1\u5305\u62ec\uff1a

1. Straight-Through Estimator (STE)\uff1a\u68af\u5ea6\u7ed5\u8fc7\u91cf\u5316\u8282\u70b9\u76f4\u63a5\u56de\u4f20\uff0c\u89e3\u51b3\u91cf\u5316\u4e0d\u53ef\u5bfc\u95ee\u9898
2. \u6ea2\u51fa\u60e9\u7f6a\uff1a\u6a21\u62df 16-bit \u6709\u7b7e\u53f7\u7d2f\u52a0\u5668\u6ea2\u51fa\u98ce\u9669\uff0c\u8bad\u7ec3\u65f6\u5bf9\u6ea2\u51fa\u6a21\u5f0f\u8d1f\u62df\u5219\u5316
3. \u91cf\u5316\u635f\u5931\u9879\uff1a\u989d\u5916\u635f\u5931\u51fd\u6570\u5c06\u6742\u91cd\u63a8\u5411 {-2, -1, 0, +1} \u7f51\u683c\u70b9
4. \u53cc\u6a21\u578b\u5e76\u884c\uff1a\u6d6e\u70b9\u6a21\u578b\u548c\u91cf\u5316\u6a21\u578b\u540c\u65f6\u8bad\u7ec3\uff0c\u7528\u6d6e\u70b9\u6a21\u578b\u8bc4\u4f30\u91cf\u5316\u7cbe\u5ea6

\u8fd9\u5957\u673a\u5236\u786e\u4fdd\u6a21\u578b\u5728\u8bad\u7ec3\u65f6\u5c31\u9002\u5e94\u4e86 2-bit \u79bb\u6563\u5316\uff0c\u907f\u514d\u4e86\u201c\u8bad\u7ec3\u65f6\u662f\u6d6e\u70b9\uff0c\u63a0\u7406\u65f6\u91cf\u5316\u540e\u6027\u80fd\u5d29\u57cb\u201d\u7684\u95ee\u9898\u3002

\u5b57\u7b26\u7ea7\u81ea\u56de\u5f52\u63a0\u7406

feedme.py \u5b9a\u4e49\u4e86\u5b57\u7b26\u7ea7\u81ea\u56de\u5f52\u6a21\u578b\u7684\u8bad\u7ec3\u6846\u67b6\uff1a\u8f93\u5165\u4e3a 128 \u7ef4 query trigram \u5411\u91cf + 128 \u7ef4\u4e0a\u4e0b\u6587\u5411\u91cf\uff08256 \u7ef4\uff09\uff0c\u8f93\u51fa\u4e3a 64 \u5b57\u7b26\u7684 softmax \u5206\u5e03\u3002\u6bcf\u6b21\u751f\u6210\u4e00\u4e2a\u5b57\u7b26\u540e\uff0c\u5c06\u5176\u7f16\u7801\u56de\u4e0a\u4e0b\u6587\u5411\u91cf\uff0c\u7ee7\u7eed\u5faa\u73af\u76f4\u5230\u9047\u5230 EOS\uff08\u7ed3\u675f\u7b26\uff09\u3002

\u8fd9\u662f\u4e00\u4e2a\u9ad8\u5ea6\u5b9a\u5236\u5316\u7684\u67b6\u6784\u2014\u2014\u6ca1\u6709\u4f7f\u7528\u4efb\u4f55\u6807\u51c6 NLP \u5e93\u7684\u81ea\u56de\u5f52\u5b9e\u73b0\uff0c\u7eaf\u7cb9\u662f\u624b\u5de5\u8bbe\u8ba1\u7684 MLP\u3002

Z80 \u6c47\u7f16\u4ee3\u7801\u751f\u6210

libz80.py \u63d0\u4f9b\u4e86\u4e00\u4e2a Z80 \u6c47\u7f16\u4ee3\u7801\u751f\u6210\u5668 Z80Builder\uff0c\u80fd\u591f\u4ee5 Python DSL \u7684\u65b9\u5f0f\u63cf\u8ff0 Z80 \u6307\u4ee4\u5e8f\u5217\uff0c\u5e76\u81ea\u52a8\u5904\u7406\u6807\u7b7e\u89e3\u6790\u548c\u76f8\u5bf9\u8df3\u8f6c\u504f\u79fb\u91cf\u8ba1\u7b97\u3002

buildz80com.py \u5219\u662f\u6574\u4e2a\u7f16\u8bd1\u94fe\u7684\u6838\u5fc3\uff1a\u52a0\u8f7d PyTorch \u6a21\u578b\u6743\u91cd \u2192 \u5e94\u7528 2-bit \u6253\u5305\uff084\u4e2a\u503c/\u5b57\u8282\uff09 \u2192 \u751f\u6210 Z80 \u6c47\u7f16\u6307\u4ee4\u5e8f\u5217 \u2192 \u8f93\u51fa CP/M .COM \u53ef\u6267\u884c\u6587\u4ef6\u3002buildz80tap.py \u8d1f\u8d23\u751f\u6210 ZX Spectrum \u4e13\u7528\u7684 .TAP \u78c1\u5e55\u683c\u5f0f\u3002

\u793a\u4f8b\u5e94\u7528

\u9879\u76ee\u81ea\u5e26\u4e24\u4e2a\u9884\u8bad\u7ec3\u793a\u4f8b\uff1a

• tinychat\uff1a\u95f2\u804a\u673a\u5668\u4eba\uff0c\u7528 | \u5206\u9694\u7684\u8bad\u7ec3\u5bf9\u6570\u636e\u8bad\u7ec3\uff08\u5982 hello|HI, are you real|MAYBE\uff09\uff0c\u5b66\u4f1a\u4e86\u7b80\u6d01\u3001\u6709\u4e2a\u6027\u7684\u56de\u590d\u98ce\u683c
• guess\uff1a20 Questions \u731c\u8c1c\u6e38\u620f\uff0c\u6a21\u578b\u5185\u90e8\u5b58\u50a8\u4e00\u4e2a\u795e\u79d8\u4e3b\u9898\uff0c\u901a\u8fc7 YES/NO/MAYBE \u56de\u7b54\u7528\u6237\u731c\u6d4b

\u8bad\u7ec3\u6570\u636e\u751f\u6210\u5de5\u5177 gendata.py \u652f\u6301\u8c03\u7528 Ollama \u6216 Claude API \u6279\u91cf\u751f\u6210\u95ee\u7b54\u5bf9\uff0c\u5e76\u4f7f\u7528 balance.py \u5e73\u8861\u7c7b\u522b\u5206\u5e03\u3002

\u4e0a\u624b\u95e8\u69db\uff1a\u8bad\u7ec3\u4e0e\u8fd0\u884c\u4e24\u9636\u6bb5

\u9636\u6bb5	\u6240\u9700\u5de5\u5177	\u6280\u672f\u95e8\u69db
\u6a21\u578b\u8bad\u7ec3	Python 3.8+ / PyTorch / NumPy	\u4e2d\u7b49\uff08\u9700\u7406\u89e3 QAT \u539f\u7406\uff09
\u4e8c\u8fdb\u5236\u7f16\u8bd1	buildz80com.py\uff08\u81ea\u52a8\u8c03\u7528 Z80Builder\uff09	\u4f4e\uff08\u6709 Python \u5373\u53ef\uff09
\u8fd0\u884c\uff08CP/M\uff09	iz-cpm \u6a21\u62df\u5668	\u6781\u4f4e\uff08\u4e0b\u8f7d\u6a21\u62df\u5668 + \u8fd0\u884c .COM\uff09
\u8fd0\u884c\uff08ZX Spectrum\uff09	FUSE \u6a21\u62df\u5668	\u6781\u4f4e\uff08\u4e0b\u8f7d\u6a21\u62df\u5668 + \u52a0\u8f7d .TAP\uff09

\u8bad\u7ec3\u6d41\u7a0b\u7b80\u6d01\u660e\u4e86\uff1acat data/*.txt | ./feedme.py --epochs 300 --chat\uff0c\u6309 Ctrl+C \u53ef\u968f\u65f6\u4e2d\u65ad\u5e76\u4fdd\u5b58\u6700\u4f73\u68c0\u67e5\u70b9\u3002\u7f16\u8bd1\u53ea\u9700\uff1a./buildz80com.py -m model.pt -o chat.com\u3002

\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c**\u6ca1\u6709 GPU \u4e5f\u80fd\u8bad\u7ec3**\u2014\u2014\u5bf9\u4e8e\u8fd9\u4e2a\u89c4\u6a21\u7684\u5c0f\u6a21\u578b\uff08256\u2192192\u2192128\u219264\uff0c\u7ea6 10 \u4e07\u53c2\u6570\u7ea7\u522b\uff09\uff0cCPU \u8bad\u7ec3\u5b8c\u5168\u53ef\u884c\uff0c\u51e0\u5341\u5230\u6570\u767e\u4e2a epoch \u5373\u53ef\u6536\u655b\u3002

\u4e89\u8bae\u4e0e\u5c40\u9650\uff1a\u8bda\u5b9e\u7684\u8fb9\u754c

\u5fc5\u987b\u627f\u8ba4 Z80-\u03bCLM \u7684\u5c40\u9650\u6027\uff1a

1. \u751f\u6210\u8d28\u91cf\u6709\u9650\uff1a\u6a21\u578b\u53ea\u80fd\u8f93\u51fa\u8bad\u7ec3\u6570\u636e\u4e2d\u51fa\u73b0\u8fc7\u7684\u6709\u9650\u8bcd\u6c47\uff0c\u56de\u590d\u98ce\u683c\u9ad8\u5ea6\u4f9d\u8d56\u4e8e\u8bad\u7ec3\u6570\u636e\u89c4\u6a21\u3002\u4e0d\u53ef\u80fd\u50cf GPT \u90a3\u6837\u751f\u6210\u5f00\u653e\u57df\u7684\u957f\u6587\u672c\u3002

2. 2-bit \u91cf\u5316\u7684\u7cbe\u5ea6\u635f\u5931\uff1a\u4ece FP32 \u91cf\u5316\u5230 2-bit\uff084\u4e2a\u79bb\u6563\u503c\uff09\uff0c\u4fe1\u606f\u635f\u5931\u663e\u8457\u3002\u6587\u6863\u7279\u522b\u63d0\u9192\u8bad\u7ec3\u65f6\u8981\u5173\u6ce8 IntAcc\uff08\u6574\u6570\u7cbe\u5ea6\uff09\u800c\u975e Acc\uff08\u6d6e\u70b9\u7cbe\u5ea6\uff09\uff1a\u201c\u6d6e\u70b9\u7cbe\u5ea6\u53ef\u8fbe 99%\uff0c\u4f46\u6574\u6570\u7cbe\u5ea6\u53ef\u80fd\u53ea\u6709 60%\u201d\u3002

3. \u8d85\u53c2\u6570\u654f\u611f\uff1a--epochs \u6570\u91cf\u3001\u91cf\u5316\u6e29\u5ea6\u66f2\u7ebf\u8bbe\u8ba1\u90fd\u9700\u8981\u53cd\u590d\u5b9e\u9a8c\u3002\u6587\u6863\u4e2d\u7279\u522b\u63d0\u5230\u201c\u4e0d\u8981\u5728 QTemp=1.0 \u65f6\u7acb\u5373\u505c\u6b62\u8bad\u7ec3\u201d\uff0c\u5426\u5219\u6a21\u578b\u65e0\u6cd5\u5145\u5206\u5728\u91cf\u5316\u7ea6\u675f\u4e0b\u7cbe\u8c03\u3002

4. \u529f\u80fd\u5355\u4e00\uff1a\u6a21\u578b\u4ec5\u652f\u6301\u5b57\u7b26\u7ea7\u81ea\u56de\u5f52\uff0c\u65e0\u6cd5\u5904\u7406\u591a\u8f6e\u590d\u6742\u5bf9\u8bdd\uff0c\u4e5f\u65e0\u6cd5\u6267\u884c\u4efb\u4f55\u5de5\u5177\u8c03\u7528\u6216\u5916\u90e8\u67e5\u8be2\u3002

\u884c\u4e1a\u610f\u4e49\uff1a\u6781\u9650\u6311\u6218\u7684\u4ef7\u503c

\u5c3d\u7ba1 Z80-\u03bCLM \u663e\u7136\u4e0d\u662f\u4e00\u4e2a\u751f\u4ea7\u7ea7 AI \u5de5\u5177\uff0c\u4f46\u5b83\u5728\u4ee5\u4e0b\u65b9\u9762\u5177\u6709\u72ec\u7279\u4ef7\u503c\uff1a

\u5bf9 AI \u7814\u7a76\u793e\u533a\uff1a\u5b83\u63d0\u4f9b\u4e86\u4e00\u4e2a\u6781\u7aef\u91cf\u5316\u7684\u5b9e\u9a8c\u5e73\u53f0\uff0c\u5c55\u793a\u4e86 2-bit QAT \u7684\u53ef\u884c\u6027\uff0c\u8bc1\u660e\u4e86\u5373\u4f7f\u5728\u6781\u5176\u53d7\u9650\u7684\u6743\u91cd\u7a7a\u95f4\u5185\uff0c\u795e\u7ecf\u7f51\u7edc\u4ecd\u80fd\u5b66\u4e60\u5230\u6709\u610f\u4e49\u7684\u884c\u4e3a\u6a21\u5f0f\u3002\u8fd9\u5bf9\u4e8e\u7814\u7a76\u6a21\u578b\u538b\u7f29\u3001\u795e\u7ecf\u7f51\u7edc\u526a\u679d\u8fb9\u754c\u7684\u5b66\u8005\u6709\u53c2\u8003\u4ef7\u503c\u3002

\u5bf9\u6781\u5ba2\u793e\u533a\uff1a\u5b83\u5c06 AI \u80fd\u529b\u5e26\u5165\u4e86\u590d\u53e4\u8ba1\u7b97\u8fd9\u4e2a\u5145\u6ee1\u70ed\u60c5\u7684\u5c0f\u4f17\u5708\u5b50\uff0c\u8ba9 ZX Spectrum\u3001CP/M \u65f6\u4ee3\u7684\u673a\u5668\u4e5f\u80fd\u62e5\u6709\u201c\u5bf9\u8bdd\u201d\u4f53\u9a8c\uff0c\u8d4b\u4e88\u8001\u786c\u4ef6\u65b0\u7684\u751f\u547d\u529b\u3002

\u5bf9\u6559\u80b2\u8005\uff1a\u6574\u4e2a\u9879\u76ee\u4ee3\u7801\u91cf\u9002\u4e2d\uff0820KB \u7c7b\u578b\u7684\u6838\u5fc3 Python \u6587\u4ef6\uff09\uff0c\u67b6\u6784\u6e05\u6670\uff0c\u662f\u5b66\u4e60\u795e\u7ecf\u7f51\u7edc\u91cf\u5316\u3001Z80 \u6c47\u7f16\u3001\u5d4c\u5165\u5f0f ML \u7684\u7edd\u4f73\u5b9e\u6218\u6559\u6750\u3002

---

\u672c\u62a5\u544a\u7531 PIFS \u81ea\u52a8\u5316\u5206\u6790\u7cfb\u7edf\u751f\u6210\u3002\u6570\u636e\u6765\u6e90\uff1aGitHub API + \u4ee3\u7801\u9759\u6001\u5206\u6790\u3002