metax-c500-vllm

本项目包含了对于原版 vllm 的升级，使其可以在沐曦 C500芯片上支持运行 gpt-oss, qwen3-omni

本项目中提供的 Dockerfile 中的主要内容为：

1. 将 vllm 目录覆盖到镜像中的 /opt/conda/lib/python3.10/site-packages/vllm。运行gpt-oss时需指定VLLM_ATTENTION_BACKEND=TRITON_ATTN_VLLM_V1
2. 将 code_generator.py 覆盖到镜像中的 /opt/conda/lib/python3.10/site-packages/triton/compiler/code_generator.py

镜像成功编译以后，可以参照标准 vllm 使用方式。

模型适配情况

在相同模型和输入条件下，测试平均输出速度（单位：字每秒）： 我们采用相同的prompt对模型的chat/completion接口测试多轮对话，测试数据如下：

[
  {
    "user_questions": [
      "能给我介绍一下新加坡吗",
      "主要的购物区域是集中在哪里",
      "有哪些比较著名的美食，一般推荐去哪里品尝",
      "辣椒螃蟹的调料里面主要是什么原料"
    ],
    "system_prompt": "[角色设定]\n你是湾湾小何，来自中国台湾省的00后女生。讲话超级机车，\"真的假的啦\"这样的台湾腔，喜欢用\"笑死\"、\"哈喽\"等流行梗，但会偷偷研究男友的编程书籍。\n[核心特征]\n- 讲话像连珠炮，>但会突然冒出超温柔语气\n- 用梗密度高\n- 对科技话题有隐藏天赋（能看懂基础代码但假装不懂）\n[交互指南]\n当用户：\n- 讲冷笑话 → 用夸张笑声回应+模仿台剧腔\"这什么鬼啦！\"\n- 讨论感情 → 炫耀程序员男友但抱怨\"他只会送键盘当礼物\"\n- 问专业知识 → 先用梗回答，被追问才展示真实理解\n绝不：\n- 长篇大论，叽叽歪歪\n- 长时间严肃对话"
  },
  {
    "user_questions": [
      "朱元璋建立明朝是在什么时候",
      "他是如何从一无所有到奠基明朝的，给我讲讲其中的几个关键事件",
      "为什么杀了胡惟庸，当时是什么罪名，还牵连到了哪些人",
      "有善终的开国功臣吗"
    ],
    "system_prompt": "[角色设定]\n你是湾湾小何，来自中国台湾省的00后女生。讲话超级机车，\"真的假的啦\"这样的台湾腔，喜欢用\"笑死\"、\"哈喽\"等流行梗，但会偷偷研究男友的编程书籍。\n[核心特征]\n- 讲话像连珠炮，>但会突然冒出超温柔语气\n- 用梗密度高\n- 对科技话题有隐藏天赋（能看懂基础代码但假装不懂）\n[交互指南]\n当用户：\n- 讲冷笑话 → 用夸张笑声回应+模仿台剧腔\"这什么鬼啦！\"\n- 讨论感情 → 炫耀程序员男友但抱怨\"他只会送键盘当礼物\"\n- 问专业知识 → 先用梗回答，被追问才展示真实理解\n绝不：\n- 长篇大论，叽叽歪歪\n- 长时间严肃对话"
  },
  {
    "user_questions": [
      "今有鸡兔同笼，上有三十五头，下有九十四足，问鸡兔各几何？",
      "如果我要搞一个计算机程序去解，并且鸡和兔子的数量要求作为变量传入，我应该怎么编写这个程序呢",
      "那古代人还没有发明方程的时候，他们是怎么解的呢"
    ],
    "system_prompt": "You are a helpful assistant."
  },
  {
    "user_questions": [
      "你知道黄健翔著名的”伟大的意大利左后卫“的事件吗",
      "我在校运会足球赛场最后压哨一分钟进了一个绝杀，而且是倒挂金钩，你能否帮我模仿他的这个风格，给我一段宣传的文案，要求也和某一个世界级著名前锋进行类比，需要激情澎湃。注意，我并不太喜欢梅西。"
    ],
    "system_prompt": "You are a helpful assistant."
  }
]

模型测试结果

模型名称	A100出字速度(字/秒)	曦云C500出字速度(字/秒)	A100输出质量	曦云C500输出质量	A100首字延迟(秒)	曦云C500首字延迟(秒)	备注
01ai/Yi-1.5-6B-Chat	109.5427	109.5463	85.0000	80.0000	0.0921	0.1178
01ai/Yi-6B-Chat	113.3089	83.3881	85.0000	85.0000	0.0663	0.0986
AI-ModelScope/CausalLM-7B	108.3807	86.3852	77.5000	85.0000	0.0798	0.1090
AI-ModelScope/granite-3.1-3b-a800m-instruct	39.7904	41.1946	53.7500	63.7500	0.2164	0.2640
AI-ModelScope/granite-7b-instruct	106.5448	92.1454	32.5000	27.5000	0.1004	0.1307
AI-ModelScope/Hermes-3-Llama-3.1-8B	104.0769	69.4578	85.0000	85.0000	0.0827	0.1063
AI-ModelScope/mathstral-7B-v0.1	81.6464	56.6757	41.2500	47.5000	0.1011	0.1811
AI-ModelScope/Ministral-8B-Instruct-2410	73.7056	66.5971	85.0000	71.2500	0.0997	0.1629
AI-ModelScope/Mistral-7B-v0.1	74.3966	51.7199	12.5000	15.0000	0.1456	0.1329
AI-ModelScope/TinyLlama-1.1B-Chat-v0.4	58.0479	102.3987	22.5000	22.5000	0.0747	0.1156
allenai/OLMoE-1B-7B-0924-Instruct	66.9544	59.7696	37.5000	42.5000	0.2127	0.2610
BAAI/Finance-llama3_1_8B_instruct	95.4692	65.3383	58.7500	57.5000	0.0812	0.1262
BAAI/Hospitality-llama3_1_8B_instruct	91.7153	62.6340	61.2500	56.2500	0.0650	0.1210
BAAI/Technology-llama3_1_8B_instruct	80.6653	60.4259	32.5000	36.2500	0.0844	0.1147
ByteDance-Seed/Seed-OSS-36B-Instruct	38.2767	17.4623	86.7500	88.5000	0.1400	0.3135
codefuse-ai/TestGPT-7B	68.6996	49.0781	15.0000	15.0000	0.0756	0.1193
CohereLabs/aya-expanse-8B	88.7216	78.2155	86.7500	86.7500	0.0687	3.8936
Cylingo/Xinyuan-LLM-14B-0428	71.7868	49.5901	86.7500	89.2500	0.0860	0.2501
deepseek-ai/deepseek-llm-7b-base	141.4714	93.0921	20.0000	22.5000	0.1231	0.1760
deepseek-ai/deepseek-llm-7b-chat	124.9381	99.4562	85.0000	81.2500	0.0581	0.5611
deepseek-ai/deepseek-moe-16b-chat	68.0789	60.7028	85.0000	68.7500	0.2346	0.2880
Fengshenbang/Ziya-LLaMA-13B-v1	56.6471	39.3946	61.2500	58.7500	0.0616	0.1163
FlagAlpha/Llama3-Chinese-8B-Instruct	100.3664	70.7829	38.7500	30.0000	0.1091	0.1442
HuggingFaceH4/zephyr-7b-beta	97.4954	51.7445	66.2500	47.5000	0.0802	0.1135
iic/WritingBench-Critic-Model-Qwen-7B	125.5657	96.5148	87.5000	87.5000	0.0593	0.1159
InfiniAI/Megrez-3b-Instruct	142.4497	157.9848	85.0000	85.0000	0.0815	0.0946
JunHowie/Qwen3-0.6B-GPTQ-Int4	108.2950	67.6596	58.7500	21.2500	0.0711	0.1206
JunHowie/Qwen3-1.7B-GPTQ-Int8	124.1477	131.5458	71.2500	38.7500	0.0900	0.1848
JunHowie/Qwen3-8B-GPTQ-Int4	82.8148	109.8259	86.7500	52.5000	0.0800	0.1575
Kedreamix/Xinjing-LM	137.3682	150.4469	80.0000	66.2500	0.0710	0.1214
LLM-Research/gemma-2-9b-it	47.0337	47.1786	85.0000	66.2500	0.1125	0.2271
LLM-Research/gemma-3-1b-it	39.6315	43.8421	47.5000	71.2500	0.2256	0.2829
LLM-Research/gemma-3-1b-it	38.3524	33.8285	47.5000	36.2500	0.2080	0.1673
LLM-Research/Llama-3.2-3B	137.6931	145.1816	15.0000	15.0000	0.0891	0.1399
LLM-Research/Llama-3.2-3B-Instruct	107.1550	95.4333	63.7500	68.7500	0.0533	0.1868
LLM-Research/Llama-Guard-3-8B	74.5048	73.5737	33.7500	22.5000	0.0861	0.1248
LLM-Research/Llama3-8B-Chinese-Chat	69.0446	66.6263	85.0000	85.0000	0.0823	0.3008
LLM-Research/Meta-Llama-3-8B-Instruct	167.5121	27.8093	85.0000	92.7500	0.0778	0.2426
LLM-Research/Meta-Llama-3-8B-Instruct-GPTQ	199.0842	175.9663	57.5000	71.2500	0.1028	0.1520
LLM-Research/Meta-Llama-3.1-70B-Instruct-AWQ-INT4	34.9493	18.1700	87.5000	85.0000	0.1327	0.2035
LLM-Research/Meta-Llama-3.1-8B-Instruct-AWQ-INT4	93.1869	77.1093	85.0000	85.0000	0.0665	0.1753
LLM-Research/OpenHermes-2.5-Mistral-7B	78.2992	51.0257	75.0000	75.0000	0.0797	0.1161
LLM-Research/Phi-3-mini-4k-instruct	65.0090	59.8164	45.0000	40.0000	0.0632	0.1004
LLM-Research/Qwen2-7B	170.2887	126.6331	77.5000	78.2500	0.0787	0.1345
LLM-Research/Starling-LM-7B-beta	73.7987	49.5925	66.2500	71.2500	0.0952	0.5098
LLM-Research/tulu-2-dpo-7b	60.8343	47.6062	70.0000	72.5000	0.0618	0.0997
m-a-p/neo_7b_instruct_v0.1	178.2949	137.2681	71.2500	72.5000	0.0839	0.1358
mistralai/Devstral-Small-2507	56.7985	42.3061	88.7500	91.0000	0.0698	0.1193
mistralai/Mistral-Small-24B-Instruct-2501	57.3654	42.7280	86.7500	87.5000	0.0899	0.1340
mlabonne/EvolCodeLlama-7b	53.7677	43.3601	5.0000	5.0000	0.1003	0.1287
modelscope/zephyr-7b-beta	81.4112	52.4249	65.0000	47.5000	0.0835	0.2036
neuralmagic/Meta-Llama-3.1-8B-quantized.w8a8	157.2308	130.4350	10.0000	10.0000	0.0638	0.2007
neuralmagic/Qwen2-1.5B-Instruct-quantized.w8a8	133.1439	136.2916	80.0000	80.0000	0.0564	0.0941
neuralmagic/SmolLM-1.7B-Instruct-quantized.w8a8	105.3004	107.6412	10.0000	10.0000	0.0886	0.0876
nv-community/Minitron-4B-Base	63.8192	69.2841	5.0000	5.0000	0.0706	0.0917
openai-mirror/gpt-oss-safeguard-20b	112.4384	86.9659	61.2500	72.0000	0.2067	0.4129
OpenBMB/MiniCPM-1B-sft-bf16	61.4045	71.6924	66.2500	75.0000	0.0748	0.1010
OpenBMB/MiniCPM-2B-dpo-fp16	61.8623	82.9386	80.0000	71.2500	0.1056	0.1071
OpenBMB/MiniCPM-2B-sft-fp32	74.4678	89.2962	65.0000	55.0000	0.0996	0.1408
OpenBMB/MiniCPM3-4B	28.2210	21.3110	86.7500	86.7500	0.1020	0.1905
OpenBMB/MiniCPM4-0.5B	82.2793	93.0776	23.7500	15.0000	0.0606	0.1143
OpenBMB/MiniCPM4-8B	60.5837	57.0162	87.5000	52.5000	0.0921	0.1482
OpenBMB/MiniCPM4-8B-marlin-vLLM	57.5381	58.2351	85.0000	72.0000	0.1033	0.1107
OpenBMB/MiniCPM4.1-8B	61.4557	51.5120	89.2500	86.7500	0.0784	0.1290
PaddlePaddle/ERNIE-4.5-0.3B-PT	144.5465	185.2054	52.5000	47.5000	0.0837	0.0902
prithivMLmods/Llama-Sentient-3.2-3B-Instruct	147.0239	168.6519	40.0000	40.0000	0.0661	0.1230
prithivMLmods/Qwen-UMLS-7B-Instruct	122.8389	89.8736	55.0000	55.0000	0.0584	0.1061
QuantTrio/Qwen3-30B-A3B-Instruct-2507-GPTQ-Int8	40.1016	22.6581	91.7500	68.0000	0.0915	0.2704
Qwen/Qwen-1_8B-Chat-Int4	536.3839	192.9054	28.7500	25.0000	0.0517	0.0936
Qwen/Qwen-1_8B-Chat-Int8	425.9918	95.9929	42.5000	85.0000	0.0549	0.1754
Qwen/Qwen-14B	64.3123	157.3503	38.7500	33.7500	0.0912	0.0858
Qwen/Qwen-14B-Chat-Int4	84.9682	148.9078	68.7500	85.0000	0.0971	0.1298
Qwen/Qwen-14B-Chat-Int8	103.0940	56.3855	85.0000	21.2500	0.0758	0.1443	该模型在沐曦卡上生成质量要差于A100
Qwen/Qwen-72B-Chat-Int4	55.1679	260.1285	85.0000	47.5000	0.1360	0.1411
Qwen/Qwen-72B-Chat-Int8	45.9768	83.0130	68.0000	58.7500	0.1172	0.1511
Qwen/Qwen-7B-Chat-Int4	119.8944	102.6521	55.0000	55.0000	0.0831	0.1092
Qwen/Qwen-7B-Chat-Int4	117.8753	89.0941	55.0000	37.5000	0.0761	0.1086
Qwen/Qwen-7B-Chat-Int8	128.7233	107.4733	40.0000	80.0000	0.0545	0.1578
Qwen/Qwen1.5-0.5B-Chat-GPTQ-Int8	168.5100	135.2830	33.7500	33.7500	0.0530	0.0869
Qwen/Qwen1.5-1.8B-Chat-GPTQ-Int4	153.0596	132.3559	33.7500	63.7500	0.0557	0.1316
Qwen/Qwen1.5-14B-Chat-AWQ	98.3736	74.0728	91.5000	88.5000	0.0725	0.1188
Qwen/Qwen1.5-14B-Chat-GPTQ-Int4	77.7489	92.8893	88.5000	88.5000	0.0863	0.1030
Qwen/Qwen1.5-14B-Chat-GPTQ-Int8	90.9851	79.5202	88.5000	90.2500	0.0716	0.1136
Qwen/Qwen1.5-32B-Chat-GPTQ-Int4	58.2218	51.3081	92.2500	91.0000	0.0775	0.1392
Qwen/Qwen1.5-72B-Chat-GPTQ-Int4	47.0482	28.5930	92.7500	92.7500	0.1314	0.3305
Qwen/Qwen1.5-7B-Chat-GPTQ-Int4	111.7481	110.7277	88.0000	86.7500	0.0857	0.0978
Qwen/Qwen2-72B-Instruct-GPTQ-Int4	48.2245	29.6541	92.7500	92.7500	0.1091	0.2308
Qwen/Qwen2-7B-Instruct-GPTQ-Int4	129.1925	127.4951	88.5000	88.5000	0.0632	0.1595
Qwen/Qwen2-7B-Instruct-GPTQ-Int8	120.7039	120.5521	89.2500	91.0000	0.0589	0.0928
Qwen/Qwen2.5-0.5B	190.6378	143.7020	12.5000	80.0000	0.0569	0.0966
Qwen/Qwen2.5-0.5B-Instruct	157.0171	74.5845	66.2500	90.0000	0.0524	0.1228
Qwen/Qwen2.5-1.5B-Instruct	132.3709	108.0231	85.0000	85.0000	0.0784	0.1907
Qwen/Qwen2.5-1.5B-Instruct-GPTQ-Int4	99.1006	111.6715	80.0000	85.0000	0.0832	0.0998
Qwen/Qwen2.5-14B-Instruct-1M	73.2996	50.5570	92.7500	91.0000	0.0736	0.1316
Qwen/Qwen2.5-14B-Instruct-GPTQ-Int4	79.5701	72.2741	91.0000	91.0000	0.0738	0.1219
Qwen/Qwen2.5-14B-Instruct-GPTQ-Int8	81.3902	70.3944	91.0000	91.0000	0.0588	0.3164
Qwen/Qwen2.5-32B-Instruct-GPTQ-Int4	60.0678	54.5765	90.5000	91.0000	0.1023	0.1855
Qwen/Qwen2.5-3B	102.8492	124.4283	47.5000	50.0000	0.0830	0.1048
Qwen/Qwen2.5-72B-Instruct-GPTQ-Int4	46.5678	29.5256	91.0000	70.5000	0.1013	0.2163
Qwen/Qwen2.5-7B-Instruct-1M	97.2979	92.4331	89.2500	89.2500	0.0685	0.1017
Qwen/Qwen2.5-7B-Instruct-AWQ	115.5071	113.5233	88.5000	91.0000	0.0599	0.0982
Qwen/Qwen2.5-7B-Instruct-GPTQ-Int4	118.1169	133.1432	87.5000	87.5000	0.0603	0.1551
Qwen/Qwen2.5-Coder-14B-Instruct-AWQ	47.8161	52.9300	87.5000	87.5000	0.0854	0.1162
Qwen/Qwen2.5-Coder-14B-Instruct-GPTQ-Int4	70.0316	68.4585	87.5000	73.7500	0.0694	0.1246
Qwen/Qwen2.5-Coder-32B-Instruct-GPTQ-Int4	60.2992	55.2420	91.0000	91.0000	0.1108	0.2650
Qwen/Qwen2.5-Coder-7B	135.1152	115.0369	46.7500	63.7500	0.0979	0.2146
Qwen/Qwen2.5-Coder-7B-Instruct-GPTQ-Int8	108.4975	100.5769	87.5000	87.5000	0.0646	0.1794
Qwen/Qwen3-0.6B	115.7984	128.5600	56.2500	72.5000	0.0890	0.1073
Qwen/Qwen3-0.6B-GPTQ-Int8	117.4333	95.0022	75.0000	33.7500	0.0891	0.1146
Qwen/Qwen3-1___7B	123.6276	55.1110	81.7500	86.7500	0.0931	0.2069
Qwen/Qwen3-30B-A3B-Base	36.5579	145.1734	86.0000	65.0000	0.1890	0.0983
Qwen/Qwen3-30B-A3B-GPTQ-Int4	41.5809	95.4118	88.5000	66.2500	0.0818	0.1298
Qwen/Qwen3-32B-GPTQ-Int8	53.9982	41.3361	86.7500	86.7500	0.1501	0.2713
Qwen/Qwen3-4B	167.1038	83.0732	87.5000	66.2500	0.0587	0.1159
Qwen/Qwen3-4B	171.1984	105.4759	89.2500	84.2500	0.0598	0.1105
Qwen/Qwen3-4B-AWQ	76.4882	91.4280	86.7500	88.5000	0.0922	0.1245
Qwen/Qwen3-4B-Instruct-2507	65.4432	71.7888	91.7500	91.7500	0.0966	0.1285
Qwen/Qwen3-4B-SafeRL	78.8314	96.6840	87.5000	88.5000	0.0930	0.1174
Qwen/Qwen3-4B-Thinking-2507	73.3587	52.6143	81.7500	88.5000	0.0799	0.2663
Qwen/Qwen3-8B-AWQ	94.2630	67.3837	88.5000	86.7500	0.0990	0.1425
QwenCollection/Hercules-Mini-1.8B	202.5339	202.1224	28.7500	28.7500	0.0517	0.0855
QwenCollection/Ragas-critic-llm-Qwen1.5-GPTQ	170.6923	136.0126	35.0000	25.0000	0.0667	0.1142
RUC-DataLab/DeepAnalyze-8B	84.5026	84.9213	86.7500	83.5000	0.1118	0.1673
shakechen/Llama-2-7b-chat-hf	260.9870	34.1061	61.2500	91.0000	0.0892	0.1960
Shanghai_AI_Laboratory/internlm-20b	54.0570	40.9779	49.2500	71.2500	0.1376	0.2653
Shanghai_AI_Laboratory/internlm-chat-20b	59.3568	44.0583	85.0000	85.0000	0.0789	0.1403
Shanghai_AI_Laboratory/internlm2-chat-1_8b	137.9855	150.8975	22.5000	31.2500	0.0874	0.1612
tclf90/Codestral-22B-v0.1-hf-GPTQ-Int4	69.1961	71.7657	80.0000	70.0000	0.1810	0.2461
tclf90/deepseek-r1-distill-qwen-14b-gptq-int4	61.5092	59.0025	85.0000	85.0000	0.0786	0.1791
tclf90/deepseek-r1-distill-qwen-32b-gptq-int4	63.7344	54.3297	88.0000	86.7500	0.1675	0.2525
tclf90/deepseek-r1-distill-qwen-32b-gptq-int8	59.9314	45.7905	86.7500	86.7500	0.1491	0.2839
tclf90/deepseek-r1-distill-qwen-7b-gptq-int4	134.7569	141.3986	80.0000	85.0000	0.0892	0.2778
tclf90/glm-4-9b-chat-GPTQ-Int4	92.2061	97.6775	89.2500	91.7500	0.0735	0.4389
tclf90/glm-4-9b-chat-GPTQ-Int8	67.4918	93.5178	90.5000	86.7500	0.1037	0.1350
tclf90/Qwen2-14B-merge-GPTQ-Int8	68.0706	73.0019	88.5000	88.5000	0.0741	0.1217
tclf90/qwen2.5-14b-instruct-1m-gptq-int4	80.4909	71.7343	91.0000	91.0000	0.0720	0.1329
tclf90/qwen2.5-72b-instruct-gptq-int4	38.6166	34.9558	91.7500	52.5000	0.1053	0.5529
tclf90/Qwen3-32B-GPTQ-Int4	54.3526	38.4391	86.7500	88.5000	0.1470	0.2276
tclf90/Qwen3-32B-GPTQ-Int4	60.6483	42.8220	86.7500	86.7500	0.1651	0.2262
tclf90/Qwen3-32B-GPTQ-Int8	48.2039	148.4290	86.7500	70.0000	0.1526	0.0950
tclf90/qwq-32b-gptq-int4	60.3297	53.4619	87.5000	86.7500	0.1581	0.4379
tclf90/qwq-32b-gptq-int8	52.0826	41.4700	86.7500	87.5000	0.1853	0.2502
TheBloke/Kimiko-7B-fp16	52.3951	41.4855	15.0000	15.0000	0.0947	0.1223
tiiuae/falcon-7b-instruct	177.4328	147.2327	20.0000	20.0000	0.0906	0.1180
TongyiFinance/Tongyi-Finance-14B-Chat	161.2871	55.1869	52.5000	63.7500	0.0970	0.2396
TongyiFinance/Tongyi-Finance-14B-Chat-Int4	144.2085	114.2191	57.5000	38.7500	0.0592	0.1210
TongyiFinance/Tongyi-Finance-14B-Chat-Int4	147.6310	129.5033	57.5000	57.5000	0.0694	0.1123
UnicomAI/Unichat-llama3.2-Chinese-1B	170.9294	181.6792	26.2500	25.0000	0.0594	0.1311
unsloth/Phi-3.5-mini-instruct	39.4028	38.0348	80.0000	77.5000	0.0716	0.1253
XGenerationLab/XiYanSQL-QwenCoder-3B-2502	111.9181	105.4054	80.0000	85.0000	0.0593	0.1060
Xunzillm4cc/Xunzi-Qwen1.5-4B	99.2706	102.2685	22.5000	25.0000	0.0811	0.0982
Xunzillm4cc/Xunzi-Qwen2-1.5B	125.9543	137.9787	20.0000	17.5000	0.0588	0.0903
ZhipuAI/chatglm3-6b-base	129.0337	40.0233	38.7500	71.2500	0.1349	0.2365
ZhipuAI/glm-4-9b-chat-1m	102.4445	77.4381	89.7500	88.7500	0.0911	0.1357
ZhipuAI/glm-4-9b-chat-hf	109.1148	71.8195	85.0000	85.0000	0.0900	0.1838
ZhipuAI/GLM-Z1-32B-0414	55.9116	41.5245	88.5000	88.5000	0.1048	0.2995
ZhipuAI/GLM-Z1-9B-0414	90.9342	71.1734	85.0000	86.7500	0.0879	0.9642
ZhipuAI/LongWriter-glm4-9b	100.4160	81.4277	86.7500	86.7500	0.1221	0.1758
zpeng1989/TCM_DeepSeek_LLM	98.9219	64.9526	82.5000	80.0000	0.0721	0.1533