|
6 | 6 | "VLLM_RPC_TIMEOUT": 100000,
|
7 | 7 | "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
|
8 | 8 | "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
|
| 9 | + "VLLM_CPU_NUM_OF_RESERVED_CPU": 0, |
9 | 10 | "VLLM_CPU_KVCACHE_SPACE": 40
|
10 | 11 | },
|
11 | 12 | "server_parameters": {
|
|
18 | 19 | "trust_remote_code": "",
|
19 | 20 | "disable_log_stats": "",
|
20 | 21 | "disable_log_requests": "",
|
| 22 | + "enforce_eager": "", |
21 | 23 | "load_format": "dummy"
|
22 | 24 | },
|
23 | 25 | "client_parameters": {
|
|
35 | 37 | "VLLM_RPC_TIMEOUT": 100000,
|
36 | 38 | "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
|
37 | 39 | "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
|
| 40 | + "VLLM_CPU_NUM_OF_RESERVED_CPU": 0, |
38 | 41 | "VLLM_CPU_KVCACHE_SPACE": 40
|
39 | 42 | },
|
40 | 43 | "server_parameters": {
|
|
47 | 50 | "trust_remote_code": "",
|
48 | 51 | "disable_log_stats": "",
|
49 | 52 | "disable_log_requests": "",
|
| 53 | + "enforce_eager": "", |
50 | 54 | "load_format": "dummy"
|
51 | 55 | },
|
52 | 56 | "client_parameters": {
|
|
64 | 68 | "VLLM_RPC_TIMEOUT": 100000,
|
65 | 69 | "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
|
66 | 70 | "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
|
| 71 | + "VLLM_CPU_NUM_OF_RESERVED_CPU": 0, |
67 | 72 | "VLLM_CPU_KVCACHE_SPACE": 40
|
68 | 73 | },
|
69 | 74 | "server_parameters": {
|
|
76 | 81 | "trust_remote_code": "",
|
77 | 82 | "disable_log_stats": "",
|
78 | 83 | "disable_log_requests": "",
|
| 84 | + "enforce_eager": "", |
79 | 85 | "load_format": "dummy"
|
80 | 86 | },
|
81 | 87 | "client_parameters": {
|
|
93 | 99 | "VLLM_RPC_TIMEOUT": 100000,
|
94 | 100 | "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
|
95 | 101 | "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
|
| 102 | + "VLLM_CPU_NUM_OF_RESERVED_CPU": 0, |
96 | 103 | "VLLM_CPU_KVCACHE_SPACE": 40
|
97 | 104 | },
|
98 | 105 | "server_parameters": {
|
|
106 | 113 | "enable_chunked_prefill": "",
|
107 | 114 | "disable_log_stats": "",
|
108 | 115 | "disable_log_requests": "",
|
| 116 | + "enforce_eager": "", |
109 | 117 | "load_format": "dummy"
|
110 | 118 | },
|
111 | 119 | "client_parameters": {
|
|
125 | 133 | "VLLM_RPC_TIMEOUT": 100000,
|
126 | 134 | "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
|
127 | 135 | "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
|
| 136 | + "VLLM_CPU_NUM_OF_RESERVED_CPU": 0, |
128 | 137 | "VLLM_CPU_KVCACHE_SPACE": 40
|
129 | 138 | },
|
130 | 139 | "server_parameters": {
|
|
138 | 147 | "enable_chunked_prefill": "",
|
139 | 148 | "disable_log_stats": "",
|
140 | 149 | "disable_log_requests": "",
|
| 150 | + "enforce_eager": "", |
141 | 151 | "load_format": "dummy"
|
142 | 152 | },
|
143 | 153 | "client_parameters": {
|
|
0 commit comments