cjw/rocm_vllm_nightly
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

x

Browse Source
This commit is contained in:
SZCJW
2026-03-29 17:41:28 +08:00
parent 7ff076530a
commit d84cdc00e5
1 changed files with 7 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
config.json
+7 -7
View File
@@ -24,7 +24,7 @@
"ctx": "24576", "ctx": "24576",
"trust_remote": true, "trust_remote": true,
"valid_tp": [2], "valid_tp": [2],
"max_num_seqs": "32", "max_num_seqs": "8",
"max_tokens": "16384", "max_tokens": "16384",
"gpu_util": "0.98", "gpu_util": "0.98",
"enforce_eager": false, "enforce_eager": false,
@@ -41,7 +41,7 @@
"ctx": "32768", "ctx": "32768",
"trust_remote": true, "trust_remote": true,
"valid_tp": [1, 2], "valid_tp": [1, 2],
"max_num_seqs": "64", "max_num_seqs": "8",
"max_tokens": "32768", "max_tokens": "32768",
"gpu_util": "0.98", "gpu_util": "0.98",
"tool_call_parser": "qwen3_xml", "tool_call_parser": "qwen3_xml",
@@ -54,7 +54,7 @@
"ctx": "65536", "ctx": "65536",
"trust_remote": true, "trust_remote": true,
"valid_tp": [1, 2], "valid_tp": [1, 2],
"max_num_seqs": "64", "max_num_seqs": "8",
"max_tokens": "32768", "max_tokens": "32768",
"gpu_util": "0.98", "gpu_util": "0.98",
"tool_call_parser": "qwen3_xml", "tool_call_parser": "qwen3_xml",
@@ -69,9 +69,9 @@
"ctx": "65536", "ctx": "65536",
"trust_remote": true, "trust_remote": true,
"valid_tp": [1, 2], "valid_tp": [1, 2],
"max_num_seqs": "64", "max_num_seqs": "8",
"max_tokens": "32768", "max_tokens": "4096",
"gpu_util": "0.98", "gpu_util": "0.92",
"tool_call_parser": "qwen3_xml", "tool_call_parser": "qwen3_xml",
"enable_auto_tool_choice": true, "enable_auto_tool_choice": true,
"served_model_name": "Qwen3.5-35B-A3B-GPTQ-Int4", "served_model_name": "Qwen3.5-35B-A3B-GPTQ-Int4",
@@ -83,7 +83,7 @@
"ctx": "65536", "ctx": "65536",
"trust_remote": true, "trust_remote": true,
"valid_tp": [1, 2], "valid_tp": [1, 2],
"max_num_seqs": "64", "max_num_seqs": "8",
"max_tokens": "32768", "max_tokens": "32768",
"gpu_util": "0.98", "gpu_util": "0.98",
"tool_call_parser": "qwen3_xml", "tool_call_parser": "qwen3_xml",