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
2026-04-01 01:19:14 +08:00
parent 1f807347ee
commit 6574035969
1 changed files with 18 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files
config.json
+18 -18
View File
@@ -22,24 +22,6 @@
"default": "Qwen3.5-35B-A3B-GPTQ-Int4",
"selected": "Qwen3.5-35B-A3B-GPTQ-Int4",
"profiles": {
"Qwen3-Next-80B-A3B-Instruct-AWQ-4bit": {
"local_path": "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit",
"ctx": "24576",
"trust_remote": true,
"valid_tp": [2],
"max_num_seqs": "8",
"max_tokens": "16384",
"gpu_util": "0.98",
"enforce_eager": false,
"env": {
"VLLM_USE_TRITON_AWQ": "1"
},
"tool_call_parser": "qwen3_coder",
"reasoning_parser": "qwen3",
"enable_auto_tool_choice": true,
"served_model_name": "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit",
"hf_model_id": "cpatonn/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit"
},
"GLM-4.7-Flash-AWQ": {
"local_path": "GLM-4.7-Flash-AWQ",
"ctx": "32768",
@@ -102,6 +84,24 @@
"enable_auto_tool_choice": true,
"served_model_name": "Qwen3.5-35B-A3B-FP8",
"hf_model_id": "Qwen/Qwen3.5-35B-A3B-FP8"
},
"Qwen3-Next-80B-A3B-Instruct-AWQ-4bit": {
"local_path": "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit",
"ctx": "65536",
"trust_remote": true,
"valid_tp": [2],
"max_num_seqs": "4",
"max_tokens": "8192",
"gpu_util": "0.90",
"enforce_eager": false,
"env": {
"VLLM_USE_TRITON_AWQ": "1"
},
"tool_call_parser": "qwen3_coder",
"reasoning_parser": "qwen3",
"enable_auto_tool_choice": true,
"served_model_name": "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit",
"hf_model_id": "cpatonn/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit"
}
}
}