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:
chenjw28
2026-04-02 14:57:07 +08:00
parent 87aaef7f0a
commit fc86b4426b
2 changed files with 10 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.gitignore
+2
View File
@@ -0,0 +1,2 @@
.idea/
venv/
config.json
+8 -3
View File
@@ -87,10 +87,11 @@
}, },
"Qwen3-Next-80B-A3B-Instruct-AWQ-4bit": { "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit": {
"local_path": "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit", "local_path": "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit",
"ctx": "262144", "dtype": "auto",
"ctx": "131072",
"trust_remote": true, "trust_remote": true,
"valid_tp": [2], "valid_tp": [2],
"max_num_seqs": "4", "max_num_seqs": "6",
"max_tokens": "4096", "max_tokens": "4096",
"gpu_util": "0.90", "gpu_util": "0.90",
"enforce_eager": false, "enforce_eager": false,
@@ -101,7 +102,11 @@
"reasoning_parser": "qwen3", "reasoning_parser": "qwen3",
"enable_auto_tool_choice": true, "enable_auto_tool_choice": true,
"served_model_name": "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit", "served_model_name": "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit",
"hf_model_id": "cpatonn/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit" "hf_model_id": "cpatonn/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit",
"temperature": 0.7,
"top_p": 0.8,
"top_k": 20,
"min_p": 0
} }
} }
} }