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
2026-04-02 23:10:43 +08:00
parent 9151dcf576
commit 7ab9fe6e59
1 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
config.json
+4 -4
View File
@@ -20,7 +20,7 @@
"revision": "", "revision": "",
"models": { "models": {
"default": "Qwen3.5-35B-A3B-GPTQ-Int4", "default": "Qwen3.5-35B-A3B-GPTQ-Int4",
"selected": "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit", "selected": "Qwen3.5-35B-A3B-FP8",
"profiles": { "profiles": {
"GLM-4.7-Flash-AWQ": { "GLM-4.7-Flash-AWQ": {
"local_path": "GLM-4.7-Flash-AWQ", "local_path": "GLM-4.7-Flash-AWQ",
@@ -73,13 +73,13 @@
}, },
"Qwen3.5-35B-A3B-FP8": { "Qwen3.5-35B-A3B-FP8": {
"local_path": "Qwen3.5-35B-A3B-FP8", "local_path": "Qwen3.5-35B-A3B-FP8",
"dtype": "bfloat16", "dtype": "float16",
"ctx": "65536", "ctx": "65536",
"trust_remote": true, "trust_remote": true,
"valid_tp": [1, 2], "valid_tp": [1, 2],
"max_num_seqs": "8", "max_num_seqs": "8",
"max_tokens": "32768", "max_tokens": "4096",
"gpu_util": "0.98", "gpu_util": "0.92",
"tool_call_parser": "qwen3_coder", "tool_call_parser": "qwen3_coder",
"enable_auto_tool_choice": true, "enable_auto_tool_choice": true,
"served_model_name": "Qwen3.5-35B-A3B-FP8", "served_model_name": "Qwen3.5-35B-A3B-FP8",