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
2026-04-23 14:54:02 +08:00
parent 405e17ea3c
commit b1f0fd6a39
1 changed files with 4 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
config.json
+4 -5
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": "Qwen3.6-35B-A3B-FP8", "selected": "Qwen3.6-27B",
"profiles": { "profiles": {
"Qwen3.6-35B-A3B-FP8": { "Qwen3.6-35B-A3B-FP8": {
"local_path": "Qwen3.6-35B-A3B-FP8", "local_path": "Qwen3.6-35B-A3B-FP8",
@@ -43,8 +43,9 @@
}, },
"Qwen3.6-27B": { "Qwen3.6-27B": {
"local_path": "Qwen3.6-27B", "local_path": "Qwen3.6-27B",
"dtype": "auto", "ctx": "262144",
"ctx": "131072", "max_tokens": "32768",
"max_num_batched_tokens": 16384,
"trust_remote": true, "trust_remote": true,
"enforce_eager": false, "enforce_eager": false,
"valid_tp": [ "valid_tp": [
@@ -52,11 +53,9 @@
2 2
], ],
"max_num_seqs": 32, "max_num_seqs": 32,
"max_tokens": "16384",
"gpu_util": "0.90", "gpu_util": "0.90",
"kv_cache_dtype": "fp8", "kv_cache_dtype": "fp8",
"enable_prefix_caching": true, "enable_prefix_caching": true,
"max_num_batched_tokens": 16384,
"tool_call_parser": "qwen3_coder", "tool_call_parser": "qwen3_coder",
"reasoning_parser": "qwen3", "reasoning_parser": "qwen3",
"enable_auto_tool_choice": true, "enable_auto_tool_choice": true,