gahow/aituner
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Switch 27B TP A/B to length-aware TTFT SLO (4s + L_in/8k), widen search

Browse Source 
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
Gahow Wang
2026-06-15 20:35:23 +08:00
parent ed2bbe0323
commit 7678c7d5e8
1 changed files with 4 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
configs/examples/dash0_qwen27b_tp_ab.json
View File

@@ -145,20 +145,9 @@
"slo": { "slo": {
"target_pass_rate": 0.95, "target_pass_rate": 0.95,
"ttft_rule": { "ttft_rule": {
"kind": "step_ms", "kind": "linear_ms",
"buckets": [ "intercept_ms": 4000,
{ "per_token_ms": 0.125
"max_input_tokens": 4096,
"threshold_ms": 2000
},
{
"max_input_tokens": 32768,
"threshold_ms": 4000
},
{
"threshold_ms": 6000
}
]
}, },
"tpot_rule": { "tpot_rule": {
"kind": "fixed_ms", "kind": "fixed_ms",
@@ -167,7 +156,7 @@
}, },
"search": { "search": {
"low": 0.0, "low": 0.0,
"high": 0.25, "high": 0.5,
"tolerance": 0.001, "tolerance": 0.001,
"max_probes": 7, "max_probes": 7,
"sample_seed": 20260325 "sample_seed": 20260325