Skip to content

Commit ebb0730

Browse files
committed
bench: add leaderboard-safe DeepSeek max manifest
1 parent 023ea2d commit ebb0730

1 file changed

Lines changed: 27 additions & 0 deletions

File tree

Lines changed: 27 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,27 @@
1+
# Leaderboard-eligible persona v0 baseline: Meli solo with DeepSeek V4 Flash
2+
# at max reasoning effort on Databricks. Unlike the local-study manifest, this
3+
# deliberately leaves Harbor's task-specific CPU, memory, and storage requests
4+
# unchanged; leaderboard static validation rejects resource overrides.
5+
schema_version: "1"
6+
condition: tb-persona-v0-deepseek-max-leaderboard
7+
roster:
8+
- id: meli
9+
kind: orchestrator
10+
role: solo
11+
count: 1
12+
endpoint: databricks-deepseek-v4-flash-0731
13+
model_revision: deepseek-v4-flash-20260731
14+
prompt:
15+
path: personas/bench/meli-solo-v0.md
16+
sha256: de6330190fe22cb4b58cd5c2ca1cb7fc080e9b7f97b298c8960e08854e7a7a8b
17+
generation:
18+
thinking_effort: max
19+
prices:
20+
databricks-deepseek-v4-flash-0731:
21+
# Provisional public per-token model rates; receipts preserve measured tokens.
22+
input_per_million_usd: 0.14
23+
cached_input_per_million_usd: 0.14
24+
output_per_million_usd: 0.28
25+
cache_read_rate: 0.0
26+
trial_budget:
27+
timeout_seconds: 36000

0 commit comments

Comments
 (0)