ziyuzhou02 commited on
Commit
e6f4fb7
·
verified ·
1 Parent(s): 42fa53c

UI-only mode: update README.md

Browse files
Files changed (1) hide show
  1. README.md +10 -13
README.md CHANGED
@@ -13,21 +13,18 @@ python_version: 3.11.9
13
 
14
  # TSFM Realworld Bench Leaderboard
15
 
16
- Live benchmark: every **1 hour** the Space collects TS-Bench data, runs all TSFM models via TSFM.ai, and updates the leaderboard.
17
 
18
- ## Enable live evaluation (required)
 
19
 
20
- In **Settings → Repository secrets**, add:
21
 
22
- | Secret | Value |
23
- |---|---|
24
- | `TSFM_API_KEY` | Your key from [tsfm.ai](https://tsfm.ai/) |
 
25
 
26
- Optional variables (Settings → Variables):
27
-
28
- | Variable | Default | Purpose |
29
- |---|---|---|
30
- | `TSFM_BENCH_INTERVAL_SECONDS` | `3600` | Collect + eval cycle (1 hour) |
31
-
32
- Without `TSFM_API_KEY`, the leaderboard only reloads static CSV files and **does not** call the API.
33
 
 
 
13
 
14
  # TSFM Realworld Bench Leaderboard
15
 
16
+ **UI-only** leaderboard for the TSFM Realworld Benchmark.
17
 
18
+ Evaluations run **locally** (or via GitHub Actions) every hour and push `results/` here.
19
+ The Space does **not** call TSFM.ai on startup — it only displays CSV results.
20
 
21
+ ## Update results (local)
22
 
23
+ ```bash
24
+ # .env: TSFM_API_KEY, HF_TOKEN
25
+ python scripts/run_online_daemon.py --once
26
+ ```
27
 
28
+ ## Optional: run eval inside Space (not recommended)
 
 
 
 
 
 
29
 
30
+ Set `TSFM_LIVE_EVAL=1` and `TSFM_API_KEY` in Space secrets. This blocks startup for ~3 min per eval cycle.