vanishingradient commited on
Commit
d91c758
·
verified ·
1 Parent(s): 7ca973d

Add kaggle/KAGGLE_SETUP.md

Browse files
Files changed (1) hide show
  1. kaggle/KAGGLE_SETUP.md +8 -4
kaggle/KAGGLE_SETUP.md CHANGED
@@ -9,11 +9,12 @@
9
  ## 1. Local grader (keep running for the whole kernel)
10
 
11
  ```bash
12
- # Docker Desktop must be up
13
- python scripts/swe_grader_server.py --host 127.0.0.1 --port 8787
14
 
15
  # separate terminal — public HTTPS URL
16
- cloudflared tunnel --url http://127.0.0.1:8787
 
17
  ```
18
 
19
  Optional auth: set `SWE_GRADER_TOKEN` in `.env` (same value in Kaggle secrets).
@@ -67,7 +68,10 @@ kaggle kernels push -p notebooks/
67
 
68
  ## 6. Watch
69
 
70
- - Langfuse: filter `kaggle-ckpt-eval`
 
 
 
71
  - W&B: project `deepagents-rl-coding`, run `kaggle-grpo-swe`
72
  - HF: https://huggingface.co/vanishingradient/deepagents-rl-coding/tree/main
73
  - Local grader logs: each `/grade` call
 
9
  ## 1. Local grader (keep running for the whole kernel)
10
 
11
  ```bash
12
+ # Docker Desktop must be up (port 8787 may be taken — default 8788)
13
+ python scripts/swe_grader_server.py --host 127.0.0.1 --port 8788
14
 
15
  # separate terminal — public HTTPS URL
16
+ ngrok http 8788
17
+ # or: cloudflared tunnel --url http://127.0.0.1:8788
18
  ```
19
 
20
  Optional auth: set `SWE_GRADER_TOKEN` in `.env` (same value in Kaggle secrets).
 
68
 
69
  ## 6. Watch
70
 
71
+ - Langfuse:
72
+ - `grpo-swe-rollout` — **every train/eval episode**: nested `ls` / `read_file` / … tool spans + `reward_grader`
73
+ - `kaggle-ckpt-eval` — checkpoint aggregate (+ links to rollout trace ids)
74
+ - ckpt eval also logs `llm-turn-*` generation spans under each rollout
75
  - W&B: project `deepagents-rl-coding`, run `kaggle-grpo-swe`
76
  - HF: https://huggingface.co/vanishingradient/deepagents-rl-coding/tree/main
77
  - Local grader logs: each `/grade` call