Hugging Face's logo

Datasets:
FineEnvs
/
MiMo-V2.6-RL-harbor-code

RL Environment
Tasks:
Other
Languages:
English
Chinese
Size:
1K<n<10K
Tags:
reinforcement-learning
harbor
agent
mimo
mimo-v2.6-rl
code
Libraries:
Harbor
License:
Dataset card Files
xet
 Community
AdithyaSK HF Staff commited on
Commit
5746e2f
·
verified ·
1 Parent(s): d3c288c

Parity results: parity_experiment.json

Browse files
Files changed (1) hide show
  1. parity_experiment.json +114 -0
parity_experiment.json ADDED
@@ -0,0 +1,114 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "adapter_name": "mimo-v2.6-rl",
4
+ "agent": "opencode@1.18.32 (mimo_opencode)",
5
+ "model": "zai-org/GLM-5.3 via deepinfra, reasoning_effort low",
6
+ "date": "2026-09-27",
7
+ "adapted_benchmark_size": 7780,
8
+ "parity_benchmark_size": 36,
9
+ "number_of_runs": 3,
10
+ "notes": "Original side: the MiMo RL Environment Explorer runner (Xiaomi's harness: XiaomiMiMo/verl a2ad9f6 + mimoagent 467f0a1). Parity subset: 6 tasks per dataset, fixed by sha256(task id) (--split parity). Same agent version, model, provider, thinking level, step limits, 65,536-token replies, judges (Inkling, Llama-4-Maverick) and sandbox flavors on both sides. Per-run score = mean reward over the tasks that side scored; trials that failed for infrastructure reasons (sandbox API timeouts) were rerun, and arvo_49407 run 1 failed on both sides (the agent starved a 2-vCPU sandbox) and is excluded on both. Music differs by design: one completion on the original side, up to 4 write-only steps in Harbor.",
11
+ "original_parity_repo": "https://github.com/adithya-s-k/FineEnvs/tree/mimo-explorer-rollouts/mimo-explorer",
12
+ "adapter_pr": [],
13
+ "dataset_pr": [],
14
+ "parity_pr": [],
15
+ "metrics": [
16
+ {
17
+ "benchmark_name": "MiMo-V2.6-RL Code",
18
+ "metric": "mean reward",
19
+ "original": "0.556 ± 0.056",
20
+ "harbor": "0.556 ± 0.056",
21
+ "original_runs": [
22
+ 0.6667,
23
+ 0.5,
24
+ 0.5
25
+ ],
26
+ "harbor_runs": [
27
+ 0.6667,
28
+ 0.5,
29
+ 0.5
30
+ ]
31
+ },
32
+ {
33
+ "benchmark_name": "MiMo-V2.6-RL Cyber",
34
+ "metric": "mean reward",
35
+ "original": "0.822 ± 0.011",
36
+ "harbor": "0.833 ± 0.096",
37
+ "original_runs": [
38
+ 0.8,
39
+ 0.8333,
40
+ 0.8333
41
+ ],
42
+ "harbor_runs": [
43
+ 1.0,
44
+ 0.6667,
45
+ 0.8333
46
+ ]
47
+ },
48
+ {
49
+ "benchmark_name": "MiMo-V2.6-RL General",
50
+ "metric": "mean reward",
51
+ "original": "0.844 ± 0.008",
52
+ "harbor": "0.844 ± 0.024",
53
+ "original_runs": [
54
+ 0.8423,
55
+ 0.8315,
56
+ 0.8591
57
+ ],
58
+ "harbor_runs": [
59
+ 0.8899,
60
+ 0.8107,
61
+ 0.8315
62
+ ]
63
+ },
64
+ {
65
+ "benchmark_name": "MiMo-V2.6-RL Music",
66
+ "metric": "mean reward",
67
+ "original": "0.111 ± 0.063",
68
+ "harbor": "0.317 ± 0.093",
69
+ "original_runs": [
70
+ 0.0,
71
+ 0.1135,
72
+ 0.2193
73
+ ],
74
+ "harbor_runs": [
75
+ 0.1382,
76
+ 0.4493,
77
+ 0.3633
78
+ ]
79
+ },
80
+ {
81
+ "benchmark_name": "MiMo-V2.6-RL Terminal",
82
+ "metric": "mean reward",
83
+ "original": "0.278 ± 0.056",
84
+ "harbor": "0.111 ± 0.056",
85
+ "original_runs": [
86
+ 0.3333,
87
+ 0.3333,
88
+ 0.1667
89
+ ],
90
+ "harbor_runs": [
91
+ 0.1667,
92
+ 0.1667,
93
+ 0.0
94
+ ]
95
+ },
96
+ {
97
+ "benchmark_name": "MiMo-V2.6-RL Webdev",
98
+ "metric": "mean reward",
99
+ "original": "0.749 ± 0.049",
100
+ "harbor": "0.738 ± 0.054",
101
+ "original_runs": [
102
+ 0.6505,
103
+ 0.8033,
104
+ 0.7933
105
+ ],
106
+ "harbor_runs": [
107
+ 0.7994,
108
+ 0.6306,
109
+ 0.7852
110
+ ]
111
+ }
112
+ ]
113
+ }
114
+ ]