Dataset Viewer
Auto-converted to Parquet Duplicate
model
stringclasses
4 values
task
stringclasses
3 values
seed
int64
13
777
score
float64
0.58
0.99
slo_recovery
float64
0.6
1
action_efficiency
float64
0.8
1
time_efficiency
float64
0.27
0.95
steps_used
int64
1
50
terminated
bool
1 class
termination_reason
stringclasses
2 values
source_note
stringclasses
3 values
untrained-llama
easy
13
0.97
1
1
0.8
2
true
resolved
logged in _diag/eval_done.log
untrained-llama
easy
99
0.764
0.8
1
0.36
10
true
timeout
logged in _diag/eval_done.log
untrained-llama
easy
777
0.7256
0.75
1
0.3375
10
true
timeout
logged in _diag/eval_done.log
untrained-llama
medium
13
0.9925
1
1
0.95
1
true
resolved
logged in _diag/eval_done.log
untrained-llama
medium
99
0.8407
0.9
1
0.405
20
true
timeout
logged in _diag/eval_done.log
untrained-llama
medium
777
0.9925
1
1
0.95
1
true
resolved
logged in _diag/eval_done.log
untrained-llama
hard
13
0.6105
0.6
1
0.27
50
true
timeout
logged in _diag/eval_done.log
untrained-llama
hard
99
0.6384
0.6364
1
0.2864
50
true
timeout
logged in _diag/eval_done.log
untrained-llama
hard
777
0.6617
0.6667
1
0.3
50
true
timeout
logged in _diag/eval_done.log
sft-primary
easy
13
0.97
1
1
0.8
2
true
resolved
logged in _diag/eval_done.log
sft-primary
easy
99
0.764
0.8
1
0.36
10
true
timeout
logged in _diag/eval_done.log
sft-primary
easy
777
0.7256
0.75
1
0.3375
10
true
timeout
logged in _diag/eval_done.log
sft-primary
medium
13
0.9925
1
1
0.95
1
true
resolved
logged in _diag/eval_done.log
sft-primary
medium
99
0.8407
0.9
1
0.405
20
true
timeout
logged in _diag/eval_done.log
sft-primary
medium
777
0.9925
1
1
0.95
1
true
resolved
logged in _diag/eval_done.log
sft-primary
hard
13
0.5805
0.6
0.8
0.27
50
true
timeout
logged in _diag/eval_done.log
sft-primary
hard
99
0.6384
0.6364
1
0.2864
50
true
timeout
logged in _diag/eval_done.log
sft-primary
hard
777
0.6617
0.6667
1
0.3
50
true
timeout
logged in _diag/eval_done.log
grpo-primary
easy
13
0.97
1
1
0.8
2
true
resolved
final confirmed replay: flat vs untrained baseline
grpo-primary
easy
99
0.764
0.8
1
0.36
10
true
timeout
final confirmed replay: flat vs untrained baseline
grpo-primary
easy
777
0.7256
0.75
1
0.3375
10
true
timeout
final confirmed replay: flat vs untrained baseline
grpo-primary
medium
13
0.9925
1
1
0.95
1
true
resolved
final confirmed replay: flat vs untrained baseline
grpo-primary
medium
99
0.8407
0.9
1
0.405
20
true
timeout
final confirmed replay: flat vs untrained baseline
grpo-primary
medium
777
0.9925
1
1
0.95
1
true
resolved
final confirmed replay: flat vs untrained baseline
grpo-primary
hard
13
0.6105
0.6
1
0.27
50
true
timeout
final confirmed replay: flat vs untrained baseline
grpo-primary
hard
99
0.6384
0.6364
1
0.2864
50
true
timeout
final confirmed replay: flat vs untrained baseline
grpo-primary
hard
777
0.6617
0.6667
1
0.3
50
true
timeout
final confirmed replay: flat vs untrained baseline
grpo-stability
easy
13
0.97
1
1
0.8
2
true
resolved
partial final eval only: Easy/Medium
grpo-stability
easy
99
0.764
0.8
1
0.36
10
true
timeout
partial final eval only: Easy/Medium
grpo-stability
easy
777
0.7256
0.75
1
0.3375
10
true
timeout
partial final eval only: Easy/Medium
grpo-stability
medium
13
0.9925
1
1
0.95
1
true
resolved
partial final eval only: Easy/Medium
grpo-stability
medium
99
0.8407
0.9
1
0.405
20
true
timeout
partial final eval only: Easy/Medium
grpo-stability
medium
777
0.9925
1
1
0.95
1
true
resolved
partial final eval only: Easy/Medium

SevZero held-out eval results

Per-seed held-out evaluation rows for the OpenEnv India Hackathon 2026 SevZero submission.

  • Environment: Mist-ic/sevzero-env
  • Seeds: 13, 99, 777
  • Tasks: easy, medium, hard
  • Score fields: final grader score plus slo_recovery, action_efficiency, time_efficiency, steps_used, and termination metadata.

Coverage

Core comparison rows are complete for:

  • untrained-llama
  • sft-primary
  • grpo-primary

grpo-primary is flat against the untrained baseline on the final held-out replay: Easy 0.8199, Medium 0.9419, Hard 0.6369, Mean 0.7996.

grpo-stability is intentionally partial. Easy and Medium rows are included; Hard-tier eval was deferred because hard episodes run to the 50-action ceiling and did not fit the final compute budget.

Aggregate table

Model Easy Medium Hard Mean
Untrained Llama-3.1-8B-Instruct 0.8199 0.9419 0.6369 0.7996
SFT-primary 0.8199 0.9419 0.6269 0.7962
GRPO-primary 0.8199 0.9419 0.6369 0.7996
GRPO-stability 0.8199 0.9419 n/a partial

The full write-up is in the Space blog: https://huggingface.co/spaces/Mist-ic/sevzero-env/blob/main/BLOG.md

Downloads last month
43

Space using Mist-ic/sevzero-eval-results 1