Dataset Viewer
Auto-converted to Parquet Duplicate
model
stringclasses
8 values
file
stringclasses
8 values
file_sha256
stringclasses
8 values
task
stringclasses
100 values
seed
int64
0
2
attempted_at_8gb_ctx
bool
2 classes
source
stringclasses
3 values
pass_v1
bool
2 classes
pass_v2
bool
2 classes
graded_by
stringclasses
2 values
prompt_tokens
int64
238
1.28k
generated_tokens
int64
42
262k
finish_reason
stringclasses
2 values
looped_v3
bool
2 classes
ctx
int64
25.1k
262k
⌀
scored_ctx
int64
25.1k
262k
card
stringclasses
7 values
runtime
stringclasses
3 values
flags
stringclasses
5 values
sampling
stringclasses
1 value
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
1873_d
0
true
run1
true
true
kit-v4 regrade
654
2,995
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
1873_d
1
true
run1
true
true
kit-v4 regrade
654
600
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
1873_d
2
true
run1
true
true
kit-v4 regrade
654
4,663
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
2808
0
true
run1
true
true
kit-v4 regrade
497
6,187
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
2808
1
true
run1
false
false
kit-v4 regrade
497
834
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
2808
2
true
run1
false
false
kit-v4 regrade
497
27,585
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
2848
0
true
run1
true
true
kit-v4 regrade
334
923
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
2848
1
true
run1
true
true
kit-v4 regrade
334
709
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
2848
2
true
run1
true
true
kit-v4 regrade
334
846
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
2850
0
true
run1
false
false
kit-v4 regrade
451
11,512
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
2850
1
true
run1
true
true
kit-v4 regrade
451
2,639
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
2850
2
true
run1
false
false
kit-v4 regrade
451
8,722
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
2857
0
true
run1
false
false
kit-v4 regrade
423
1,150
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
2857
1
true
run1
false
false
kit-v4 regrade
423
1,908
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
2857
2
true
run1
false
false
kit-v4 regrade
423
1,054
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
2954
0
true
run1
true
true
kit-v4 regrade
607
1,188
stop
true
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
2954
1
true
run1
false
false
kit-v4 regrade
607
1,524
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
2954
2
true
run1
true
true
kit-v4 regrade
607
623
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
2955
0
true
run1
true
true
kit-v4 regrade
434
697
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
2955
1
true
run1
true
true
kit-v4 regrade
434
742
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
2955
2
true
run1
true
true
kit-v4 regrade
434
591
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3034
0
true
run1
true
true
kit-v4 regrade
370
624
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3034
1
true
run1
true
true
kit-v4 regrade
370
403
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3034
2
true
run1
true
true
kit-v4 regrade
370
521
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3046
0
true
run1
true
true
kit-v4 regrade
465
5,645
stop
true
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3046
1
true
run1
false
false
kit-v4 regrade
465
1,078
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3046
2
true
run1
true
true
kit-v4 regrade
465
11,788
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3163
0
true
run1
true
true
kit-v4 regrade
488
532
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3163
1
true
run1
true
true
kit-v4 regrade
488
487
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3163
2
true
run1
true
true
kit-v4 regrade
488
662
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3193
0
true
run1
true
true
kit-v4 regrade
598
938
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3193
1
true
run1
true
true
kit-v4 regrade
598
1,105
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3193
2
true
run1
true
true
kit-v4 regrade
598
928
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3194
0
true
run1
true
true
kit-v4 regrade
377
254
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3194
1
true
run1
true
true
kit-v4 regrade
377
337
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3194
2
true
run1
true
true
kit-v4 regrade
377
230
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3195
0
true
run1
true
true
kit-v4 regrade
442
1,039
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3195
1
true
run1
true
true
kit-v4 regrade
442
426
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3195
2
true
run1
true
true
kit-v4 regrade
442
870
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3199
0
true
run1
true
true
kit-v4 regrade
391
634
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3199
1
true
run1
false
false
kit-v4 regrade
391
972
stop
true
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3199
2
true
run1
true
true
kit-v4 regrade
391
3,887
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3206
0
true
run1
true
true
kit-v4 regrade
458
785
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3206
1
true
run1
true
true
kit-v4 regrade
458
575
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3206
2
true
run1
true
true
kit-v4 regrade
458
292
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3229
0
true
run1
true
true
kit-v4 regrade
824
18,873
stop
true
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3229
1
true
run1
false
false
kit-v4 regrade
824
1,014
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3229
2
true
run1
true
true
kit-v4 regrade
824
10,333
stop
true
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3265
0
true
run1
true
true
kit-v4 regrade
546
7,116
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3265
1
true
run1
true
true
kit-v4 regrade
546
1,061
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3265
2
true
run1
true
true
kit-v4 regrade
546
13,742
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3299
0
true
run1
true
true
kit-v4 regrade
475
13,347
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3299
1
true
run1
true
true
kit-v4 regrade
475
22,383
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3299
2
true
run1
true
true
kit-v4 regrade
475
9,412
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3308
0
true
run1
true
true
kit-v4 regrade
389
2,541
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3308
1
true
run1
false
false
kit-v4 regrade
389
6,901
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3308
2
true
run1
false
false
kit-v4 regrade
389
1,480
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3320
0
true
run1
true
true
kit-v4 regrade
425
617
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3320
1
true
run1
true
true
kit-v4 regrade
425
563
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3320
2
true
run1
true
true
kit-v4 regrade
425
640
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3353
0
true
run1
true
true
kit-v4 regrade
315
670
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3353
1
true
run1
true
true
kit-v4 regrade
315
483
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3353
2
true
run1
true
true
kit-v4 regrade
315
549
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3360
0
true
run1
true
true
kit-v4 regrade
474
1,944
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3360
1
true
run1
false
false
kit-v4 regrade
474
661
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3360
2
true
run1
true
true
kit-v4 regrade
474
2,728
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3387
0
true
run1
false
false
kit-v4 regrade
449
2,913
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3387
1
true
run1
true
true
kit-v4 regrade
449
18,000
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3387
2
true
run1
true
true
kit-v4 regrade
449
2,698
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3406
0
true
run1
true
true
kit-v4 regrade
629
9,357
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3406
1
true
run1
false
false
kit-v4 regrade
629
3,470
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3406
2
true
run1
false
false
kit-v4 regrade
629
12,067
stop
true
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3438
0
true
run1
false
false
kit-v4 regrade
620
9,126
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3438
1
true
run1
false
false
kit-v4 regrade
620
1,211
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3438
2
true
run1
false
false
kit-v4 regrade
620
3,030
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3450
0
true
run1
true
true
kit-v4 regrade
593
1,752
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3450
1
true
run1
true
true
kit-v4 regrade
593
1,473
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3450
2
true
run1
true
true
kit-v4 regrade
593
1,601
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3455
0
true
run1
true
true
kit-v4 regrade
387
6,876
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3455
1
true
run1
true
true
kit-v4 regrade
387
18,738
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3455
2
true
run1
true
true
kit-v4 regrade
387
1,958
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3468
0
true
run1
true
true
kit-v4 regrade
363
936
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3468
1
true
run1
true
true
kit-v4 regrade
363
514
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3468
2
true
run1
true
true
kit-v4 regrade
363
417
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3508
0
true
run1
true
true
kit-v4 regrade
360
1,168
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3508
1
true
run1
true
true
kit-v4 regrade
360
489
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3508
2
true
run1
true
true
kit-v4 regrade
360
1,204
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3518
0
true
run1
true
true
kit-v4 regrade
485
9,153
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3518
1
true
run1
true
true
kit-v4 regrade
485
4,681
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3518
2
true
run1
true
true
kit-v4 regrade
485
71,976
stop
true
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3535
0
true
run1
true
true
kit-v4 regrade
450
14,323
stop
true
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3535
1
true
run1
true
true
kit-v4 regrade
450
8,074
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3535
2
true
run1
false
false
kit-v4 regrade
450
5,590
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3553
0
true
run1
true
true
kit-v4 regrade
345
671
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3553
1
true
run1
true
true
kit-v4 regrade
345
371
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3553
2
true
run1
true
true
kit-v4 regrade
345
417
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3608
0
true
run1
false
false
kit-v4 regrade
678
37,234
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3608
1
true
run1
false
false
kit-v4 regrade
678
34,230
stop
false
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3608
2
true
run1
false
false
kit-v4 regrade
678
48,376
stop
true
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
empero Qwen3.8-9B-Distill Q4_K_M
Qwen3.8-9B-Q4_K_M.gguf
df13d66021cef676f82be74053220fd75af6bf2a6a7fb77f5222ab9e50744a7a
3637
0
true
run1
true
true
kit-v4 regrade
443
11,519
stop
true
131,072
103,936
RTX 3090 24 GB
upstream llama.cpp 52975342f
temp 1.0, top_p 0.95, top_k 20, min_p 0, seed = run seed, enable_thinking, reasoning_effort xhigh, no max_tokens
End of preview. Expand in Data Studio

LiveCodeBench v6 at 8 GB context: per-task-seed results

Per-task-seed pass/fail for qwen3.8-27b-Swift1.5-Uncensored-ATX-AI2-8GB (an 8.39 GB GGUF of a 27B model, run with llamAmpere) and small GGUF models that also fit an 8 GB card. Every model got the same 100 pinned LiveCodeBench v6 problems, 3 seeds each, so 300 task-seeds per model. Problem statements are not included. Tasks are identified by their LiveCodeBench id (task) and the run seed (seed).

Results

Model 8 GB context Strict v1 /300 Fixed v2 /300 v1 % Ours minus model, v1 points [95% CI], p v2 points [95% CI], p
Qwen3.8-27B ATX AI2 8 GB (ours) 204,800 244 253 81.3% - -
Mellum2.1-12B-A2.5B-Thinking IQ3_XS (bartowski) 128,000 246 257 82.0% -0.7 [-5.0, +3.7], p=0.88 -1.3 [-5.7, +3.0], p=0.66
Mellum2.1-12B-A2.5B-Thinking MTP-MXFP4_MOE (JetBrains) 25,344 236 245 78.7% +2.7 [-2.0, +7.7], p=0.37 +2.7 [-2.0, +8.0], p=0.37
LFM3-1.7B Q8_0 (LiquidAI) 262,144 215 n/a 71.7% +9.7 [+3.3, +16.3], p=0.0055 n/a
Ornith-1.5-9B Q4_K_M (ornith-ai) 103,936 213 222 71.0% +10.3 [+5.3, +15.3], p=6.6e-05 +10.3 [+5.3, +15.3], p=6.6e-05
Qwen3.5-9B Q4_K_M + MTP (unsloth) † 262,144 on a 24 GB card (8 GB: 67,072) 207 214 69.0% +12.3 [+6.3, +18.3], p=0.00011 +13.0 [+7.0, +19.3], p=5.9e-05
TwIL-8B Q4_K_M verbose (webAI) 25,088 206 212 68.7% +12.7 [+6.0, +20.0], p=0.00043 +13.7 [+7.0, +21.0], p=0.00018
Qwen3.8-9B-Distill Q4_K_M (empero) 103,936 175 180 58.3% +23.0 [+16.3, +30.0], p=1.9e-10 +24.3 [+17.7, +31.3], p=3e-11
Unsloth Qwen3.8-27B UD-IQ2_S * does not fit (OOM all-GPU; 4,096 with CPU offload) - - - - -

The two Mellum2.1 files average 241/300, a tie with ours (the CI crosses 0). Every other model is significantly below ours.

† Qwen3.5-9B was run without a context cap on 24 GB cards (262,144 per slot), so it is not an 8 GB result. On an 8 GB card it fits only up to 67,072 tokens of context.

* Does not fit an 8 GB card at a useful context, so it has no LiveCodeBench run.

Rules

  • 8 GB context = the largest context that fits a real RTX 3070 8 GB with peak GPU memory at or under 7,692 MiB (8,192 minus 500). For the other models it was measured with upstream llama.cpp --fit on -fitt 500, -ngl 99, q8_0 K/V cache, one slot. For ours it is 204,800 tokens, with a peak of 7,607 MiB on an RTX 3070.
  • No credit for work not done. Each model attempted all 300 task-seeds. An answer whose prompt plus generated tokens is longer than the model's 8 GB context counts as a fail, as does a task-seed that was never generated (none were).
  • Some answers were generated on 24 GB cards with a larger context and then scored at the model's 8 GB context: anything that would not have fit counts as a fail. Our answers were all generated this way, on an RTX 3090 Ti with llamAmpere v0.7, and 3 of ours are counted as fails for being longer than 204,800 tokens. The card, ctx and scored_ctx fields show this for every row.

Harness (the same for every model)

  • LiveCodeBench generic prompt and system prompt; the last fenced code block in the reply is the solution.
  • Sampling: temperature 1.0, top_p 0.95, top_k 20, min_p 0, per-request seed = run seed (0/1/2), thinking on with reasoning effort xhigh, no max_tokens (generation is limited only by the context).
  • Grading: strict v1 and fixed v2 graders, CPython 3.11.15. An answer cut by the length limit is a fail. Where an old answer's solution file was not kept (LFM3 rows from the earlier run), pass_v2 is null and pass_v1 is that run's strict v1 result.
  • Other models ran on upstream llama.cpp (commit in runtime, flags in flags).

Files

One JSONL per model, one line per task-seed, 300 lines each: ours_qwen38-27b-ATX-AI2-8GB.jsonl, mellum21-iq3xs.jsonl, mellum21-mtp-mxfp4.jsonl, lfm3-1.7b-q8.jsonl, ornith-1.5-9b-q4km.jsonl, qwen35-9b-mtp-q4km.jsonl, twil-8b-q4km.jsonl, empero-qwen38-9b-q4km.jsonl.

Fields: model, file, file_sha256, task, seed, attempted_at_8gb_ctx, source (run1 = earlier run, mostly on 24 GB cards; run2 = rerun at the 8 GB context; ours = our run), pass_v1, pass_v2, graded_by, prompt_tokens, generated_tokens, finish_reason, looped_v3 (loop audit flag, informational), ctx (context per slot during generation), scored_ctx (the context the answer is scored at), card, runtime, flags, sampling.

Statistics

Paired gap = ours minus the other model on the same 100 problems, using the per-problem difference in 3-seed pass counts. The 95% CI comes from 10,000 bootstrap resamples over problems. p is an exact two-sided sign-flip test.

Downloads last month
36