Upload etat/carte-d0wveneybza9f4.log with huggingface_hub
Browse files
etat/carte-d0wveneybza9f4.log
CHANGED
|
@@ -6,3 +6,12 @@ vllm 0.27.1 cap (12, 0) SM 170
|
|
| 6 |
[ 60 s] Using FLASHINFER attention backend out of potential backends: ['FLASHINFER', 'TRITON_ATTN'].
|
| 7 |
[240 s] Using uncalibrated q_scale 1.0 and/or prob_scale 1.0 with fp8 attention. This may cause accuracy issues. Pleas
|
| 8 |
[300 s] Using flashinfer Mamba SSU backend.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
[ 60 s] Using FLASHINFER attention backend out of potential backends: ['FLASHINFER', 'TRITON_ATTN'].
|
| 7 |
[240 s] Using uncalibrated q_scale 1.0 and/or prob_scale 1.0 with fp8 attention. This may cause accuracy issues. Pleas
|
| 8 |
[300 s] Using flashinfer Mamba SSU backend.
|
| 9 |
+
Using FLASHINFER attention backend out of potential backends: ['FLASHINFER', 'TRITON_ATTN'].
|
| 10 |
+
Using 'MARLIN' NvFp4 MoE backend out of potential backends: ['FLASHINFER_TRTLLM', 'FLASHINFER_CUTEDSL', 'FLASHINFER_CUTEDSL_BATCHED', 'FLASH
|
| 11 |
+
Using MarlinNvFp4LinearKernel for NVFP4 GEMM
|
| 12 |
+
GPU KV cache size: 319,488 tokens
|
| 13 |
+
Maximum concurrency for 65,536 tokens per request: 4.88x
|
| 14 |
+
NE DEMARRE PAS
|
| 15 |
+
> ValueError: max_num_seqs (256) exceeds available Mamba cache blocks (117). Each decode sequence requires one Mamba cache block, so CUDA graph capture cannot proc
|
| 16 |
+
> RuntimeError: Engine core initialization failed. See root cause above. Failed core proc(s): {}
|
| 17 |
+
[CARTE] 19:01:44 TERMINE
|