Page values for "DeepSeek V4.1 Flash on 4x 8 GB (GNmac)"
From CMP Wiki
More actions
"Recipe" values
1 row is stored for this page| Field | Field type | Allowed values | Value |
|---|---|---|---|
| page | Page | DeepSeek V4.1 Flash on 4x 8 GB (GNmac) | |
| model | String | DeepSeek-V4.1-Flash-1017 | |
| quant | String | Hybrid: routed experts EXL3 (mostly 3 bpw, 2 bpw on layers 18-22) from LibertAI's 2.9 bpw pack; attention, shared experts, indexer and DSpark drafter official FP8; KV fp8_ds_mla | |
| engine | String | vLLM with a pinned sm_80 patch stack (built locally from source) | |
| engine_version | String | recipe 1.2.0 | |
| cards | String | 8 GB · 10 GB · both | 8 GB |
| card_count | Integer | 4 | |
| topology | String | TP4 (default) or PP4 for x1/x4 links | |
| context | String | 262,144 (up to 1,048,576) | |
| decode_tps | Float | 258.0 | |
| prefill_tps | Float | 2,416.0 | |
| vram_per_card | String | 64 GiB, GPU_MEMORY_UTILIZATION 0.95 | |
| power_cap | String | 180 W (nvidia-smi -pl 180) | |
| driver | String | CUDA 13 capable; measured on 610.43.02 | |
| unlock_build | String | cmpunlocker (64 GiB, 74 SMs) with a P2P-capable driver for TP4 | |
| flags | Text | LAYOUT=tp4, MAX_MODEL_LEN 262144, MAX_NUM_SEQS 8, GPU_MEMORY_UTILIZATION 0.95, P2P 1, DSV41_GATE_TAU 0.10, DSV41_DYNK accept:3,5 | |
| status | String | Proven · Reported · Theory · Contested · Warning | Proven |
| tested_by | String | greencandle9753 | |
| tested_on | Date | 2026-10-26 | |
| source | String | https://github.com/GNmac-lab/deepseek-v41-flash-1017-cmp170hx | |
| files | Wikitext |