Add hlwq_config.json (rename from polar_config.json)
Browse files- hlwq_config.json +30 -0
hlwq_config.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"method": "HLWQ",
|
| 3 |
+
"weight_bits": 5,
|
| 4 |
+
"kv_cache_bits": 3,
|
| 5 |
+
"block_size": 128,
|
| 6 |
+
"head_dim": 128,
|
| 7 |
+
"num_layers": 64,
|
| 8 |
+
"quantized_layers": 497,
|
| 9 |
+
"skipped_layers": 0,
|
| 10 |
+
"int4_group_size": 128,
|
| 11 |
+
"vram_gb": 19.1,
|
| 12 |
+
"base_model": "huihui-ai/Huihui-Qwopus3.5-27B-v3-abliterated",
|
| 13 |
+
"benchmark": {
|
| 14 |
+
"FP16": {
|
| 15 |
+
"tok_s": 21.804779849809538,
|
| 16 |
+
"peak_gb": 70.363202048,
|
| 17 |
+
"time": 4.586150407791138
|
| 18 |
+
},
|
| 19 |
+
"Q3": {
|
| 20 |
+
"tok_s": 22.092763850247692,
|
| 21 |
+
"peak_gb": 19.130801664,
|
| 22 |
+
"time": 4.526368935902913
|
| 23 |
+
},
|
| 24 |
+
"Q2": {
|
| 25 |
+
"tok_s": 22.038845106697618,
|
| 26 |
+
"peak_gb": 19.130801664,
|
| 27 |
+
"time": 4.537442843119304
|
| 28 |
+
}
|
| 29 |
+
}
|
| 30 |
+
}
|