Fix model config
#4
by BitPhinix - opened
- config.json +9 -0
config.json
CHANGED
|
@@ -36,6 +36,15 @@
|
|
| 36 |
"q_lora_rank": 1536,
|
| 37 |
"qk_nope_head_dim": 128,
|
| 38 |
"qk_rope_head_dim": 64,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 39 |
"rms_norm_eps": 1e-06,
|
| 40 |
"rope_scaling": {
|
| 41 |
"beta_fast": 32,
|
|
|
|
| 36 |
"q_lora_rank": 1536,
|
| 37 |
"qk_nope_head_dim": 128,
|
| 38 |
"qk_rope_head_dim": 64,
|
| 39 |
+
"quantization_config": {
|
| 40 |
+
"activation_scheme": "dynamic",
|
| 41 |
+
"fmt": "e4m3",
|
| 42 |
+
"quant_method": "fp8",
|
| 43 |
+
"weight_block_size": [
|
| 44 |
+
128,
|
| 45 |
+
128
|
| 46 |
+
]
|
| 47 |
+
},
|
| 48 |
"rms_norm_eps": 1e-06,
|
| 49 |
"rope_scaling": {
|
| 50 |
"beta_fast": 32,
|