Add FP8 KV cache quantization

#1
Files changed (2) hide show
  1. config.json +5 -0
  2. hf_quant_config.json +1 -1
config.json CHANGED
@@ -394,6 +394,11 @@
394
  "model.layers.9.self_attn*"
395
  ],
396
  "quant_algo": "NVFP4",
 
 
 
 
 
397
  "producer": {
398
  "name": "modelopt",
399
  "version": "0.45.0.dev44+gc273ddb8a.d20260509"
 
394
  "model.layers.9.self_attn*"
395
  ],
396
  "quant_algo": "NVFP4",
397
+ "kv_cache_scheme": {
398
+ "dynamic": false,
399
+ "num_bits": 8,
400
+ "type": "float"
401
+ },
402
  "producer": {
403
  "name": "modelopt",
404
  "version": "0.45.0.dev44+gc273ddb8a.d20260509"
hf_quant_config.json CHANGED
@@ -5,7 +5,7 @@
5
  },
6
  "quantization": {
7
  "quant_algo": "NVFP4",
8
- "kv_cache_quant_algo": null,
9
  "group_size": 16,
10
  "exclude_modules": [
11
  "lm_head",
 
5
  },
6
  "quantization": {
7
  "quant_algo": "NVFP4",
8
+ "kv_cache_quant_algo": "FP8",
9
  "group_size": 16,
10
  "exclude_modules": [
11
  "lm_head",