From 69866c897645d6b31f8fbd31af2473fceb01da94 Mon Sep 17 00:00:00 2001 From: Danny Date: Wed, 3 Dec 2025 14:06:05 +0200 Subject: [PATCH 1/2] Block matmul and kv_cache in dynamic quantization --- .../maxabs_quant_dynamic_quantization.json | 13 +++++++++++++ 1 file changed, 13 insertions(+) diff --git a/calibration/quantization_config/maxabs_quant_dynamic_quantization.json b/calibration/quantization_config/maxabs_quant_dynamic_quantization.json index fa4ad093b..99a5ff3c2 100644 --- a/calibration/quantization_config/maxabs_quant_dynamic_quantization.json +++ b/calibration/quantization_config/maxabs_quant_dynamic_quantization.json @@ -4,5 +4,18 @@ "scale_format": "CONST", "scale_method": "act_maxabs_pcs_pow2_weight_maxabs_pts_pow2_hw", "dynamic_quantization": true, + "blocklist": { + "types": [ + "Matmul", + "KVCache", + "VLLMKVCache" + ], + "names": [ + "matmul_qk", + "matmul_av", + "k_cache", + "v_cache" + ] + }, "dump_stats_path": "" } From 817de1453a8caa6bbf2b049d89457b02953b95c8 Mon Sep 17 00:00:00 2001 From: Danny Semiat Date: Mon, 8 Dec 2025 11:06:59 +0200 Subject: [PATCH 2/2] Update maxabs_quant_dynamic_quantization.json Signed-off-by: Danny Semiat --- .../maxabs_quant_dynamic_quantization.json | 6 ------ 1 file changed, 6 deletions(-) diff --git a/calibration/quantization_config/maxabs_quant_dynamic_quantization.json b/calibration/quantization_config/maxabs_quant_dynamic_quantization.json index 99a5ff3c2..c82e0b96d 100644 --- a/calibration/quantization_config/maxabs_quant_dynamic_quantization.json +++ b/calibration/quantization_config/maxabs_quant_dynamic_quantization.json @@ -9,12 +9,6 @@ "Matmul", "KVCache", "VLLMKVCache" - ], - "names": [ - "matmul_qk", - "matmul_av", - "k_cache", - "v_cache" ] }, "dump_stats_path": ""