添加 U8+u16混合量化的Image Encoder

.gitattributes

@@ -55,3 +55,5 @@ Qwen3-VL-4B-Instruct-AX650-c128_p1152-int4/qwen3_vl_text_p128_l21_together.axmod
 Qwen3-VL-4B-Instruct-AX650-c128_p1152-int4/qwen3_vl_text_p128_l28_together.axmodel filter=lfs diff=lfs merge=lfs -text
 main_ax650 filter=lfs diff=lfs merge=lfs -text
 main_axcl_aarch64 filter=lfs diff=lfs merge=lfs -text
+main_axcl_x86 filter=lfs diff=lfs merge=lfs -text
+Qwen3-VL-4B-Instruct-AX650-c128_p1152-int4/Qwen3-VL-4B-Instruct_vision_u8.axmodel filter=lfs diff=lfs merge=lfs -text

Qwen3-VL-4B-Instruct-AX650-c128_p1152-int4/Qwen3-VL-4B-Instruct_vision_u8.axmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0318af7d86d0df92ceee2344eb1239fd1c74b3fedf84a7cd2cdcbf91341d9121
+size 441899962

README.md

@@ -52,6 +52,16 @@ For those who are interested in model conversion, you can try to export axmodel
 |--|--|--|--|--|--|--|--|
 |AX650| 384*384 | 8  | 773 ms | 1887 ms | 7.1 tokens/sec| 5.6GiB | 5.6GiB |
 
+**Image Process (Image Encoder U8+U16 Quantization)**
+|Chips| input size | image num | image encoder | ttft(320 tokens) | w4a16 | CMM | Flash |
+|--|--|--|--|--|--|--|--|
+|AX650| 384*384 | 1 | 171 ms | 678 ms | 7.0 tokens/sec| 5.6GiB | 5.6GiB |
+
+**Video Process (Image Encoder U8+U16 Quantization)**
+|Chips| input size | image num | image encoder |ttft(600 tokens) | w4a16 | CMM | Flash |
+|--|--|--|--|--|--|--|--|
+|AX650| 384*384 | 8  | 498 ms | 1887 ms | 7.1 tokens/sec| 5.6GiB | 5.6GiB |
+
 The DDR capacity refers to the CMM memory that needs to be consumed. Ensure that the CMM memory allocation on the development board is greater than this value.
 
 ## How to use

run_image_ax650_imgu8.sh

@@ -0,0 +1,23 @@
+AXMODEL_DIR=./Qwen3-VL-4B-Instruct-AX650-c128_p1152-int4/
+
+./main_ax650 \
+--template_filename_axmodel "${AXMODEL_DIR}/qwen3_vl_text_p128_l%d_together.axmodel" \
+--axmodel_num 36 \
+--filename_image_encoder_axmodedl "${AXMODEL_DIR}/Qwen3-VL-4B-Instruct_vision_u8.axmodel" \
+--bos 0 --eos 0 \
+--dynamic_load_axmodel_layer 0 \
+--use_mmap_load_embed 1 \
+--filename_tokenizer_model "http://127.0.0.1:8080" \
+--filename_post_axmodel "${AXMODEL_DIR}/qwen3_vl_text_post.axmodel" \
+--use_topk 0 \
+--filename_tokens_embed "${AXMODEL_DIR}/model.embed_tokens.weight.bfloat16.bin" \
+--tokens_embed_num 151936 \
+--tokens_embed_size 2560 \
+--patch_size 16 \
+--live_print 1 \
+--continue 1 \
+--video 0 \
+--img_width 384 \
+--img_height 384 \
+--vision_start_token_id 151652 \
+--post_config_path post_config.json 

run_video_ax650_imgu8.sh

@@ -0,0 +1,23 @@
+AXMODEL_DIR=./Qwen3-VL-4B-Instruct-AX650-c128_p1152-int4/
+
+./main_ax650  \
+--template_filename_axmodel "${AXMODEL_DIR}/qwen3_vl_text_p128_l%d_together.axmodel" \
+--axmodel_num 36 \
+--filename_image_encoder_axmodedl "${AXMODEL_DIR}/Qwen3-VL-4B-Instruct_vision_u8.axmodel" \
+--bos 0 --eos 0 \
+--dynamic_load_axmodel_layer 0 \
+--use_mmap_load_embed 1 \
+--filename_tokenizer_model "http://127.0.0.1:8080" \
+--filename_post_axmodel "${AXMODEL_DIR}/qwen3_vl_text_post.axmodel" \
+--use_topk 0 \
+--filename_tokens_embed "${AXMODEL_DIR}/model.embed_tokens.weight.bfloat16.bin" \
+--tokens_embed_num 151936 \
+--tokens_embed_size 2560 \
+--patch_size 16 \
+--live_print 1 \
+--continue 1 \
+--video 1 \
+--img_width 384 \
+--img_height 384 \
+--vision_start_token_id 151652 \
+--post_config_path post_config.json