tckm.yan/aha
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

update format

Browse Source
This commit is contained in:
jhqxxx
2025-10-28 14:31:02 +08:00
parent cff51418c7
commit ff5f8f87e1
2 changed files with 7 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/models/qwen2_5vl/processor.rs
+3 -2
View File
@@ -111,8 +111,9 @@ impl Qwen2_5VLProcessor {
// img_tensor: (t, c, h, w) // img_tensor: (t, c, h, w)
let t = img_tensor.dim(0)?; let t = img_tensor.dim(0)?;
let img_tensor = if t % self.vision_setting.temporal_patch_size != 0 { let img_tensor = if t % self.vision_setting.temporal_patch_size != 0 {
let repeat_num = self.vision_setting.temporal_patch_size - t % self.vision_setting.temporal_patch_size; let repeat_num = self.vision_setting.temporal_patch_size
let repeats = img_tensor.i(t-1)?.repeat((repeat_num, 1, 1, 1))?; - t % self.vision_setting.temporal_patch_size;
let repeats = img_tensor.i(t - 1)?.repeat((repeat_num, 1, 1, 1))?;
Tensor::cat(&[img_tensor, &repeats], 0)? Tensor::cat(&[img_tensor, &repeats], 0)?
} else { } else {
img_tensor.clone() img_tensor.clone()
src/models/qwen3vl/processor.rs
+4 -3
View File
@@ -162,8 +162,9 @@ impl Qwen3VLProcessor {
// img_tensor: (t, c, h, w) // img_tensor: (t, c, h, w)
let t = img_tensor.dim(0)?; let t = img_tensor.dim(0)?;
let img_tensor = if t % self.img_process_cfg.temporal_patch_size != 0 { let img_tensor = if t % self.img_process_cfg.temporal_patch_size != 0 {
let repeat_num = self.img_process_cfg.temporal_patch_size - t % self.img_process_cfg.temporal_patch_size; let repeat_num = self.img_process_cfg.temporal_patch_size
let repeats = img_tensor.i(t-1)?.repeat((repeat_num, 1, 1, 1))?; - t % self.img_process_cfg.temporal_patch_size;
let repeats = img_tensor.i(t - 1)?.repeat((repeat_num, 1, 1, 1))?;
Tensor::cat(&[img_tensor, &repeats], 0)? Tensor::cat(&[img_tensor, &repeats], 0)?
} else { } else {
img_tensor.clone() img_tensor.clone()
@@ -251,7 +252,7 @@ impl Qwen3VLProcessor {
.broadcast_sub(img_mean)? .broadcast_sub(img_mean)?
.broadcast_div(img_std)? .broadcast_div(img_std)?
.contiguous()?; .contiguous()?;
// t // t
let (video_tensor, video_grid_thw) = self.process_vision_tensor(&video_tensor)?; let (video_tensor, video_grid_thw) = self.process_vision_tensor(&video_tensor)?;
pixel_values_vec.push(video_tensor); pixel_values_vec.push(video_tensor);
vision_grid_thws_vec.push(video_grid_thw); vision_grid_thws_vec.push(video_grid_thw);