mirror of
https://github.com/likelovewant/ollama-for-amd.git
synced 2025-12-21 14:26:30 +00:00
fix: qwen2.5 vl rope (#13486)
* qwen25vl: bump max pixels * qwen25vl: mrope fix qwen2.5vl window * qwen25vl: vision rope
This commit is contained in:
@@ -1534,7 +1534,8 @@ func (t *Tensor) RoPE(ctx ml.Context, positions ml.Tensor, ropeDim int, ropeBase
|
||||
unsafe.SliceData(mropeSections),
|
||||
C.int(opts.Type),
|
||||
cmp.Or(C.int(opts.YaRN.OriginalContextLength), 128<<10),
|
||||
C.float(ropeBase), C.float(ropeScale),
|
||||
C.float(ropeBase),
|
||||
C.float(ropeScale),
|
||||
C.float(opts.YaRN.ExtrapolationFactor),
|
||||
cmp.Or(C.float(opts.YaRN.AttentionFactor), 1),
|
||||
cmp.Or(C.float(opts.YaRN.BetaFast), 32),
|
||||
@@ -1546,9 +1547,11 @@ func (t *Tensor) RoPE(ctx ml.Context, positions ml.Tensor, ropeDim int, ropeBase
|
||||
dequant,
|
||||
positions.(*Tensor).t,
|
||||
opts.Factors.(*Tensor).t,
|
||||
C.int(ropeDim), C.int(opts.Type),
|
||||
C.int(ropeDim),
|
||||
C.int(opts.Type),
|
||||
cmp.Or(C.int(opts.YaRN.OriginalContextLength), 128<<10),
|
||||
C.float(ropeBase), C.float(ropeScale),
|
||||
C.float(ropeBase),
|
||||
C.float(ropeScale),
|
||||
C.float(opts.YaRN.ExtrapolationFactor),
|
||||
cmp.Or(C.float(opts.YaRN.AttentionFactor), 1),
|
||||
cmp.Or(C.float(opts.YaRN.BetaFast), 32),
|
||||
|
||||
@@ -77,6 +77,13 @@ func WithMRoPE(sections []int) func(*Options) {
|
||||
}
|
||||
}
|
||||
|
||||
func WithVision(sections []int) func(*Options) {
|
||||
return func(opts *Options) {
|
||||
opts.Type |= 1<<3 | 1<<4
|
||||
opts.MRoPE.Sections = sections
|
||||
}
|
||||
}
|
||||
|
||||
func WithInterleaveMRoPE(sections []int) func(*Options) {
|
||||
return func(opts *Options) {
|
||||
opts.Type |= 1<<3 | 1<<5
|
||||
|
||||
Reference in New Issue
Block a user