highperfocused/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2025-03-18 05:41:43 +01:00
Code Issues Packages Projects Releases Wiki Activity

compat with upstream gguf

Browse Source
This commit is contained in:
Michael Yang 2025-03-10 11:08:37 -07:00
parent c5cbe4fc2a
commit 6b32a2d549
3 changed files with 14 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
convert/convert_gemma3.go
View File

@ -76,19 +76,19 @@ func (p *gemma3Model) KV(t *Tokenizer) ggml.KV {
switch p.Architecture {
case "Gemma3ForCausalLM":
kv["gemma3.context_length"] = p.MaxPositionEmbeddings
kv["gemma3.text.attention.layer_norm_rms_epsilon"] = p.RMSNormEPS
kv["gemma3.attention.layer_norm_rms_epsilon"] = p.RMSNormEPS
kv["gemma3.attention.key_length"] = p.HeadDim
kv["gemma3.attention.value_length"] = p.HeadDim
kv["gemma3.text.attention.sliding_window"] = p.SlidingWindow
kv["gemma3.text.final_logit_softcapping"] = p.FinalLogitSoftcap
kv["gemma3.text.rope.local.freq_base"] = p.RopeLocalTheta
kv["gemma3.text.rope.global.freq_base"] = p.RopeGlobalTheta
kv["gemma3.attention.sliding_window"] = p.SlidingWindow
kv["gemma3.final_logit_softcapping"] = p.FinalLogitSoftcap
kv["gemma3.rope.local.freq_base"] = p.RopeLocalTheta
kv["gemma3.rope.global.freq_base"] = p.RopeGlobalTheta
kv["gemma3.embedding_length"] = p.HiddenSize
kv["gemma3.text.feed_forward_length"] = p.IntermediateSize
kv["gemma3.feed_forward_length"] = p.IntermediateSize
default:
kv["gemma3.embedding_length"] = p.TextModel.HiddenSize
kv["gemma3.text.feed_forward_length"] = p.TextModel.IntermediateSize
kv["gemma3.text.attention.sliding_window"] = p.TextModel.SlidingWindow
kv["gemma3.feed_forward_length"] = p.TextModel.IntermediateSize
kv["gemma3.attention.sliding_window"] = p.TextModel.SlidingWindow
kv["gemma3.vision.block_count"] = p.VisionModel.NumHiddenLayers
kv["gemma3.vision.embedding_length"] = p.VisionModel.HiddenSize
kv["gemma3.vision.feed_forward_length"] = p.VisionModel.IntermediateSize

2
model/models/gemma3/model.go
View File

@ -62,7 +62,7 @@ func New(c ml.Config) (model.Model, error) {
TextModel: newTextModel(c),
}
slidingWindowLen := int32(c.Uint("text.attention.sliding_window"))
slidingWindowLen := int32(c.Uint("attention.sliding_window"))
m.Cache = kvcache.NewWrapperCache(kvcache.NewSWACache(slidingWindowLen, m.Shift), kvcache.NewCausalCache(m.Shift))
return &m, nil

10
model/models/gemma3/model_text.go
View File

@ -62,11 +62,11 @@ func newTextModel(c ml.Config) *TextModel {
numKVHeads: int(c.Uint("attention.head_count_kv")),
attnKeyLen: int(c.Uint("attention.key_length", 256)),
attnValLen: int(c.Uint("attention.value_length", 256)),
eps: c.Float("text.attention.layer_norm_rms_epsilon", 1e-06),
ropeLocalBase: c.Float("text.rope.local.freq_base", 10000.0),
ropeGlobalBase: c.Float("text.rope.global.freq_base", 1000000.0),
ropeScale: c.Float("text.rope.freq_scale", 1.0),
finalLogitSoftcap: c.Float("text.final_logit_softcapping", 30.0),
eps: c.Float("attention.layer_norm_rms_epsilon", 1e-06),
ropeLocalBase: c.Float("rope.local.freq_base", 10000.0),
ropeGlobalBase: c.Float("rope.global.freq_base", 1000000.0),
ropeScale: c.Float("rope.freq_scale", 1.0),
finalLogitSoftcap: c.Float("final_logit_softcapping", 30.0),
},
}