baalajimaestro/ollama
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

refactor kv estimation

Browse source
This commit is contained in:
Michael Yang 2024-10-31 13:46:30 -07:00
parent 8c238e70ab
commit d07cf41a97
2 changed files with 20 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

17
llm/ggml.go
View file

@ -360,7 +360,7 @@ func DecodeGGML(rs io.ReadSeeker, maxArraySize int) (*GGML, int64, error) {
}, offset, nil }, offset, nil
} }
func (llm GGML) GraphSize(context, batch uint64) (partialOffload, fullOffload uint64) { func (llm GGML) GraphSize(context, batch uint64) (kv, partialOffload, fullOffload uint64) {
embedding := llm.KV().EmbeddingLength() embedding := llm.KV().EmbeddingLength()
heads := llm.KV().HeadCount() heads := llm.KV().HeadCount()
headsKV := llm.KV().HeadCountKV() headsKV := llm.KV().HeadCountKV()
@ -368,9 +368,12 @@ func (llm GGML) GraphSize(context, batch uint64) (partialOffload, fullOffload ui
embeddingHeads := llm.KV().EmbeddingHeadCount() embeddingHeads := llm.KV().EmbeddingHeadCount()
embeddingHeadsK := llm.KV().EmbeddingHeadCountK() embeddingHeadsK := llm.KV().EmbeddingHeadCountK()
embeddingHeadsV := llm.KV().EmbeddingHeadCountV()
layers := llm.Tensors().Layers() layers := llm.Tensors().Layers()
kv = 2 * context * llm.KV().BlockCount() * (embeddingHeadsK + embeddingHeadsV) * headsKV
switch llm.KV().Architecture() { switch llm.KV().Architecture() {
case "llama": case "llama":
fullOffload = max( fullOffload = max(
@ -403,6 +406,18 @@ func (llm GGML) GraphSize(context, batch uint64) (partialOffload, fullOffload ui
case "mllama": case "mllama":
var visionTokens, tiles uint64 = 1601, 4 var visionTokens, tiles uint64 = 1601, 4
if crossAttentionLayers, ok := llm.KV()["mllama.attention.cross_attention_layers"].(*array); ok {
kv = headsKV *
(embeddingHeadsK + embeddingHeadsV) * // one for K, one for V
(2* // sizeof(float16)
(llm.KV().BlockCount()-uint64(crossAttentionLayers.size))* // num non-cross attention layers
context +
4* // sizeof(float32)
uint64(crossAttentionLayers.size)* // num cross attention layers
visionTokens*
tiles)
}
fullOffload = max( fullOffload = max(
4*batch*(2+3*embedding+embeddingHeadsK*heads+context*(1+heads)), 4*batch*(2+3*embedding+embeddingHeadsK*heads+context*(1+heads)),
// vocab graph // vocab graph

11
llm/memory.go
View file

@ -123,13 +123,7 @@ func EstimateGPULayers(gpus []discover.GpuInfo, ggml *GGML, projectors []string,
slog.Warn("model missing blk.0 layer size") slog.Warn("model missing blk.0 layer size")
} }
// fp16 k,v = sizeof(float16) * n_ctx * n_layer * (n_embd_head_k + n_embd_head_v) * n_head_kv kv, graphPartialOffload, graphFullOffload := ggml.GraphSize(uint64(opts.NumCtx), uint64(min(opts.NumCtx, opts.NumBatch)))
var kv uint64 = 2 * uint64(opts.NumCtx) * ggml.KV().BlockCount() * (ggml.KV().EmbeddingHeadCountK() + ggml.KV().EmbeddingHeadCountV()) * ggml.KV().HeadCountKV()
// KV is proportional to the number of layers
layerSize += kv / ggml.KV().BlockCount()
graphPartialOffload, graphFullOffload = ggml.GraphSize(uint64(opts.NumCtx), uint64(min(opts.NumCtx, opts.NumBatch)))
if graphPartialOffload == 0 { if graphPartialOffload == 0 {
graphPartialOffload = ggml.KV().GQA() * kv / 6 graphPartialOffload = ggml.KV().GQA() * kv / 6
} }
@ -137,6 +131,9 @@ func EstimateGPULayers(gpus []discover.GpuInfo, ggml *GGML, projectors []string,
graphFullOffload = graphPartialOffload graphFullOffload = graphPartialOffload
} }
// KV is proportional to the number of layers
layerSize += kv / ggml.KV().BlockCount()
// on metal there's no partial offload overhead // on metal there's no partial offload overhead
if gpus[0].Library == "metal" { if gpus[0].Library == "metal" {
graphPartialOffload = graphFullOffload graphPartialOffload = graphFullOffload