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

typo

Browse source
This commit is contained in:
Michael Yang 2024-05-13 14:14:10 -07:00
parent 50b9056e09
commit 1d359e737e
1 changed files with 13 additions and 9 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

22
llm/memory.go
View file

@ -54,8 +54,10 @@ func EstimateGPULayers(gpus []gpu.GpuInfo, ggml *GGML, projectors []string, opts
} }
layers := ggml.Tensors().Layers() layers := ggml.Tensors().Layers()
// add one layer worth of memorr as a buffer // add one layer worth of memory as a buffer
memoryMinimum += layers["blk.0"].size() if blk0, ok := layers["blk.0"]; ok {
memoryMinimum += blk0.size()
}
// fp16 k,v = (1 (k) + 1 (v)) * sizeof(float16) * n_ctx * n_layer * n_embd / n_head * n_head_kv // fp16 k,v = (1 (k) + 1 (v)) * sizeof(float16) * n_ctx * n_layer * n_embd / n_head * n_head_kv
var kv uint64 = 2 * 2 * uint64(opts.NumCtx) * ggml.KV().BlockCount() * ggml.KV().EmbeddingLength() / ggml.KV().HeadCount() * ggml.KV().HeadCountKV() var kv uint64 = 2 * 2 * uint64(opts.NumCtx) * ggml.KV().BlockCount() * ggml.KV().EmbeddingLength() / ggml.KV().HeadCount() * ggml.KV().HeadCountKV()
@ -102,15 +104,17 @@ func EstimateGPULayers(gpus []gpu.GpuInfo, ggml *GGML, projectors []string, opts
var layerCount int var layerCount int
for i := 0; i < int(ggml.KV().BlockCount()); i++ { for i := 0; i < int(ggml.KV().BlockCount()); i++ {
memoryLayer := layers[fmt.Sprintf("blk.%d", i)].size() if blk, ok := layers[fmt.Sprintf("blk.%d", i)]; ok {
memoryLayer := blk.size()
// KV is proportional to the number of layers // KV is proportional to the number of layers
memoryLayer += kv / ggml.KV().BlockCount() memoryLayer += kv / ggml.KV().BlockCount()
memoryRequiredTotal += memoryLayer memoryRequiredTotal += memoryLayer
if (opts.NumGPU >= 0 && layerCount+1 <= opts.NumGPU) || (opts.NumGPU < 0 && memoryAvailable > memoryRequiredPartial+memoryLayer) { if (opts.NumGPU >= 0 && layerCount+1 <= opts.NumGPU) || (opts.NumGPU < 0 && memoryAvailable > memoryRequiredPartial+memoryLayer) {
memoryRequiredPartial += memoryLayer memoryRequiredPartial += memoryLayer
layerCount++ layerCount++
}
} }
} }