highperfocused/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2025-07-13 01:57:20 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #1552 from jmorganca/mxyng/lint-test

Browse Source 
add lint and test on pull_request
This commit is contained in:
Michael Yang
2024-01-11 09:37:45 -08:00
committed by GitHub
parent 3bc8b9832b f921e2696e
commit f4f939de28
17 changed files with 141 additions and 82 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
llm/llm.go
View File

@ -47,7 +47,7 @@ func New(workDir, model string, adapters, projectors []string, opts api.Options)
kv := 2 * 2 * int64(opts.NumCtx) * int64(ggml.NumLayers()) * int64(ggml.NumEmbed()) * int64(ggml.NumHeadKv()) / int64(ggml.NumHead())
// this amount is the overhead + tensors in memory
// TODO: get this from the llama.cpp's graph calcluations instead of
// TODO: get this from the llama.cpp's graph calculations instead of
// estimating it's 1/6 * kv_cache_size * num_gqa
graph := int64(ggml.NumGQA()) * kv / 6