Get up and running with Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.
gptoss: fix memory calc (#11700)
Michael Yangfcec04bf4249
diff --git a/fs/ggml/ggml.go b/fs/ggml/ggml.goindex fb993a288..45d58bc47 100644--- a/fs/ggml/ggml.go+++ b/fs/ggml/ggml.go@@ -676,7 +676,7 @@ func (f GGML) GraphSize(context, batch uint64, numParallel int, kvCacheType stri}}fullOffload = 4 * f.KV().HeadCountMax() / cmp.Or(f.KV().HeadCountKVMin(), 1) * kvTotal / 6- partialOffload = 2 * fullOffload+ partialOffload = fullOffload}return