mirrors/ollama-for-amd
1
0
Fork 0
mirror of https://github.com/likelovewant/ollama-for-amd.git synced 2025-12-21 22:33:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat(model): add qwen3vl (#12665)

Browse Source
This commit is contained in:
Michael Yang
2025-10-28 17:39:47 -07:00
committed by GitHub
parent 36d64fb531
commit 7d25b9e194
22 changed files with 1502 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
server/routes.go
View File

@@ -142,7 +142,10 @@ func (s *Server) scheduleRunner(ctx context.Context, name string, caps []model.C
// This model is much more capable with a larger context, so set that
// unless it would penalize performance too much
if !s.lowVRAM && slices.Contains([]string{"gptoss", "gpt-oss"}, model.Config.ModelFamily) {
if !s.lowVRAM && slices.Contains([]string{
"gptoss", "gpt-oss",
"qwen3vl", "qwen3vlmoe",
}, model.Config.ModelFamily) {
opts.NumCtx = max(opts.NumCtx, 8192)
}

6
server/sched.go
View File

@@ -390,11 +390,11 @@ func (s *Scheduler) load(req *LlmRequest, f *ggml.GGML, systemInfo ml.SystemInfo
numParallel = 1
}
// `mllama` is a snowflake and uses an encoder cache which cannot be used with num_parallel > 1
// `mllama`, `qwen3vl`, and `qwen3vlmoe` are snowflakes and uses an encoder cache which cannot be used with num_parallel > 1
// ref: https://github.com/ollama/ollama/issues/4165
if slices.Contains(req.model.Config.ModelFamilies, "mllama") && numParallel != 1 {
if slices.Contains([]string{"mllama", "qwen3vl", "qwen3vlmoe"}, req.model.Config.ModelFamily) && numParallel != 1 {
numParallel = 1
slog.Warn("mllama does not currently support parallel requests")
slog.Warn("model architecture does not currently support parallel requests", "architecture", req.model.Config.ModelFamily)
}
sessionDuration := envconfig.KeepAlive()