mirrors/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2025-05-11 18:36:41 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

Workaround gfx900 SDMA bugs

Browse source 
Implement support for GPU env var workarounds, and leverage
this for the Vega RX 56 which needs
HSA_ENABLE_SDMA=0 set to work properly
This commit is contained in:
Daniel Hiltgen 2024-05-31 16:15:21 -07:00
parent 45cacbaf05
commit da3bf23354
3 changed files with 18 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

10
llm/server.go
View file

@ -320,6 +320,10 @@ func NewLlamaServer(gpus gpu.GpuInfoList, model string, ggml *GGML, adapters, pr
s.cmd.Stdout = os.Stdout
s.cmd.Stderr = s.status
envWorkarounds := [][2]string{}
for _, gpu := range gpus {
envWorkarounds = append(envWorkarounds, gpu.EnvWorkarounds...)
}
visibleDevicesEnv, visibleDevicesEnvVal := gpus.GetVisibleDevicesEnv()
pathEnvVal := strings.Join(libraryPaths, string(filepath.ListSeparator))
@ -334,6 +338,12 @@ func NewLlamaServer(gpus gpu.GpuInfoList, model string, ggml *GGML, adapters, pr
} else if devicesNeeded && strings.EqualFold(cmp[0], visibleDevicesEnv) {
s.cmd.Env[i] = visibleDevicesEnv + "=" + visibleDevicesEnvVal
devicesNeeded = false
} else if len(envWorkarounds) != 0 {
for _, kv := range envWorkarounds {
if strings.EqualFold(cmp[0], kv[0]) {
s.cmd.Env[i] = kv[0] + "=" + kv[1]
}
}
}
}
if pathNeeded {