Daniel Hiltgen
|
17ce203a26
Merge pull request #4875 from dhiltgen/rocm_gfx900_workaround
|
10 mesi fa |
Daniel Hiltgen
|
6be309e1bd
Centralize GPU configuration vars
|
1 anno fa |
Daniel Hiltgen
|
da3bf23354
Workaround gfx900 SDMA bugs
|
11 mesi fa |
Daniel Hiltgen
|
6f351bf586
review comments and coverage
|
11 mesi fa |
Daniel Hiltgen
|
6fd04ca922
Improve multi-gpu handling at the limit
|
11 mesi fa |
Daniel Hiltgen
|
43ed358f9a
Refine GPU discovery to bootstrap once
|
11 mesi fa |
Daniel Hiltgen
|
b32ebb4f29
Use DRM driver for VRAM info for amd
|
11 mesi fa |
Daniel Hiltgen
|
8727a9c140
Record more GPU information
|
1 anno fa |
Daniel Hiltgen
|
e592e8fccb
Support Fedoras standard ROCm location
|
1 anno fa |
Daniel Hiltgen
|
0d6687f84c
AMD gfx patch rev is hex
|
1 anno fa |
Daniel Hiltgen
|
34b9db5afc
Request and model concurrency
|
1 anno fa |
Daniel Hiltgen
|
58d95cc9bd
Switch back to subprocessing for llama.cpp
|
1 anno fa |
Michael Yang
|
f31f2bedf4
Update troubleshooting link
|
1 anno fa |
Daniel Hiltgen
|
82b0c7c27e
Fix iGPU detection for linux
|
1 anno fa |
Daniel Hiltgen
|
bc13da2bfe
Avoid rocm runner and dependency clash
|
1 anno fa |
Daniel Hiltgen
|
00ec269321
Add ollama executable peer dir for rocm
|
1 anno fa |
Daniel Hiltgen
|
4a5c9b8035
Finish unwinding idempotent payload logic
|
1 anno fa |
Daniel Hiltgen
|
6c5ccb11f9
Revamp ROCm support
|
1 anno fa |