 Michael Yang
|
cef3cf353a
estimate log value
|
4 months ago |
|
Michael Yang
|
b7943d941d
merge ggml file decoding
|
5 months ago |
 frob
|
63269668c0
Prevent underflow when FreeMemory < overhead (#8014)
|
4 months ago |
 Sam
|
539be43640
llm: normalise kvct parameter handling (#7926)
|
5 months ago |
|
Sam
|
1bdab9fdb1
llm: introduce k/v context quantization (vRAM improvements) (#6279)
|
5 months ago |
|
Michael Yang
|
d07cf41a97
refactor kv estimation
|
6 months ago |
 Patrick Devine
|
c7cb0f0602
image processing for llama3.2 (#6963)
|
6 months ago |
 Daniel Hiltgen
|
05cd82ef94
Rename gpu package discover (#7143)
|
6 months ago |
|
Daniel Hiltgen
|
56318fb365
Improve logging on GPU too small (#6666)
|
7 months ago |
|
Daniel Hiltgen
|
b05c9e83d9
Introduce GPU Overhead env var (#5922)
|
7 months ago |
|
Michael Yang
|
8e0641a9bf
handle asymmetric embedding KVs
|
10 months ago |
 Daniel Hiltgen
|
359b15a597
Handle models with divergent layer sizes
|
10 months ago |
|
Daniel Hiltgen
|
7784ca33ce
Tighten up memory prediction logging
|
10 months ago |
|
Daniel Hiltgen
|
17df6520c8
Remove mmap related output calc logic
|
10 months ago |
|
Daniel Hiltgen
|
6f351bf586
review comments and coverage
|
11 months ago |
|
Daniel Hiltgen
|
6fd04ca922
Improve multi-gpu handling at the limit
|
11 months ago |
|
Michael Yang
|
6297f85606
gofmt, goimports
|
11 months ago |
|
Michael Yang
|
e40145a39d
lint
|
11 months ago |
 Patrick Devine
|
4cc3be3035
Move envconfig and consolidate env vars (#4608)
|
11 months ago |
|
Michael Yang
|
1d359e737e
typo
|
11 months ago |
|
Michael Yang
|
50b9056e09
count memory up to NumGPU
|
11 months ago |
 Jeffrey Morgan
|
bb6fd02298
Don't clamp ctx size in `PredictServerFit` (#4317)
|
11 months ago |
|
Daniel Hiltgen
|
bee2f4a3b0
Record GPU usage information
|
1 year ago |
|
Michael Yang
|
4736391bfb
llm: add minimum based on layer size
|
1 year ago |
|
Daniel Hiltgen
|
f56aa20014
Centralize server config handling
|
1 year ago |
|
Jeffrey Morgan
|
f0c454ab57
gpu: add 512MiB to darwin minimum, metal doesn't have partial offloading overhead (#4068)
|
1 year ago |
|
Michael Yang
|
f81f308118
fix gemma, command-r layer weights
|
1 year ago |
|
Michael Yang
|
7bb7cb8a60
only count output tensors
|
1 year ago |
|
Daniel Hiltgen
|
5445aaa94e
Add back memory escape valve
|
1 year ago |
|
Daniel Hiltgen
|
34b9db5afc
Request and model concurrency
|
1 year ago |
