 Georgi Gerganov
|
39173bcacb
context : reserve new scheduler when graph topology changes (#18547)
|
2 settimane fa |
|
Georgi Gerganov
|
cd5e3b5754
server : support unified cache across slots (#16736)
|
3 mesi fa |
|
Georgi Gerganov
|
e58174cecb
llama : bump max seq limit from 64 to 256 (#15916)
|
4 mesi fa |
|
Georgi Gerganov
|
9ebebef62f
llama : remove KV cache defragmentation logic (#15473)
|
5 mesi fa |
|
Georgi Gerganov
|
225e7a1438
llama : add high-throughput mode (#14363)
|
6 mesi fa |
|
Georgi Gerganov
|
c311ac664d
cparams : rename LLAMA_MAX_PARALLEL_SEQUENCES to LLAMA_MAX_SEQ (#14188)
|
7 mesi fa |
|
Georgi Gerganov
|
b9912ac570
batch : auto-gen positions + verify multi-sequence input (#14177)
|
7 mesi fa |
|
Georgi Gerganov
|
de2ef53a4b
kv-cache : rework kv_cell (#13706)
|
8 mesi fa |
 David Huang
|
7f323a589f
Add `--no-op-offload` to improve `-ot` pp perf in MoE models like llama4 400B (#13386)
|
8 mesi fa |
 fairydreaming
|
8fcb563613
Load all MoE experts during warmup (#11571)
|
10 mesi fa |
|
Georgi Gerganov
|
f66f582927
llama : refactor `src/llama.cpp` (#10902)
|
1 anno fa |
