 Georgi Gerganov
|
3600cc2886
llama : use n_swa + n_ubatch cells for SWA cache (#13833)
|
hai 7 meses |
|
Georgi Gerganov
|
3f55f781f1
llama : auto-batch preparation (#13845)
|
hai 7 meses |
|
Georgi Gerganov
|
12d0188c0d
kv-cache : refactor + add llama_memory_state_i (#13746)
|
hai 7 meses |
|
Georgi Gerganov
|
4f81b33e32
llama : validate seq id batch input (#13809)
|
hai 8 meses |
|
Georgi Gerganov
|
79c137f776
examples : allow extracting embeddings from decoder contexts (#13797)
|
hai 8 meses |
|
Georgi Gerganov
|
de2ef53a4b
kv-cache : rework kv_cell (#13706)
|
hai 8 meses |
|
Georgi Gerganov
|
797f2ac062
kv-cache : simplify the interface (#13660)
|
hai 8 meses |
|
Georgi Gerganov
|
a4090d1174
llama : remove llama_kv_cache_view API + remove deprecated (#13653)
|
hai 8 meses |
|
Georgi Gerganov
|
e298d2fbd0
kv-cache : add SWA support (#13194)
|
hai 8 meses |
 Sigbjørn Skjæret
|
f5170c1d7a
editorconfig : fix trailing whitespace from #13542 (#13546)
|
hai 8 meses |
 Gilad S.
|
017f10b5fa
fix: crash when calling `llama_state_get_size` on a context without a KV cache (#13542)
|
hai 8 meses |
 Johannes Gäßler
|
10d2af0eaa
llama/ggml: add LLM training support (#10544)
|
hai 8 meses |
|
Georgi Gerganov
|
064cc596ac
context : fix state io for memory-less contexts (#13470)
|
hai 8 meses |
 David Huang
|
7f323a589f
Add `--no-op-offload` to improve `-ot` pp perf in MoE models like llama4 400B (#13386)
|
hai 8 meses |
|
Georgi Gerganov
|
6562e5a4d6
context : allow cache-less context for embeddings (#13108)
|
hai 8 meses |
|
Georgi Gerganov
|
51fb96b1ff
context : remove logits_all flag (#13284)
|
hai 8 meses |
|
Georgi Gerganov
|
a75cb30dc9
context : fix reorder logic (#13267)
|
hai 8 meses |
|
Georgi Gerganov
|
c642bc014c
kv-cache : separate recurrent vs non-recurrent impl (#12799)
|
hai 8 meses |
 ddh0
|
16a457facd
fix typo: `n_ctx_pre_seq` -> `n_ctx_per_seq` (#13221)
|
hai 8 meses |
 pockers21
|
fb0471d175
context : do not clear output buffer on reserve (#13152)
|
hai 9 meses |
 Diego Devesa
|
295354ea68
llama : fix K-shift with quantized K and BLAS backend (#13113)
|
hai 9 meses |
|
Georgi Gerganov
|
2f74c354c0
graph : make FA compatible with MLA + add initial Metal kernels (#12953)
|
hai 9 meses |
 Juk Armstrong
|
daa422881a
llama : DeepSeek V2/V3 MLA implementation (#12801)
|
hai 9 meses |
|
Georgi Gerganov
|
3e1d29348b
kv-cache : simplify + fix warning for recurrent models (#12756)
|
hai 9 meses |
|
Diego Devesa
|
e0e912f49b
llama : add option to override model tensor buffers (#11397)
|
hai 9 meses |
|
Georgi Gerganov
|
a10b36c91a
llama : refactor kv cache guard (#12695)
|
hai 9 meses |
 Xuan-Son Nguyen
|
af6ae1efb2
llama : fix non-causal mask for gemma 3 (#12615)
|
hai 10 meses |
|
Georgi Gerganov
|
b4ae50810e
metal : improve FA + improve MoE (#12612)
|
hai 10 meses |
|
Georgi Gerganov
|
2d77d88e70
context : fix worst-case reserve outputs (#12545)
|
hai 10 meses |
 fairydreaming
|
568013d0cd
context : clear sets containing encoder output sequence ids before storing new values (#12470)
|
hai 10 meses |
