 Georgi Gerganov
|
7b50d589a8
kv-cells : fix tracking of seq_pos (#14339)
|
6 bulan lalu |
 Ed Addario
|
fa4a9f2a1c
quantize : handle user-defined pruning of whole layers (blocks) (#13037)
|
6 bulan lalu |
 Ruikai Peng
|
dd6e6d0b6a
vocab : prevent tokenizer overflow (#14301)
|
7 bulan lalu |
 Sigbjørn Skjæret
|
88fc854b4b
llama : improve sep token handling (#14272)
|
7 bulan lalu |
|
Georgi Gerganov
|
89fea80d29
server : fix incorrect usage of llama_get_embeddings() (#14225)
|
7 bulan lalu |
|
Georgi Gerganov
|
d3e64b9f49
llama : rework embeddings logic (#14208)
|
7 bulan lalu |
|
Georgi Gerganov
|
b9912ac570
batch : auto-gen positions + verify multi-sequence input (#14177)
|
7 bulan lalu |
|
Georgi Gerganov
|
745aa5319b
llama : deprecate llama_kv_self_ API (#14030)
|
7 bulan lalu |
|
Sigbjørn Skjæret
|
d17a809ef0
llama : support multiple classifier outputs and labels (#13940)
|
7 bulan lalu |
|
Georgi Gerganov
|
7f37b6cf1e
memory : migrate from llama_kv_cache to more generic llama_memory (#14006)
|
7 bulan lalu |
|
Georgi Gerganov
|
803f8baf4f
llama : deprecate explicit kv_self defrag/update calls (#13921)
|
7 bulan lalu |
|
Georgi Gerganov
|
3600cc2886
llama : use n_swa + n_ubatch cells for SWA cache (#13833)
|
7 bulan lalu |
|
Georgi Gerganov
|
12d0188c0d
kv-cache : refactor + add llama_memory_state_i (#13746)
|
7 bulan lalu |
|
Georgi Gerganov
|
22229314fc
llama : clarify deprecation message (#13794)
|
7 bulan lalu |
|
Georgi Gerganov
|
de2ef53a4b
kv-cache : rework kv_cell (#13706)
|
7 bulan lalu |
|
Georgi Gerganov
|
797f2ac062
kv-cache : simplify the interface (#13660)
|
8 bulan lalu |
|
Georgi Gerganov
|
a4090d1174
llama : remove llama_kv_cache_view API + remove deprecated (#13653)
|
8 bulan lalu |
|
Georgi Gerganov
|
e298d2fbd0
kv-cache : add SWA support (#13194)
|
8 bulan lalu |
 Diego Devesa
|
cf0a43bb64
llama-bench : add defrag-thold, check for invalid ranges (#13487)
|
8 bulan lalu |
 Johannes Gäßler
|
10d2af0eaa
llama/ggml: add LLM training support (#10544)
|
8 bulan lalu |
 David Huang
|
7f323a589f
Add `--no-op-offload` to improve `-ot` pp perf in MoE models like llama4 400B (#13386)
|
8 bulan lalu |
|
Sigbjørn Skjæret
|
d2a4ef05c6
vocab : add ByteDance-Seed/Seed-Coder (#13423)
|
8 bulan lalu |
|
Georgi Gerganov
|
6562e5a4d6
context : allow cache-less context for embeddings (#13108)
|
8 bulan lalu |
|
Georgi Gerganov
|
51fb96b1ff
context : remove logits_all flag (#13284)
|
8 bulan lalu |
|
Georgi Gerganov
|
d9d398f84f
sampling : when top-k <= 0 -> noop (#13173)
|
8 bulan lalu |
 Xuan-Son Nguyen
|
ecda2ec4b3
mtmd : Support Pixtral 12B (#13065)
|
8 bulan lalu |
|
Ed Addario
|
71e90e8813
quantize: Handle user-defined quantization levels for additional tensors (#12511)
|
9 bulan lalu |
|
Xuan-Son Nguyen
|
1466621e73
llama : Support llama 4 text-only (#12791)
|
9 bulan lalu |
|
Diego Devesa
|
e0e912f49b
llama : add option to override model tensor buffers (#11397)
|
9 bulan lalu |
|
Sigbjørn Skjæret
|
2c3f8b850a
llama : support BailingMoE (Ling) (#12634)
|
9 bulan lalu |
