3 a l i 2bf318fd2f model : add JAIS-2 architecture support (#19488) ... * model: add JAIS-2 architecture support Add support for the JAIS-2 family of Arabic-English bilingual models from Inception AI (https://huggingface.co/inceptionai/Jais-2-8B-Chat). Architecture characteristics: - LayerNorm (not RMSNorm) with biases - ReLU² (ReLU squared) activation function - Separate Q/K/V projections with biases - Simple MLP without gate projection (up -> act -> down) - RoPE positional embeddings - GPT-2 BPE tokenizer Supported model sizes: - Jais-2-8B (32 layers, 26 heads, 3328 hidden) - Jais-2-70B (68 layers, 56 heads, 7168 hidden) Tested with quantizations: BF16, Q8_0, Q6_K, Q5_K_M, Q5_0, Q4_K_M, Q4_0, Q3_K_M, Q2_K Note: JAIS-2 requires F32 precision accumulators for numerical stability and uses standard attention (not flash attention) on CUDA backends. * fix: run convert_hf_to_gguf_update.py for jais-2 tokenizer hash * fix: use NEOX RoPE type for JAIS2 * fix: remove Q/K permutation (NEOX RoPE doesn't need it) * fix: enable flash attention for JAIS2 (fixed by #19115) * fix: add dedicated JAIS2 pre-tokenizer type and control vector support - Add LLAMA_VOCAB_PRE_TYPE_JAIS2 with cascading whitespace regex - Include original regex from tokenizer.json as comment - Add build_cvec call for control vector support * no longer necessary to override set_vocab --------- Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>		2026-02-19 13:30:17 +01:00
..
afmoe.cpp	llama : refactor rope_freq_base/scale_swa conversion and init (#18553)	2026-01-05 09:14:04 +01:00
apertus.cpp
arcee.cpp
arctic.cpp
arwkv7.cpp
baichuan.cpp
bailingmoe.cpp
bailingmoe2.cpp
bert.cpp	model : add support for JinaBertModel with non-gated ffn (#18475)	2026-01-01 18:38:51 +01:00
bitnet.cpp
bloom.cpp
chameleon.cpp
chatglm.cpp
codeshell.cpp
cogvlm.cpp	graph : reduce topology branching (#18548)	2026-01-02 19:01:56 +02:00
cohere2-iswa.cpp	llama : refactor rope_freq_base/scale_swa conversion and init (#18553)	2026-01-05 09:14:04 +01:00
command-r.cpp
dbrx.cpp
deci.cpp
deepseek.cpp
deepseek2.cpp	model: support GLM MoE DSA arch (NOTE: indexer is not yet supported) (#19460)	2026-02-13 14:56:53 +01:00
delta-net-base.cpp	models : dedup Kimi Linear delta net implementation (#19668)	2026-02-19 08:15:17 +02:00
dots1.cpp
dream.cpp
ernie4-5-moe.cpp
ernie4-5.cpp	models : move build_inp_out_ids outside loop (#17151)	2025-11-10 22:55:30 +01:00
exaone-moe.cpp	model : add EXAONE MoE (#18543)	2026-01-13 23:28:38 +01:00
exaone.cpp
exaone4.cpp
falcon-h1.cpp	models : deduplicate delta-net graphs for Qwen family (#19597)	2026-02-16 14:35:04 +02:00
falcon.cpp
gemma-embedding.cpp	graph : reduce topology branching (#18548)	2026-01-02 19:01:56 +02:00
gemma.cpp
gemma2-iswa.cpp	llama : refactor rope_freq_base/scale_swa conversion and init (#18553)	2026-01-05 09:14:04 +01:00
gemma3.cpp	graph : reduce topology branching (#18548)	2026-01-02 19:01:56 +02:00
gemma3n-iswa.cpp	graph : utilize ggml_build_forward_select() to avoid reallocations (#18898)	2026-01-23 18:22:34 +02:00
glm4-moe.cpp	model: support GLM4V vision encoder (#18042)	2025-12-16 11:25:26 +01:00
glm4.cpp	model: support GLM-OCR (#19677)	2026-02-18 17:51:40 +01:00
gpt2.cpp
gptneox.cpp
granite-hybrid.cpp	models : deduplicate delta-net graphs for Qwen family (#19597)	2026-02-16 14:35:04 +02:00
granite.cpp
grok.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
grovemoe.cpp
hunyuan-dense.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
hunyuan-moe.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
internlm2.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
jais.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
jais2.cpp	model : add JAIS-2 architecture support (#19488)	2026-02-19 13:30:17 +01:00
jamba.cpp	models : deduplicate delta-net graphs for Qwen family (#19597)	2026-02-16 14:35:04 +02:00
kimi-linear.cpp	models : dedup Kimi Linear delta net implementation (#19668)	2026-02-19 08:15:17 +02:00
lfm2.cpp	model : add tokenizer from LFM2.5-Audio-1.5B (#19687)	2026-02-19 09:54:48 +01:00
llada-moe.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
llada.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
llama-iswa.cpp	llama : refactor rope_freq_base/scale_swa conversion and init (#18553)	2026-01-05 09:14:04 +01:00
llama.cpp	model : support for LlamaBidirectionalModel architecture (#18220)	2025-12-24 14:02:36 +01:00
maincoder.cpp	model : Maincoder-1B support (#18534)	2026-01-02 20:11:59 +01:00
mamba-base.cpp	models : deduplicate delta-net graphs for Qwen family (#19597)	2026-02-16 14:35:04 +02:00
mamba.cpp	models : deduplicate delta-net graphs for Qwen family (#19597)	2026-02-16 14:35:04 +02:00
mimo2-iswa.cpp	model: support MiMo-V2-Flash (#18328)	2025-12-24 23:07:08 +01:00
minicpm3.cpp	mla : make the V tensor a view of K (#18986)	2026-01-22 22:09:01 +02:00
minimax-m2.cpp
mistral3.cpp	model: support Ministral3 (#17644)	2025-12-01 12:26:52 +01:00
models.h	model : add JAIS-2 architecture support (#19488)	2026-02-19 13:30:17 +01:00
modern-bert.cpp	model : full modern bert support (#18330)	2026-02-19 08:52:21 +01:00
mpt.cpp
nemotron-h.cpp	models : deduplicate delta-net graphs for Qwen family (#19597)	2026-02-16 14:35:04 +02:00
nemotron.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
neo-bert.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
olmo.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
olmo2.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
olmoe.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
openai-moe-iswa.cpp	llama : refactor rope_freq_base/scale_swa conversion and init (#18553)	2026-01-05 09:14:04 +01:00
openelm.cpp	models : remove unnecessary cont in openelm (#19289)	2026-02-03 14:20:57 +01:00
orion.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
pangu-embedded.cpp	model : add openPangu-Embedded (#16941)	2025-11-05 10:28:58 +01:00
phi2.cpp
phi3.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
plamo.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
plamo2.cpp	models : deduplicate delta-net graphs for Qwen family (#19597)	2026-02-16 14:35:04 +02:00
plamo3.cpp	model : Plamo3 support (#17304)	2025-12-28 17:28:31 +01:00
plm.cpp	mla : make the V tensor a view of K (#18986)	2026-01-22 22:09:01 +02:00
qwen.cpp
qwen2.cpp	model : add KORMo model (#18032)	2025-12-15 18:51:43 +01:00
qwen2moe.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
qwen2vl.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
qwen3.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
qwen3moe.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
qwen3next.cpp	models : dedup Kimi Linear delta net implementation (#19668)	2026-02-19 08:15:17 +02:00
qwen3vl-moe.cpp	graph : utilize ggml_build_forward_select() to avoid reallocations (#18898)	2026-01-23 18:22:34 +02:00
qwen3vl.cpp	graph : utilize ggml_build_forward_select() to avoid reallocations (#18898)	2026-01-23 18:22:34 +02:00
qwen35.cpp	models : dedup qwen35 graphs (#19660)	2026-02-19 08:17:49 +02:00
qwen35moe.cpp	models : dedup qwen35 graphs (#19660)	2026-02-19 08:17:49 +02:00
refact.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
rnd1.cpp	models : Added support for RND1 Diffusion Language Model (#17433)	2025-11-24 14:16:56 +08:00
rwkv6-base.cpp	models : deduplicate delta-net graphs for Qwen family (#19597)	2026-02-16 14:35:04 +02:00
rwkv6.cpp
rwkv6qwen2.cpp
rwkv7-base.cpp	models : deduplicate delta-net graphs for Qwen family (#19597)	2026-02-16 14:35:04 +02:00
rwkv7.cpp
seed-oss.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
smallthinker.cpp	llama : refactor rope_freq_base/scale_swa conversion and init (#18553)	2026-01-05 09:14:04 +01:00
smollm3.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
stablelm.cpp
starcoder.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
starcoder2.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
step35-iswa.cpp	model : support Step3.5-Flash (#19283)	2026-02-06 21:06:14 +01:00
t5-dec.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
t5-enc.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
wavtokenizer-dec.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00
xverse.cpp	chore : fix models indent after refactor (#16992)	2025-11-04 12:29:15 +01:00