Gemma 4 26B-A4B GGUF Benchmarks
pinnedππ₯ 8
3
#35 opened 3 days ago
by
danielhanchen
Apr 11: Updated with Google chat template fixes + more
pinnedπβ€οΈ 16
11
#24 opened 12 days ago
by
danielhanchen
Gemma 4 Tool Calling is amazing in Unsloth Studio!
pinnedπ₯ 5
4
#4 opened 21 days ago
by
danielhanchen
Tool call tokens use hybrid format incompatible with Ollama's PARSER gemma4
#36 opened about 16 hours ago
by
kbryss
So what's the difference between the IQ4_XS and IQ4_NL quants?
#34 opened 4 days ago
by
xwildfyre
Quant for non-Latin languages
#33 opened 8 days ago
by
Theory-of-mind
How to set image budget?
1
#32 opened 8 days ago
by
IlysvlVEizbr
[BUG] Vision capability not working for Gemma4 GGUF in Ollama
2
#31 opened 9 days ago
by
notuncommon
Weird issue with closing </div>
#30 opened 9 days ago
by
Fxvoid
Roleplay viability
2
#29 opened 10 days ago
by
yano2mch
Chat template is busted
1
#28 opened 10 days ago
by
FrenzyBiscuit
Apr 11 chat template causes ~7.5s template rendering overhead per request in llama.cpp
#27 opened 11 days ago
by
btdeviant
llama.cpp flags / visual token budget
#26 opened 11 days ago
by
234r89r23u89023rui90
D:\a\llama.cpp\llama.cpp\ggml\src\ggml-cuda\ggml-cuda.cu:911: GGML_ASSERT(tensor->view_src == nullptr) failed
#25 opened 11 days ago
by
osabc
Do NOT use CUDA 13.2
β€οΈ 11
#22 opened 14 days ago
by
danielhanchen
Gemma 4 seems to work best with high temperature for coding
π 1
8
#21 opened 14 days ago
by
Reverger
Apr 8 - New GGUF Updates
πβ€οΈ 14
10
#20 opened 15 days ago
by
danielhanchen
gguf updates
π 5
1
#17 opened 16 days ago
by
tstello
Ollama Error
π 1
4
#16 opened 16 days ago
by
edm-research
Inference speed on 12GB VRAM
6
#15 opened 16 days ago
by
drakexp
Fails to run on vLLM
3
#14 opened 17 days ago
by
Skodra
Only 2nd <13GB model to one-shot the Heptagon-Tumbler
β€οΈπ₯ 3
1
#12 opened 18 days ago
by
BingoBird
New uploads adds llama.cpp fixes
π 6
16
#11 opened 19 days ago
by
danielhanchen
Commit description
π 4
1
#10 opened 19 days ago
by
Kelheor
Q4_0 and Q4_1?
π 1
#9 opened 19 days ago
by
elpirater312
How to enable thinking
πβ€οΈ 9
6
#6 opened 21 days ago
by
watchingyousleep
Tool call with dates fails
2
#5 opened 21 days ago
by
EmilPi
Model produces `<|channel><unused49><unused49><unused49>`
π 5
35
#2 opened 21 days ago
by
kyuz0