Gemma4
axteam's Gemma 4 quantizations — FP8 dynamic, AWQ INT4/INT8, NVFP4 (W4A4 & W4A16), and GGUF — for 12B-it, E4B-it, 26B-A4B(+it), 31B(+it).
Image-Text-to-Text • 12B • Updated • 257 • 4Note gemma-4-12B-it · FP8 dynamic (W8A8) — near-lossless; Hopper/Ada/Blackwell
axteam/gemma-4-12B-it-AWQ-8bit
Image-Text-to-Text • 12B • Updated • 254 • 4Note gemma-4-12B-it · AWQ W8A16 INT8 — near-lossless
axteam/gemma-4-12B-it-AWQ-4bit
Image-Text-to-Text • 12B • Updated • 451 • 2Note gemma-4-12B-it · AWQ W4A16 INT4 — ~1/4 size
axteam/gemma-4-12B-it-NVFP4A16
Image-Text-to-Text • 12B • Updated • 247 • 1Note gemma-4-12B-it · NVFP4 W4A16 (FP4 weights, BF16 activations) — best 4-bit quality
axteam/gemma-4-12B-it-NVFP4
Image-Text-to-Text • 12B • Updated • 295 • 2Note gemma-4-12B-it · NVFP4 W4A4 (FP4 weights + activations) — fastest on Blackwell
axteam/gemma-4-12B-it-GGUF
Text Generation • 12B • Updated • 2.94k • 2Note gemma-4-12B-it · GGUF (Q4_K_M / Q8_0 / BF16) for llama.cpp / Ollama
axteam/gemma-4-E4B-it-FP8-dynamic
Image-Text-to-Text • 8B • Updated • 366 • 4Note gemma-4-E4B-it · FP8 dynamic (W8A8) — near-lossless; Hopper/Ada/Blackwell
axteam/gemma-4-E4B-it-AWQ-8bit
Image-Text-to-Text • 8B • Updated • 321 • 3Note gemma-4-E4B-it · AWQ W8A16 INT8 — near-lossless
axteam/gemma-4-E4B-it-AWQ-4bit
Image-Text-to-Text • 8B • Updated • 315 • 2Note gemma-4-E4B-it · AWQ W4A16 INT4 — ~1/4 size
axteam/gemma-4-E4B-it-NVFP4A16
Image-Text-to-Text • 8B • Updated • 297 • 3Note gemma-4-E4B-it · NVFP4 W4A16 (FP4 weights, BF16 activations) — best 4-bit quality
axteam/gemma-4-E4B-it-NVFP4
Image-Text-to-Text • 8B • Updated • 361 • 4Note gemma-4-E4B-it · NVFP4 W4A4 (FP4 weights + activations) — fastest on Blackwell
axteam/gemma-4-E4B-it-GGUF
Text Generation • 8B • Updated • 2.7k • 5Note gemma-4-E4B-it · GGUF (Q4_K_M / Q8_0 / BF16) for llama.cpp / Ollama
axteam/gemma-4-26B-A4B-FP8-dynamic
Image-Text-to-Text • 26B • Updated • 258 • 2Note gemma-4-26B-A4B · FP8 dynamic (W8A8) — near-lossless; Hopper/Ada/Blackwell
axteam/gemma-4-26B-A4B-AWQ-8bit
Image-Text-to-Text • 9B • Updated • 191 • 1Note gemma-4-26B-A4B · AWQ W8A16 INT8 — near-lossless
axteam/gemma-4-26B-A4B-AWQ-4bit
Image-Text-to-Text • 26B • Updated • 273 • 2Note gemma-4-26B-A4B · AWQ W4A16 INT4 — ~1/4 size
axteam/gemma-4-26B-A4B-NVFP4A16
Image-Text-to-Text • 26B • Updated • 205 • 1Note gemma-4-26B-A4B · NVFP4 W4A16 (FP4 weights, BF16 activations) — best 4-bit quality
axteam/gemma-4-26B-A4B-NVFP4
Image-Text-to-Text • 26B • Updated • 200 • 1Note gemma-4-26B-A4B · NVFP4 W4A4 (FP4 weights + activations) — fastest on Blackwell
axteam/gemma-4-26B-A4B-GGUF
Text Generation • 25B • Updated • 1.91k • 2Note gemma-4-26B-A4B · GGUF (Q4_K_M / Q8_0 / BF16) for llama.cpp / Ollama
axteam/gemma-4-26B-A4B-it-FP8-dynamic
Image-Text-to-Text • 26B • Updated • 220 • 1Note gemma-4-26B-A4B-it · FP8 dynamic (W8A8) — near-lossless; Hopper/Ada/Blackwell
axteam/gemma-4-26B-A4B-it-AWQ-8bit
Image-Text-to-Text • 9B • Updated • 191 • 1Note gemma-4-26B-A4B-it · AWQ W8A16 INT8 — near-lossless
axteam/gemma-4-26B-A4B-it-AWQ-4bit
Image-Text-to-Text • 26B • Updated • 688 • 2Note gemma-4-26B-A4B-it · AWQ W4A16 INT4 — ~1/4 size
axteam/gemma-4-26B-A4B-it-NVFP4A16
Image-Text-to-Text • 26B • Updated • 365 • 5Note gemma-4-26B-A4B-it · NVFP4 W4A16 (FP4 weights, BF16 activations) — best 4-bit quality
axteam/gemma-4-26B-A4B-it-NVFP4
Image-Text-to-Text • 26B • Updated • 268 • 2Note gemma-4-26B-A4B-it · NVFP4 W4A4 (FP4 weights + activations) — fastest on Blackwell
axteam/gemma-4-26B-A4B-it-GGUF
Text Generation • 25B • Updated • 2.65k • 4Note gemma-4-26B-A4B-it · GGUF (Q4_K_M / Q8_0 / BF16) for llama.cpp / Ollama
axteam/gemma-4-31B-FP8-dynamic
Image-Text-to-Text • 31B • Updated • 332 • 2Note gemma-4-31B · FP8 dynamic (W8A8) — near-lossless; Hopper/Ada/Blackwell
axteam/gemma-4-31B-AWQ-8bit
Image-Text-to-Text • 10B • Updated • 328 • 3Note gemma-4-31B · AWQ W8A16 INT8 — near-lossless
axteam/gemma-4-31B-AWQ-4bit
Image-Text-to-Text • 7B • Updated • 247 • 2Note gemma-4-31B · AWQ W4A16 INT4 — ~1/4 size
axteam/gemma-4-31B-NVFP4A16
Image-Text-to-Text • 18B • Updated • 288 • 2Note gemma-4-31B · NVFP4 W4A16 (FP4 weights, BF16 activations) — best 4-bit quality
axteam/gemma-4-31B-NVFP4
Image-Text-to-Text • 18B • Updated • 327 • 2Note gemma-4-31B · NVFP4 W4A4 (FP4 weights + activations) — fastest on Blackwell
axteam/gemma-4-31B-GGUF
Text Generation • 31B • Updated • 2.09k • 2Note gemma-4-31B · GGUF (Q4_K_M / Q8_0 / BF16) for llama.cpp / Ollama
axteam/gemma-4-31B-it-FP8-dynamic
Image-Text-to-Text • 31B • Updated • 280 • 4Note gemma-4-31B-it · FP8 dynamic (W8A8) — near-lossless; Hopper/Ada/Blackwell
axteam/gemma-4-31B-it-AWQ-8bit
Image-Text-to-Text • 10B • Updated • 210 • 2Note gemma-4-31B-it · AWQ W8A16 INT8 — near-lossless
axteam/gemma-4-31B-it-AWQ-4bit
Image-Text-to-Text • 7B • Updated • 570 • 2Note gemma-4-31B-it · AWQ W4A16 INT4 — ~1/4 size
axteam/gemma-4-31B-it-NVFP4A16
Image-Text-to-Text • 18B • Updated • 346 • 5Note gemma-4-31B-it · NVFP4 W4A16 (FP4 weights, BF16 activations) — best 4-bit quality
axteam/gemma-4-31B-it-NVFP4
Image-Text-to-Text • 18B • Updated • 362 • 7Note gemma-4-31B-it · NVFP4 W4A4 (FP4 weights + activations) — fastest on Blackwell
axteam/gemma-4-31B-it-GGUF
Text Generation • 31B • Updated • 1.88k • 5Note gemma-4-31B-it · GGUF (Q4_K_M / Q8_0 / BF16) for llama.cpp / Ollama
axteam/gemma-4-E2B-it-GGUF
Text Generation • 5B • Updated • 2.04k • 3Note gemma-4-E2B-it · GGUF (Q4_K_M / Q8_0 / BF16) for llama.cpp / Ollama
axteam/gemma-4-E2B-it-FP8-dynamic
Image-Text-to-Text • 5B • Updated • 312 • 2Note gemma-4-E2B-it · FP8 dynamic (W8A8) — near-lossless; Hopper/Ada/Blackwell
axteam/gemma-4-E2B-it-AWQ-8bit
Image-Text-to-Text • 5B • Updated • 299 • 2Note gemma-4-E2B-it · AWQ W8A16 INT8 — near-lossless
axteam/gemma-4-E2B-it-AWQ-4bit
Image-Text-to-Text • 5B • Updated • 319 • 2Note gemma-4-E2B-it · AWQ W4A16 INT4 — ~1/4 size
axteam/gemma-4-E2B-it-NVFP4A16
Image-Text-to-Text • 5B • Updated • 330 • 2Note gemma-4-E2B-it · NVFP4 W4A16 (FP4 weights, BF16 activations) — best 4-bit quality
axteam/gemma-4-E2B-it-NVFP4
Image-Text-to-Text • 5B • Updated • 326 • 2Note gemma-4-E2B-it · NVFP4 W4A4 (FP4 weights + activations) — fastest on Blackwell