google/gemma-4-31B-it · Discussions

#87 opened about 7 hours ago by

sigjhl

Jobayer

#85 opened 1 day ago by

Jy0018

When training models from the gemma4 series using GRPO, an abnormally high grad norm was observed

#84 opened 1 day ago by

mamazi00

add newlines and thinking tokens to template to avoid having to compute 3 extra tokens per generation in chat completion+reasoning

#83 opened 2 days ago by

quasar-of-mikus

Update README.md

#81 opened 4 days ago by

hectorruiz9

Gemma 4 models are way to paranoid about dates, any tips?

#80 opened 4 days ago by

Ahugm

Incorrect output in Gemma 4: seeking a solution to the problem

#79 opened 4 days ago by

Lintrarius

Fix chat_template: emit empty <|channel>thought\n<channel|> wrapper for existing asst turns

#78 opened 5 days ago by

flotherxi

[Bug] chat_template: missing <|channel>thought\n<channel|> wrapper for non-thinking SFT / multi-turn

#77 opened 5 days ago by

flotherxi

Thinking erratic at 30000+ context

#76 opened 6 days ago by

JeslynMcKenzie

Multilingual Support List

➕ 1

#75 opened 6 days ago by

abcdvzz

Will there be a small model like gemma-3-270m?

#74 opened 6 days ago by

ymcki

Unexpected loss spikes and performance degradation when fine-tuning Gemma 4 (google/gemma-4-31B-it)

#73 opened 7 days ago by

rstaruch

Add ParseBench evaluation results

4

#72 opened 12 days ago by

boyang-runllama

Will there be a small model for speculative decoding?

#71 opened 12 days ago by

Imagen 1 (2022) Should Be Open Sourced

👍 4

#70 opened 12 days ago by

Tralalabs

Question about tool-calling order in chat_template.jinja

#67 opened 13 days ago by

json0

gemma-4-31b-it unable to execute tool calling

#66 opened 13 days ago by

Naman2302

Do Gemma 4 models work well?

#65 opened 14 days ago by

fix: embed chat_template in tokenizer_config.json

#64 opened 16 days ago by

NERDDISCO

Infinite loop is not fixed even with Google API

👀 1

#63 opened 16 days ago by

alexcardo

Chat Template has a bug.

🤗 2

5

#62 opened 16 days ago by

Reithan

why print rightarrow

❤️👀 3

5

#61 opened 17 days ago by

wangtf-Kevin

Can anyone improve the model using the Rys methodology—by duplicating a block of layers?

11

#60 opened 18 days ago by

Strange behaviour of the tokenizer

#58 opened 19 days ago by

andercorral

Good Workflow

#57 opened 19 days ago by

anthoekfj

fix: function calling formatting in chat template

❤️ 2

#55 opened 20 days ago by

RyanMullins

Chat template is too complicated that even Gemma 4 itself has no idea how to parse it

#53 opened 20 days ago by

alexcardo

Hardware requirement

👍👀 3

13

#52 opened 20 days ago by

Charan01

Tokens per Image Parameter?

#51 opened 21 days ago by

buckeye17

Guys please add the MTP to this model

🔥 5

#50 opened 21 days ago by

Narutoouz

Will there be QAT models?

👍 11

#49 opened 21 days ago by

Gemma 4 E4B will be as encyclopedically well-read as the 12b model?

#48 opened 21 days ago by

Create BTS

#47 opened 21 days ago by deleted

brokersponsor

#46 opened 21 days ago by

Brokersponsor

Update README.md

#45 opened 21 days ago by

Brokersponsor

Qusetion about math_vision and mmmu_pro evaluation result

#44 opened 21 days ago by

JjjjjZzz

The Gemma 4 model is great. But...

👍 4

5

#43 opened 21 days ago by

suitup91

𝐎𝐍𝐋𝐘 𝐂𝐔𝐂𝐊𝐒 𝐀𝐍𝐃 𝐒𝐔𝐁𝐇𝐔𝐌𝐀𝐍𝐒 𝐅𝐎𝐋𝐋𝐎𝐖𝐒 𝐀𝚰 '𝐄𝐓𝐇𝚰𝐂𝐒' 𝐀𝐍𝐃 '𝐒𝐀𝐅𝐄𝐓𝐘'!

#42 opened 22 days ago by

Kickan

junk outputs

#41 opened 22 days ago by

rirv938

file size so big

#40 opened 22 days ago by

a9secondsleeper

How did you achieve such remarkable metrics?

🤝 2

6

#39 opened 22 days ago by

Rename README.md to README.md mmkk

#37 opened 22 days ago by

antromjjj

token masking and space_between_special_tokens during finetuning

#36 opened 22 days ago by

Darkhn

WHY IS NOT CODING WORK😤

🧠👀 5

4

#35 opened 22 days ago by

Drixpy

Featherless AI

#34 opened 23 days ago by

MarcosFRGames

Fantastic Model for Legal/English Language Use Cases

#33 opened 23 days ago by

md-1415

test it yourself