This model has a slightly better swipe diversity and a less flowery / verbose writing style. Reasoning tends to average out being a bit longer than the original however. Intelligence appears to be on par with the original.

Supports both thinking and non thinking.

02 SillyTavern Settings

Suggested Roleplay Format

ActionsIn plaintext

Dialogue"In quotes"

Thoughts*In asterisks*

Recommended Samplers

Temp0.8 - 1.0

MinP0.05

Instruct

Gemma 4 - Think

Gemma 4 - NoThink

03 Quantizations

GGUF

iMatrix

04 Creation Process

Creation Process: SFT > Merge

SFT on approx 49 million tokens.

Despite using 49 million tokens, this dataset is fairly modest in size. Trainable is somewhere in the rough ballpark of 10-15 million. All of the datasets were trained on the last turn only, to faithfully mirror the Gemma 4 chat template

The approach was very similar to the 24B A4B MeroMero. I trained the model aggressively for 2 epochs on my data and after testing various checkpoints, settled for the one at 1 epoch, which had the style and the least signs of overfitting.

I merged this checkpoint back into the original instruct which cleaned up any remaining overfitting while still retaining the changes of the finetune.

Trained using Axolotl.

Mergekit Config

models:
  - model: google/gemma-4-31B-it
  - model: ApocalypseParty/G4-31B-SFT-v3-1-1ep
merge_method: slerp
parameters:
  t: 0.5
base_model: google/gemma-4-31B-it
dtype: bfloat16

Axolotl Config

base_model: google/gemma-4-31B-it
 
plugins:
  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
  - axolotl.integrations.liger.LigerPlugin
liger_layer_norm: true
liger_rope: true
liger_rms_norm: true
liger_glu_activation: true
liger_rms_norm_gated: true
strict: false
cut_cross_entropy: true
 
datasets:
  - path: zerofata/prek_qwen36
val_set_size: 0.02
output_dir: ./G4-31B-SFT-v3-1
 
sequence_len: 10756
pad_to_sequence_len: true
sample_packing: true
 
load_in_4bit: false
adapter: lora
lora_r: 64
lora_alpha: 64
peft_use_rslora: true
lora_dropout: 0.0
freeze_mm_modules: true
 
lora_target_modules: 'model.language_model.layers.[\d]+.(_checkpoint_wrapped_module.)?(mlp|self_attn).(up|down|gate|q|k|v|o)_proj'
 
wandb_project: G4-31B-SFT
wandb_name: G4-31B-SFT-v3-1
 
gradient_accumulation_steps: 1
micro_batch_size: 4
num_epochs: 2
optimizer: adamw_torch_fused
lr_scheduler: constant_with_warmup
learning_rate: 1e-5
max_grad_norm: 1.0
 
bf16: auto
tf32: true
 
logging_steps: 1
 
# FA2 not supported
sdp_attention: true
#flex_attention: true
#torch_compile: true
flash_attention: false
 
warmup_ratio: 0.1
evals_per_epoch: 4
saves_per_epoch: 2
weight_decay: 0.05
special_tokens:
 
fsdp_config:
  fsdp_version: 2
  offload_params: false
  cpu_ram_efficient_loading: false
  auto_wrap_policy: TRANSFORMER_BASED_WRAP
  transformer_layer_cls_to_wrap: Gemma4TextDecoderLayer
  state_dict_type: FULL_STATE_DICT
  sharding_strategy: FULL_SHARD
  reshard_after_forward: true
  activation_checkpointing: true