初始化项目,由ModelHub XC社区提供模型

Model: mnoukhov/pythia410m-sft-tldr
Source: Original Platform
This commit is contained in:
ModelHub XC
2026-05-30 14:21:21 +08:00
commit 4bb3a5553f
215 changed files with 118927 additions and 0 deletions

View File

@@ -0,0 +1,18 @@
model_name: "EleutherAI/pythia-410m-deduped"
log_with: "wandb"
learning_rate: 1e-5
lr_scheduler_type: cosine
num_warmup_steps: 100
weight_decay: 0.05
load_in_8bit: False
use_peft: False
fp16: True
bf16: False
streaming: False
gradient_checkpointing: False
gradient_accumulation_steps: 8
per_device_train_batch_size: 16
per_device_eval_batch_size: 8
train_completions: True
num_train_epochs: 1
seq_length: 560