AfriGuardPlain-tiny-aya-global-full

AfriGuardPlain-tiny-aya-global-full

CohereLabs/tiny-aya-global fine-tuned with full fine-tuning (DeepSpeed ZeRO-3) on adzcai/AfriGuard-plain: the non-instruction-prompted version of israel/AfriGuard-inst.

Each training example is only the user prompt and a plain reply (a helpful answer for safe prompts, a brief refusal for unsafe ones), with no safety system instruction and no <safety> / <category> / <response> tags. The model therefore answers or refuses directly, without emitting a safety label. Its instruction-prompted counterpart (same base model and hyperparameters, trained on AfriGuard-inst) is israel/AfriGuard-tiny-aya-global-full.

Training config: examples/train_full/afriguard_plain_tiny-aya-global_full_sft.yaml in the AfriGuard-model (LlamaFactory) repo; 1 epoch, 1 GPU, chat template of the base model, loss on the response only.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 1
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 4
• total_train_batch_size: 4
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 0.1
• num_epochs: 1.0

Training results

Framework versions

• Transformers 5.8.0
• Pytorch 2.14.0+cu130
• Datasets 4.0.0
• Tokenizers 0.22.2