See axolotl config
axolotl version: 0.8.0.dev0
# train w/ shisa-ai/shisa-v1-athenev2-reannotated-filtered
base_model: mistralai/Mistral-Nemo-Instruct-2407
tokenizer_type: AutoTokenizer
load_in_8bit: false
load_in_4bit: false
strict: false
# User Liger
plugins:
- axolotl.integrations.liger.LigerPlugin
liger_rope: true
liger_rms_norm: true
liger_glu_activation: true
liger_fused_linear_cross_entropy: true
chat_template: tokenizer_default
datasets:
- path: shisa-ai/shisa-v1-athenev2-reannotated-filtered
# type: sharegpt deprecated
type: chat_template
field_messages: conversations
message_field_role: from
message_field_content: value
- path: shisa-ai/shisa-v2-roleplaying
type: chat_template
field_messages: conversations
message_property_mappings:
role: role
content: content
roles:
system:
- system
assistant:
- gpt
- model
- assistant
user:
- human
- user
roles_to_train: ["assistant"]
dataset_prepared_path: last_run_prepared
val_set_size: 0.05
output_dir: ./outputs/ablation-53-rafathenev2.rp-shisa-v2-mistral-nemo-12b
sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true
# marginal difference
neftune_noise_alpha: 5
use_wandb: true
wandb_project: shisa-v2
wandb_entity: augmxnt
wandb_name: ablation-53-rafathenev2.rp-shisa-v2-mistral-nemo-12b
gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 3
optimizer: paged_adamw_8bit
lr_scheduler: linear
learning_rate: 8e-6
train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false
gradient_checkpointing: true
gradient_checkpointing_kwargs:
use_reentrant: false
early_stopping_patience:
resume_from_checkpoint:
logging_steps: 1
xformers_attention:
flash_attention: true
warmup_steps: 100
evals_per_epoch: 2
eval_table_size:
saves_per_epoch: 0
save_total_limit: 1 # Only store a single checkpoint
debug:
deepspeed: zero3_bf16.json
weight_decay: 1e-4
fsdp:
fsdp_config:
special_tokens:
pad_token: "<pad>"
outputs/ablation-53-rafathenev2.rp-shisa-v2-mistral-nemo-12b
This model is a fine-tuned version of mistralai/Mistral-Nemo-Instruct-2407 on the shisa-ai/shisa-v1-athenev2-reannotated-filtered and the shisa-ai/shisa-v2-roleplaying datasets. It achieves the following results on the evaluation set:
- Loss: 0.5469
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 8e-06
- train_batch_size: 2
- eval_batch_size: 2
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- gradient_accumulation_steps: 4
- total_train_batch_size: 64
- total_eval_batch_size: 16
- optimizer: Use OptimizerNames.PAGED_ADAMW_8BIT with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 100
- num_epochs: 3.0
Training results
Training Loss | Epoch | Step | Validation Loss |
---|---|---|---|
0.7371 | 0.0039 | 1 | 0.7454 |
0.548 | 0.5015 | 127 | 0.5406 |
0.1184 | 1.0 | 254 | 0.5144 |
0.3822 | 1.5015 | 381 | 0.5184 |
0.0922 | 2.0 | 508 | 0.5090 |
0.2699 | 2.5015 | 635 | 0.5469 |
Framework versions
- Transformers 4.49.0
- Pytorch 2.6.0+cu124
- Datasets 3.2.0
- Tokenizers 0.21.0
- Downloads last month
- 25
Inference Providers
NEW
This model is not currently available via any of the supported Inference Providers.