See axolotl config

axolotl version: 0.8.0.dev0

base_model: mistralai/Mistral-Nemo-Base-2407
# optionally might have model_type or tokenizer_type
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
# Automatically upload checkpoint and final model to HF
hub_model_id: AiAF/Pretraining-SCPWiki-032025-12B-Instruct

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: AiAF/Pretraining-SCPWiki-032025-7B-Instruct-pretraining.jsonl
    type: completion
dataset_prepared_path: last_run_prepared
val_set_size: 0.1
output_dir: ./outputs/out/Pretraining-SCPWiki-032025-12B-V1

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true
eval_sample_packing: false

wandb_project: "LLM-Pretraining"
wandb_entity:
wandb_watch: "all"
wandb_name: "Pretraining-SCPWiki-032025-12B-V1"
wandb_log_model: "false"

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 1
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.000005

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

loss_watchdog_threshold: 5.0
loss_watchdog_patience: 3

save_total_limit: 30
warmup_steps: 10
evals_per_epoch: 20
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 20
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  eos_token: "<|im_end|>"
  pad_token: "<pad>"
  bos_token: "<s>"
  unk_token: "<unk>"
tokens:
  - "<|im_start|>"

Pretraining-SCPWiki-032025-12B-Instruct

This model is a fine-tuned version of mistralai/Mistral-Nemo-Base-2407 on the AiAF/Pretraining-SCPWiki-032025-7B-Instruct-pretraining.jsonl dataset. It achieves the following results on the evaluation set:

• Loss: 1.5467

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: 5e-06
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_BNB 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: 10
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss
3.1576	0.0018	1	3.5143
1.4459	0.0511	29	1.6213
1.4502	0.1022	58	1.6003
1.5545	0.1534	87	1.5870
1.3624	0.2045	116	1.5779
1.3053	0.2556	145	1.5691
1.5688	0.3067	174	1.5635
1.7144	0.3579	203	1.5594
1.5199	0.4090	232	1.5550
1.2483	0.4601	261	1.5516
1.4053	0.5112	290	1.5493
1.4238	0.5624	319	1.5486
1.4939	0.6135	348	1.5477
1.4072	0.6646	377	1.5472
1.6039	0.7157	406	1.5469
1.3127	0.7669	435	1.5468
1.4754	0.8180	464	1.5466
1.5992	0.8691	493	1.5467
1.421	0.9202	522	1.5467
1.2666	0.9714	551	1.5467

Framework versions

• Transformers 4.49.0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0