See axolotl config

axolotl version: 0.13.0.dev0

base_model: HuggingFaceTB/SmolLM3-3B
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

datasets:
  - path: woodman231/monster-sanctuary-conversations
    type: chat_template
    chat_template: chatml

dataset_prepared_path: /workspace/data/axolotl_prepared_data
val_set_size: 0.1
output_dir: /workspace/data/outputs

special_tokens:
  bos_token: <|im_start|>
  eos_token: <|im_end|>
  pad_token: <|im_end|>

# FULL PRECISION FOR QUALITY
load_in_4bit: false             # No quantization on H100
adapter: lora                   # Regular LoRA
lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true

# LARGE BATCH SETTINGS
gradient_accumulation_steps: 1  # No accumulation needed
micro_batch_size: 32            # Max batch size (can try 24-32)

# LEARNING RATE
learning_rate: 0.00004          # Slightly higher LR for full precision

# FULL TRAINING
num_epochs: 8                   # Full 8 epochs
# max_steps: not set            # Train to completion
eval_steps: 100                 # Evaluate every 100 steps

# H100 OPTIMIZATIONS
flash_attention: true           # H100 has flash attention support
bf16: true                      # BFloat16 for H100
gradient_checkpointing: true    # Trade compute for memory (optional)

# OUTPUT
hub_model_id: woodman231/SmolLM3-3B-monster-sanctuary-lora
save_steps: 100                 # Save every 100 steps
save_total_limit: 3             # Keep only 3 checkpoints

SmolLM3-3B-monster-sanctuary-lora

This model is a fine-tuned version of HuggingFaceTB/SmolLM3-3B on the woodman231/monster-sanctuary-conversations dataset. It achieves the following results on the evaluation set:

Loss: 0.0002
Memory/max Active (gib): 26.77
Memory/max Allocated (gib): 26.77
Memory/device Reserved (gib): 37.58

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: 4e-05
train_batch_size: 32
eval_batch_size: 32
seed: 42
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: 100
training_steps: 13384

Training results

Training Loss	Epoch	Step	Validation Loss	Active (gib)	Allocated (gib)	Reserved (gib)
No log	0	0	2.9715	26.28	26.28	26.43
0.8052	0.0598	100	0.7336	26.77	26.77	37.52
0.4467	0.1195	200	0.3855	26.77	26.77	37.58
0.2207	0.1793	300	0.2723	26.77	26.77	37.58
0.233	0.2391	400	0.2125	26.77	26.77	37.58
0.157	0.2989	500	0.1825	26.77	26.77	37.58
0.1464	0.3586	600	0.1664	26.77	26.77	37.58
0.1278	0.4184	700	0.1466	26.77	26.77	37.58
0.1467	0.4782	800	0.1251	26.77	26.77	37.58
0.1148	0.5380	900	0.1141	26.77	26.77	37.58
0.1086	0.5977	1000	0.1078	26.77	26.77	37.58
0.1137	0.6575	1100	0.0998	26.77	26.77	37.58
0.0975	0.7173	1200	0.0896	26.77	26.77	37.67
0.092	0.7770	1300	0.0857	26.77	26.77	37.58
0.0677	0.8368	1400	0.0838	26.77	26.77	37.58
0.0704	0.8966	1500	0.0749	26.77	26.77	37.58
0.0914	0.9564	1600	0.0683	26.77	26.77	37.58
0.0641	1.0161	1700	0.0605	26.77	26.77	37.58
0.0556	1.0759	1800	0.0569	26.77	26.77	37.67
0.0495	1.1357	1900	0.0546	26.77	26.77	37.5
0.0402	1.1955	2000	0.0507	26.77	26.77	37.58
0.0442	1.2552	2100	0.0472	26.77	26.77	37.58
0.0355	1.3150	2200	0.0446	26.77	26.77	37.58
0.0352	1.3748	2300	0.0393	26.77	26.77	37.58
0.0288	1.4345	2400	0.0373	26.77	26.77	37.48
0.036	1.4943	2500	0.0341	26.77	26.77	37.58
0.0299	1.5541	2600	0.0299	26.77	26.77	37.58
0.0282	1.6139	2700	0.0279	26.77	26.77	37.58
0.0212	1.6736	2800	0.0248	26.77	26.77	37.58
0.0252	1.7334	2900	0.0231	26.77	26.77	37.58
0.0234	1.7932	3000	0.0201	26.77	26.77	37.5
0.0175	1.8530	3100	0.0188	26.77	26.77	37.5
0.0154	1.9127	3200	0.0154	26.77	26.77	37.58
0.0166	1.9725	3300	0.0141	26.77	26.77	37.58
0.0123	2.0323	3400	0.0135	26.77	26.77	37.5
0.0112	2.0921	3500	0.0130	26.77	26.77	37.58
0.0119	2.1518	3600	0.0119	26.77	26.77	37.58
0.0079	2.2116	3700	0.0118	26.77	26.77	37.58
0.0105	2.2714	3800	0.0113	26.77	26.77	37.58
0.0072	2.3311	3900	0.0087	26.77	26.77	37.67
0.0067	2.3909	4000	0.0080	26.77	26.77	37.58
0.0081	2.4507	4100	0.0090	26.77	26.77	37.58
0.0058	2.5105	4200	0.0088	26.77	26.77	37.58
0.008	2.5702	4300	0.0075	26.77	26.77	37.58
0.006	2.6300	4400	0.0068	26.77	26.77	37.65
0.0063	2.6898	4500	0.0076	26.77	26.77	37.58
0.0066	2.7496	4600	0.0074	26.77	26.77	37.5
0.0059	2.8093	4700	0.0066	26.77	26.77	37.58
0.0044	2.8691	4800	0.0050	26.77	26.77	37.5
0.0061	2.9289	4900	0.0070	26.77	26.77	37.58
0.0043	2.9886	5000	0.0057	26.77	26.77	37.58
0.0027	3.0484	5100	0.0050	26.77	26.77	37.5
0.003	3.1082	5200	0.0048	26.77	26.77	37.58
0.0056	3.1680	5300	0.0039	26.77	26.77	37.58
0.0026	3.2277	5400	0.0044	26.77	26.77	37.48
0.0022	3.2875	5500	0.0038	26.77	26.77	37.58
0.0064	3.3473	5600	0.0033	26.77	26.77	37.58
0.004	3.4071	5700	0.0036	26.77	26.77	37.58
0.0038	3.4668	5800	0.0029	26.77	26.77	37.58
0.0017	3.5266	5900	0.0038	26.77	26.77	37.58
0.0016	3.5864	6000	0.0028	26.77	26.77	37.58
0.0018	3.6461	6100	0.0027	26.77	26.77	37.58
0.0037	3.7059	6200	0.0041	26.77	26.77	37.58
0.0016	3.7657	6300	0.0029	26.77	26.77	37.58
0.0012	3.8255	6400	0.0022	26.77	26.77	37.58
0.0027	3.8852	6500	0.0023	26.77	26.77	37.58
0.0009	3.9450	6600	0.0021	26.77	26.77	37.48
0.001	4.0048	6700	0.0029	26.77	26.77	37.58
0.0012	4.0646	6800	0.0015	26.77	26.77	37.58
0.0019	4.1243	6900	0.0016	26.77	26.77	37.58
0.0009	4.1841	7000	0.0022	26.77	26.77	37.58
0.0008	4.2439	7100	0.0016	26.77	26.77	37.58
0.0026	4.3036	7200	0.0015	26.77	26.77	37.5
0.0018	4.3634	7300	0.0012	26.77	26.77	37.58
0.0009	4.4232	7400	0.0015	26.77	26.77	37.67
0.0004	4.4830	7500	0.0018	26.77	26.77	37.58
0.001	4.5427	7600	0.0012	26.77	26.77	37.58
0.0026	4.6025	7700	0.0011	26.77	26.77	37.58
0.0005	4.6623	7800	0.0014	26.77	26.77	37.58
0.0007	4.7221	7900	0.0010	26.77	26.77	37.58
0.0011	4.7818	8000	0.0011	26.77	26.77	37.67
0.0004	4.8416	8100	0.0007	26.77	26.77	37.58
0.0003	4.9014	8200	0.0007	26.77	26.77	37.58
0.0009	4.9611	8300	0.0009	26.77	26.77	37.58
0.0004	5.0209	8400	0.0010	26.77	26.77	37.58
0.0009	5.0807	8500	0.0009	26.77	26.77	37.58
0.0006	5.1405	8600	0.0006	26.77	26.77	37.58
0.0003	5.2002	8700	0.0005	26.77	26.77	37.58
0.0002	5.2600	8800	0.0005	26.77	26.77	37.58
0.0002	5.3198	8900	0.0004	26.77	26.77	37.58
0.0005	5.3796	9000	0.0006	26.77	26.77	37.67
0.0003	5.4393	9100	0.0005	26.77	26.77	37.58
0.0006	5.4991	9200	0.0015	26.77	26.77	37.58
0.0001	5.5589	9300	0.0005	26.77	26.77	37.48
0.0002	5.6186	9400	0.0004	26.77	26.77	37.58
0.0005	5.6784	9500	0.0004	26.77	26.77	37.58
0.0006	5.7382	9600	0.0003	26.77	26.77	37.58
0.0002	5.7980	9700	0.0003	26.77	26.77	37.58
0.0001	5.8577	9800	0.0003	26.77	26.77	37.58
0.0004	5.9175	9900	0.0003	26.77	26.77	37.58
0.0003	5.9773	10000	0.0003	26.77	26.77	37.58
0.0001	6.0371	10100	0.0003	26.77	26.77	37.58
0.0001	6.0968	10200	0.0003	26.77	26.77	37.58
0.0004	6.1566	10300	0.0003	26.77	26.77	37.58
0.0001	6.2164	10400	0.0002	26.77	26.77	37.58
0.0001	6.2762	10500	0.0002	26.77	26.77	37.58
0.0001	6.3359	10600	0.0002	26.77	26.77	37.58
0.0001	6.3957	10700	0.0002	26.77	26.77	37.58
0.0003	6.4555	10800	0.0002	26.77	26.77	37.58
0.0001	6.5152	10900	0.0002	26.77	26.77	37.58
0.0001	6.5750	11000	0.0002	26.77	26.77	37.5
0.0001	6.6348	11100	0.0002	26.77	26.77	37.58
0.0001	6.6946	11200	0.0002	26.77	26.77	37.58
0.0001	6.7543	11300	0.0002	26.77	26.77	37.58
0.0001	6.8141	11400	0.0002	26.77	26.77	37.58
0.0001	6.8739	11500	0.0002	26.77	26.77	37.58
0.0001	6.9337	11600	0.0002	26.77	26.77	37.58
0.0001	6.9934	11700	0.0002	26.77	26.77	37.58
0.0001	7.0532	11800	0.0002	26.77	26.77	37.58
0.0001	7.1130	11900	0.0002	26.77	26.77	37.48
0.0	7.1727	12000	0.0002	26.77	26.77	37.58
0.0001	7.2325	12100	0.0002	26.77	26.77	37.48
0.0001	7.2923	12200	0.0002	26.77	26.77	37.5
0.0001	7.3521	12300	0.0002	26.77	26.77	37.58
0.0001	7.4118	12400	0.0002	26.77	26.77	37.58
0.0001	7.4716	12500	0.0002	26.77	26.77	37.58
0.0001	7.5314	12600	0.0002	26.77	26.77	37.58
0.0001	7.5912	12700	0.0002	26.77	26.77	37.58
0.0001	7.6509	12800	0.0002	26.77	26.77	37.58
0.0	7.7107	12900	0.0002	26.77	26.77	37.5
0.0001	7.7705	13000	0.0002	26.77	26.77	37.58
0.0001	7.8302	13100	0.0002	26.77	26.77	37.58
0.0001	7.8900	13200	0.0002	26.77	26.77	37.58
0.0001	7.9498	13300	0.0002	26.77	26.77	37.58

Framework versions

PEFT 0.18.0
Transformers 4.57.1
Pytorch 2.8.0+cu128
Datasets 4.4.1
Tokenizers 0.22.1

woodman231/SmolLM3-3B-monster-sanctuary-lora

SmolLM3-3B-monster-sanctuary-lora

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

为此模型做种