spacethinker-lora

This model is a fine-tuned version of UCSC-VLAA/VLAA-Thinker-Qwen2.5VL-3B on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.9785

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: 2e-05
• train_batch_size: 2
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• total_eval_batch_size: 2
• 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: constant
• lr_scheduler_warmup_ratio: 0.03
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
2.548	0.0280	10	2.3356
2.2084	0.0561	20	2.1089
2.0353	0.0841	30	1.9475
1.8831	0.1121	40	1.7890
1.7139	0.1402	50	1.5906
1.524	0.1682	60	1.4117
1.3676	0.1962	70	1.3128
1.3054	0.2242	80	1.2428
1.2321	0.2523	90	1.1900
1.1896	0.2803	100	1.1618
1.1824	0.3083	110	1.1319
1.122	0.3364	120	1.1163
1.1403	0.3644	130	1.1027
1.13	0.3924	140	1.0911
1.0914	0.4205	150	1.0831
1.1097	0.4485	160	1.0772
1.0938	0.4765	170	1.0724
1.0947	0.5046	180	1.0688
1.0737	0.5326	190	1.0650
1.0742	0.5606	200	1.0613
1.0755	0.5886	210	1.0583
1.0797	0.6167	220	1.0549
1.0644	0.6447	230	1.0528
1.0834	0.6727	240	1.0507
1.0617	0.7008	250	1.0481
1.0713	0.7288	260	1.0460
1.0602	0.7568	270	1.0437
1.0581	0.7849	280	1.0414
1.0652	0.8129	290	1.0395
1.0597	0.8409	300	1.0377
1.0551	0.8690	310	1.0360
1.0525	0.8970	320	1.0344
1.0442	0.9250	330	1.0327
1.032	0.9530	340	1.0307
1.0553	0.9811	350	1.0293
1.1467	1.0112	360	1.0282
1.0378	1.0392	370	1.0263
1.0568	1.0673	380	1.0247
1.0298	1.0953	390	1.0240
1.0403	1.1233	400	1.0221
1.051	1.1514	410	1.0215
1.0289	1.1794	420	1.0198
1.0389	1.2074	430	1.0190
1.0348	1.2355	440	1.0175
1.0379	1.2635	450	1.0161
1.0507	1.2915	460	1.0152
1.0195	1.3196	470	1.0142
1.0084	1.3476	480	1.0125
1.0317	1.3756	490	1.0115
1.0319	1.4036	500	1.0107
1.0193	1.4317	510	1.0094
1.034	1.4597	520	1.0089
1.0311	1.4877	530	1.0077
1.0497	1.5158	540	1.0071
1.0417	1.5438	550	1.0061
1.0307	1.5718	560	1.0049
1.0028	1.5999	570	1.0042
1.0192	1.6279	580	1.0036
1.007	1.6559	590	1.0023
1.0378	1.6840	600	1.0020
0.9979	1.7120	610	1.0011
1.0169	1.7400	620	1.0004
1.0148	1.7680	630	0.9999
1.0095	1.7961	640	0.9989
1.0252	1.8241	650	0.9984
0.9891	1.8521	660	0.9983
1.0598	1.8802	670	0.9969
1.0158	1.9082	680	0.9964
1.019	1.9362	690	0.9961
0.9979	1.9643	700	0.9949
1.0312	1.9923	710	0.9946
1.084	2.0224	720	0.9938
0.9932	2.0505	730	0.9937
0.9932	2.0785	740	0.9930
1.0138	2.1065	750	0.9921
1.002	2.1345	760	0.9921
1.0291	2.1626	770	0.9914
1.0171	2.1906	780	0.9908
0.9959	2.2186	790	0.9902
1.0181	2.2467	800	0.9897
0.9856	2.2747	810	0.9893
1.0141	2.3027	820	0.9888
1.0305	2.3308	830	0.9883
0.9911	2.3588	840	0.9875
0.996	2.3868	850	0.9877
0.984	2.4149	860	0.9869
0.9964	2.4429	870	0.9864
1.0101	2.4709	880	0.9856
0.9934	2.4989	890	0.9853
1.0432	2.5270	900	0.9848
0.9918	2.5550	910	0.9843
0.9977	2.5830	920	0.9844
1.009	2.6111	930	0.9834
0.9994	2.6391	940	0.9837
0.9972	2.6671	950	0.9830
1.0043	2.6952	960	0.9827
1.0005	2.7232	970	0.9823
0.9888	2.7512	980	0.9820
0.9917	2.7793	990	0.9813
1.0036	2.8073	1000	0.9810
0.984	2.8353	1010	0.9803
0.9696	2.8633	1020	0.9798
1.0062	2.8914	1030	0.9798
1.0001	2.9194	1040	0.9793
1.0214	2.9474	1050	0.9796
1.0106	2.9755	1060	0.9785

Framework versions

• PEFT 0.15.1
• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1