This model is a fine-tuned version of Qwen/Qwen2.5-1.5B-Instruct on the 7b_math_95k_2_train, the 7b_code_100k_2_train and the 7b_instruction_100k_2_train datasets. It achieves the following results on the evaluation set:

Loss: 0.0044
Token Mean Mae: 1667898715.5410
Token Mean Rmse: 829237.4977
Token Mean Seq Mean Mae: 58941.5716
Token Mean Seq Mean Rmse: 2405.8500
Token Mean Relerr: 0.3010
Token Mean Seq Mean Relerr: 0.3483

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: 8
eval_batch_size: 8
seed: 0
distributed_type: multi-GPU
num_devices: 8
gradient_accumulation_steps: 16
total_train_batch_size: 1024
total_eval_batch_size: 64
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
num_epochs: 2.0

Training results

Training Loss	Epoch	Step	Validation Loss	Token Mean Mae	Token Mean Rmse	Token Mean Seq Mean Mae	Token Mean Seq Mean Rmse	Token Mean Relerr	Token Mean Seq Mean Relerr
0.1091	0.0868	50	0.0079	2114620342.4301	975319.1954	74041.4268	3094.4201	0.4852	0.6094
0.0833	0.1736	100	0.0056	2004614063.9349	967153.4426	70343.2373	2846.9168	0.3504	0.4287
0.0808	0.2605	150	0.0052	1844904106.9341	901717.9613	64873.9665	2640.0881	0.3428	0.3878
0.0779	0.3473	200	0.0049	1816613973.2032	896552.7070	63922.2482	2590.4385	0.3288	0.3776
0.0752	0.4341	250	0.0049	1734559138.3595	852339.8631	61078.8151	2501.3052	0.3350	0.4076
0.0767	0.5209	300	0.0049	1777580395.4504	877835.4934	62422.9097	2532.6624	0.3442	0.4035
0.0752	0.6078	350	0.0047	1747704572.4626	869533.9426	61325.1604	2509.6978	0.3072	0.3447
0.0716	0.6946	400	0.0046	1740305536.6010	864221.9411	61402.4703	2499.7285	0.3160	0.3630
0.0693	0.7814	450	0.0046	1777698185.3464	889635.5166	62513.6725	2526.2048	0.3031	0.3338
0.0700	0.8682	500	0.0046	1742391892.7631	865465.9366	61463.7000	2489.8043	0.3309	0.3780
0.0741	0.9551	550	0.0045	1729399670.6567	861427.8125	60807.6753	2469.3901	0.2963	0.3366
0.0563	1.0417	600	0.0045	1745160175.3986	868709.1215	61411.6929	2490.3916	0.2906	0.3243
0.0574	1.1285	650	0.0045	1733981739.0859	861189.7798	61028.6462	2472.3316	0.2978	0.3285
0.0597	1.2153	700	0.0046	1746525720.9338	866094.1289	61438.0043	2493.2958	0.2862	0.3191
0.0592	1.3021	750	0.0047	1807411461.6055	890670.7002	63807.7708	2558.2878	0.2874	0.3133
0.0592	1.3890	800	0.0046	1744567760.4542	857970.4265	61819.2170	2506.4203	0.2826	0.3069
0.0571	1.4758	850	0.0045	1698453853.9144	843351.3908	59941.4237	2436.3356	0.3002	0.3399
0.0583	1.5626	900	0.0045	1674186696.0527	830514.9445	59018.2140	2414.6785	0.3086	0.3615
0.0583	1.6494	950	0.0046	1751055790.9167	865419.9649	61876.0072	2498.4548	0.2861	0.3155
0.0644	1.7363	1000	0.0044	1667831139.5895	815780.6399	59169.6027	2423.3030	0.3005	0.3342
0.0655	1.8231	1050	0.0044	1711587558.1778	854067.8211	60262.8130	2440.6906	0.2958	0.3278
0.0623	1.9099	1100	0.0044	1698326682.4013	855002.2224	59755.1559	2422.2892	0.2910	0.3298
0.0564	1.9967	1150	0.0044	1705119562.8685	854411.8960	60067.3141	2438.6407	0.2901	0.3296

Framework versions

Transformers 5.0.0
Pytorch 2.10.0+cu128
Datasets 3.6.0
Tokenizers 0.22.2

namezz/lvm-a-qwen2.5-7b-instruct-b-qwen2.5-1.5b-instruct

abs-bf16-7b-math-code-instruction-lr2e-5-g0.997-l1.0-gpu8-bs4-ga32-ep2-wu50-cut3000

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

namezz/
lvm-a-qwen2.5-7b-instruct-b-qwen2.5-1.5b-instruct