Modelpublic

RichardErkhov/danielgombas_-_llama_3b_step2_batch_v2-gguf

sourceHugging Faceupdated 2y agoView on Hugging Face

0likes999downloads

Model Card

Quantization made by Richard Erkhov.

Github

Discord

Request more models

llama3bstep2batchv2 - GGUF

—Model creator: https://huggingface.co/danielgombas/
—Original model: https://huggingface.co/danielgombas/llama3bstep2batchv2/

Name	Quant method	Size
llama_3b_step2_batch_v2.Q2_K.gguf	Q2_K	1.27GB
llama_3b_step2_batch_v2.IQ3_XS.gguf	IQ3_XS	1.38GB
llama_3b_step2_batch_v2.IQ3_S.gguf	IQ3_S	1.44GB
llama_3b_step2_batch_v2.Q3_K_S.gguf	Q3KS	1.44GB
llama_3b_step2_batch_v2.IQ3_M.gguf	IQ3_M	1.49GB
llama_3b_step2_batch_v2.Q3_K.gguf	Q3_K	1.57GB
llama_3b_step2_batch_v2.Q3_K_M.gguf	Q3KM	1.57GB
llama_3b_step2_batch_v2.Q3_K_L.gguf	Q3KL	1.69GB
llama_3b_step2_batch_v2.IQ4_XS.gguf	IQ4_XS	1.71GB
llama_3b_step2_batch_v2.Q4_0.gguf	Q4_0	1.79GB
llama_3b_step2_batch_v2.IQ4_NL.gguf	IQ4_NL	1.79GB
llama_3b_step2_batch_v2.Q4_K_S.gguf	Q4KS	1.8GB
llama_3b_step2_batch_v2.Q4_K.gguf	Q4_K	1.88GB
llama_3b_step2_batch_v2.Q4_K_M.gguf	Q4KM	1.88GB
llama_3b_step2_batch_v2.Q4_1.gguf	Q4_1	1.95GB
llama_3b_step2_batch_v2.Q5_0.gguf	Q5_0	2.11GB
llama_3b_step2_batch_v2.Q5_K_S.gguf	Q5KS	2.11GB
llama_3b_step2_batch_v2.Q5_K.gguf	Q5_K	2.16GB
llama_3b_step2_batch_v2.Q5_K_M.gguf	Q5KM	2.16GB
llama_3b_step2_batch_v2.Q5_1.gguf	Q5_1	2.28GB
llama_3b_step2_batch_v2.Q6_K.gguf	Q6_K	2.46GB
llama_3b_step2_batch_v2.Q8_0.gguf	Q8_0	3.19GB

Original model description: --- library_name: transformers tags:

—trl
—sft
—generatedfromtrainer model-index:
—name: llama3bstep2batchv2 results: [] ---

llama3bstep2batchv2

This model was trained from scratch on an unknown dataset. It achieves the following results on the evaluation set:

—Loss: 0.3132

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: 3e-05
—trainbatchsize: 2
—evalbatchsize: 40
—seed: 42
—gradientaccumulationsteps: 4
—totaltrainbatch_size: 8
—optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: linear
—num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss
0.993	0.0341	50	1.1011
1.0449	0.0682	100	0.9752
0.9894	0.1023	150	0.8698
0.6199	0.1364	200	0.7913
0.5326	0.1704	250	0.7341
0.8109	0.2045	300	0.6799
0.7554	0.2386	350	0.6332
0.9877	0.2727	400	0.5993
0.3571	0.3068	450	0.5726
0.4539	0.3409	500	0.5439
0.464	0.3750	550	0.5147
0.4051	0.4091	600	0.4904
0.5371	0.4432	650	0.4732
0.4954	0.4772	700	0.4549
0.4594	0.5113	750	0.4399
0.4755	0.5454	800	0.4281
0.2948	0.5795	850	0.4118
0.3699	0.6136	900	0.4021
0.319	0.6477	950	0.3927
0.3359	0.6818	1000	0.3802
0.4056	0.7159	1050	0.3746
0.2975	0.7500	1100	0.3643
0.3868	0.7840	1150	0.3596
0.3485	0.8181	1200	0.3512
0.3546	0.8522	1250	0.3476
0.3697	0.8863	1300	0.3416
0.4056	0.9204	1350	0.3388
0.3189	0.9545	1400	0.3332
0.4173	0.9886	1450	0.3286
0.1779	1.0228	1500	0.3338
0.2877	1.0569	1550	0.3300
0.1506	1.0910	1600	0.3301
0.2075	1.1251	1650	0.3289
0.1956	1.1592	1700	0.3271
0.162	1.1933	1750	0.3276
0.2416	1.2274	1800	0.3228
0.2364	1.2615	1850	0.3243
0.1602	1.2956	1900	0.3219
0.1566	1.3296	1950	0.3211
0.1784	1.3637	2000	0.3215
0.1627	1.3978	2050	0.3190
0.1907	1.4319	2100	0.3183
0.1182	1.4660	2150	0.3183
0.1585	1.5001	2200	0.3179
0.2261	1.5342	2250	0.3158
0.1457	1.5683	2300	0.3150
0.2589	1.6024	2350	0.3146
0.2253	1.6364	2400	0.3144
0.1741	1.6705	2450	0.3143
0.1477	1.7046	2500	0.3141
0.1886	1.7387	2550	0.3141
0.2211	1.7728	2600	0.3139
0.238	1.8069	2650	0.3138
0.2863	1.8410	2700	0.3137
0.2603	1.8751	2750	0.3135
0.2484	1.9092	2800	0.3133
0.2343	1.9432	2850	0.3132
0.254	1.9773	2900	0.3132

Framework versions

—Transformers 4.46.1
—Pytorch 2.1.0+cu118
—Datasets 3.0.2
—Tokenizers 0.20.1