Model reference · open weights

xls-r-npsc-4

Name: xls-r-npsc-4
Brand: AxForge

Available as managed deployment Audio Rolv-Arild · community Speech→text 1 variants 552 dl/mo

xls-r-npsc-4 is an open-weight audio or speech model from Rolv-Arild. AxForge deploys and operates it for you on dedicated EU-owned hardware — with the licence handled where one is required.

Request deployment All audio & music

Available as managed deployment — configured and operated for you on dedicated EU hardware, quoted per deployment.

What it is

Released by	Rolv-Arild
Type	Audio & music
Task	Speech→text
Runs with	transformers
Released	2022-03-02
Popularity	552 downloads / month
Licence	Open weights

About

What xls-r-npsc-4 is

It achieves the following results on the evaluation set:

Loss: 0.1957
Wer: 0.1697

Model description

More information needed

Intended uses & limitations

More information needed

Read the full model card

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 7.5e-05
train_batch_size: 16
eval_batch_size: 16
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 64
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 2000
num_epochs: 20.0
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.4527	0.28	250	4.0144	1.0
3.1828	0.56	500	3.1369	1.0
2.9927	0.85	750	3.0183	1.0
2.9591	1.13	1000	2.9991	1.0
2.8989	1.41	1250	2.9000	1.0000
2.4286	1.69	1500	1.7688	0.9550
1.6765	1.98	1750	0.6842	0.4855
1.4521	2.26	2000	0.5096	0.3736
1.3589	2.54	2250	0.4479	0.3335
1.3136	2.82	2500	0.4056	0.3123
1.2856	3.11	2750	0.3870	0.2987
1.2283	3.39	3000	0.3646	0.2828
1.2053	3.67	3250	0.3499	0.2748
1.2087	3.95	3500	0.3345	0.2603
1.2002	4.24	3750	0.3320	0.2523
1.1383	4.52	4000	0.3117	0.2439
1.1364	4.8	4250	0.3198	0.2383
1.158	5.08	4500	0.3071	0.2342
1.108	5.37	4750	0.3011	0.2314
1.1025	5.65	5000	0.2875	0.2289
1.0697	5.93	5250	0.2926	0.2256
1.0904	6.21	5500	0.2695	0.2245
1.0802	6.5	5750	0.2602	0.2189
1.0882	6.78	6000	0.2603	0.2168
1.0881	7.06	6250	0.2540	0.2293
1.0378	7.34	6500	0.2614	0.2193
1.0397	7.63	6750	0.2707	0.2104
1.0296	7.91	7000	0.2483	0.2119
1.0249	8.19	7250	0.2483	0.2047
1.013	8.47	7500	0.2487	0.2042
1.0064	8.76	7750	0.2456	0.2016
1.0668	9.04	8000	0.2397	0.1995
1.0129	9.32	8250	0.2374	0.1994
1.0164	9.6	8500	0.2206	0.1992
0.975	9.89	8750	0.2247	0.1973
0.9849	10.17	9000	0.2325	0.1953
0.9826	10.45	9250	0.2301	0.1934
0.9835	10.73	9500	0.2192	0.1942
0.9676	11.02	9750	0.2266	0.1913
0.9627	11.3	10000	0.2193	0.1921
0.976	11.58	10250	0.2309	0.1882
0.969	11.86	10500	0.2268	0.1886
0.9611	12.15	10750	0.2322	0.1863
0.9397	12.43	11000	0.2197	0.1844
0.9601	12.71	11250	0.2211	0.1871
0.9718	12.99	11500	0.2079	0.1898
0.9347	13.28	11750	0.2054	0.1843
0.9377	13.56	12000	0.2031	0.1842
0.934	13.84	12250	0.2059	0.1806
0.9295	14.12	12500	0.2122	0.1861
0.935	14.41	12750	0.2072	0.1787
0.9021	14.69	13000	0.2105	0.1781
0.9193	14.97	13250	0.2035	0.1786
0.9214	15.25	13500	0.2035	0.1766
0.9048	15.54	13750	0.1964	0.1758
0.9006	15.82	14000	0.1984	0.1757
0.9027	16.1	14250	0.2022	0.1743
0.9083	16.38	14500	0.1969	0.1744
0.9761	16.67	14750	0.1963	0.1728
0.9311	16.95	15000	0.1960	0.1737
0.886	17.23	15250	0.1929	0.1726
0.8969	17.51	15500	0.1928	0.1734
0.9084	17.8	15750	0.1937	0.1713
0.8795	18.08	16000	0.1978	0.1709
0.8883	18.36	16250	0.1956	0.1703
0.8901	18.64	16500	0.1933	0.1705
0.8922	18.93	16750	0.1962	0.1711
0.8765	19.21	17000	0.1962	0.1711
0.8992	19.49	17250	0.1965	0.1703
0.8778	19.77	17500	0.1957	0.1699

Framework versions

Transformers 4.17.0.dev0
Pytorch 1.10.0+cu113
Datasets 1.18.1
Tokenizers 0.11.0

From the published model card. Full card on the HuggingFace links in the sidebar.

Using it via the API

Call it like any OpenAI endpoint

Once AxForge deploys xls-r-npsc-4 for you, it answers on the OpenAI-compatible API — the same base URL and keys as every other model. (xls-r-npsc-4 below is illustrative; you get the exact model name on deployment.)

$ curl -sS https://api.axforge.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $AXFORGE_API_KEY" \
  -F model="xls-r-npsc-4" -F file=@audio.mp3

Create an account — your API key is available in the console. 3M free tokens every 30 days with every new account.

Get API access

Sizes, precisions & builds

xls-r-300m-npsc-4

— · BF16

Weights ↗

Details