Model reference · open weights

model-960hfacebook-2022.06.08

Name: model-960hfacebook-2022.06.08
Brand: AxForge

Available as managed deployment Audio Vkt · community Speech→text 1 variants 692 dl/mo

model-960hfacebook-2022.06.08 is an open-weight audio or speech model from Vkt. 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	Vkt
Type	Audio & music
Task	Speech→text
Runs with	transformers
Released	2022-06-08
Popularity	692 downloads / month
Licence	Open weights

About

What model-960hfacebook-2022.06.08 is

This model is a fine-tuned version of facebook/wav2vec2-large-960h-lv60-self on the common_voice dataset. It achieves the following results on the evaluation set:

Loss: 0.2907
Wer: 0.1804

Read the full model card

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: 0.0003
train_batch_size: 8
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 16
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
num_epochs: 25
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.7634	0.21	300	2.9743	0.9998
1.6536	0.43	600	0.8605	0.7529
0.9823	0.64	900	0.6600	0.6286
0.8708	0.86	1200	0.5780	0.5736
0.7878	1.07	1500	0.5386	0.5326
0.7033	1.29	1800	0.4986	0.4992
0.681	1.5	2100	0.4575	0.4778
0.6537	1.72	2400	0.4591	0.4482
0.6263	1.93	2700	0.4317	0.4353
0.5811	2.14	3000	0.4149	0.4159
0.5565	2.36	3300	0.4170	0.3956
0.5501	2.57	3600	0.4007	0.3929
0.5444	2.79	3900	0.3930	0.3851
0.5177	3.0	4200	0.4006	0.3630
0.4682	3.22	4500	0.3707	0.3713
0.4805	3.43	4800	0.3564	0.3583
0.4715	3.65	5100	0.3596	0.3434
0.4482	3.86	5400	0.3555	0.3394
0.4407	4.07	5700	0.3680	0.3312
0.4134	4.29	6000	0.3534	0.3328
0.4165	4.5	6300	0.3294	0.3259
0.4196	4.72	6600	0.3353	0.3214
0.4117	4.93	6900	0.3266	0.3211
0.3847	5.15	7200	0.3365	0.3156
0.3687	5.36	7500	0.3233	0.3014
0.376	5.58	7800	0.3345	0.2979
0.3732	5.79	8100	0.3105	0.2882
0.3705	6.0	8400	0.3252	0.2935
0.3311	6.22	8700	0.3266	0.2911
0.3386	6.43	9000	0.2975	0.2765
0.337	6.65	9300	0.3070	0.2826
0.3458	6.86	9600	0.3090	0.2766
0.3218	7.08	9900	0.3117	0.2748
0.3041	7.29	10200	0.2989	0.2651
0.3031	7.51	10500	0.3210	0.2672
0.3037	7.72	10800	0.3040	0.2667
0.3126	7.93	11100	0.2867	0.2613
0.3005	8.15	11400	0.3075	0.2610
0.2802	8.36	11700	0.3129	0.2608
0.2785	8.58	12000	0.3002	0.2579
0.2788	8.79	12300	0.3063	0.2476
0.286	9.01	12600	0.2971	0.2495
0.2534	9.22	12900	0.2766	0.2452
0.2542	9.44	13200	0.2893	0.2405
0.2576	9.65	13500	0.3038	0.2518
0.2552	9.86	13800	0.2851	0.2429
0.2487	10.08	14100	0.2858	0.2356
0.2441	10.29	14400	0.2999	0.2364
0.2345	10.51	14700	0.2907	0.2373
0.2352	10.72	15000	0.2885	0.2402
0.2464	10.94	15300	0.2896	0.2339
0.2219	11.15	15600	0.2999	0.2351
0.2257	11.37	15900	0.2930	0.2326
0.2184	11.58	16200	0.2980	0.2353
0.2182	11.79	16500	0.2832	0.2296
0.2224	12.01	16800	0.2797	0.2285
0.1991	12.22	17100	0.2810	0.2296
0.1993	12.44	17400	0.2949	0.2253
0.2042	12.65	17700	0.2864	0.2207
0.2083	12.87	18000	0.2860	0.2278
0.1998	13.08	18300	0.2872	0.2232
0.1919	13.3	18600	0.2894	0.2247
0.1925	13.51	18900	0.3007	0.2234
0.1966	13.72	19200	0.2831	0.2176
0.1942	13.94	19500	0.2811	0.2161
0.1778	14.15	19800	0.2901	0.2196
0.1755	14.37	20100	0.2864	0.2188
0.1795	14.58	20400	0.2927	0.2170
0.1817	14.8	20700	0.2846	0.2156
0.1754	15.01	21000	0.3036	0.2137
0.1674	15.23	21300	0.2876	0.2156
0.171	15.44	21600	0.2812	0.2106
0.1603	15.65	21900	0.2692	0.2093
0.1663	15.87	22200	0.2745	0.2094
0.1608	16.08	22500	0.2807	0.2043
0.1555	16.3	22800	0.2872	0.2036
0.1546	16.51	23100	0.2837	0.2049
0.1515	16.73	23400	0.2746	0.2031
0.1571	16.94	23700	0.2767	0.2047
0.1498	17.16	24000	0.2837	0.2050
0.143	17.37	24300	0.2745	0.2038
0.1471	17.58	24600	0.2787	0.2004
0.144

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 model-960hfacebook-2022-06-08 for you, it answers on the OpenAI-compatible API — the same base URL and keys as every other model. (model-960hfacebook-2022-06-08 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="model-960hfacebook-2022-06-08" -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

model-960hfacebook-2022.06.08

— · BF16

Weights ↗

Details

Trained / evaluated on

common_voice