Hugging Face's logo

waxal-benchmarking
/
omniasr-llm-300m-waxal-ach

Automatic Speech Recognition
Acoli
african-languages
waxal
omnilingual-asr
fairseq2
Model card Files
xet
 Community
Professor commited on
Commit
1b10ff7
·
verified ·
1 Parent(s): c0d4b14

Upload model.yaml with huggingface_hub

Browse files
Files changed (1) hide show
  1. model.yaml +113 -0
model.yaml ADDED
@@ -0,0 +1,113 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: checkpoint
2
+ model_family: wav2vec2_llama
3
+ model_config:
4
+ wav2vec2_asr_config:
5
+ encoder_config:
6
+ model_dim: 1024
7
+ max_seq_len: 4096
8
+ feature_dim: 512
9
+ use_fbank: false
10
+ first_pass_dropout_p: 0.0
11
+ layer_norm_features: false
12
+ feature_extractor_layer_descs:
13
+ - - 512
14
+ - 10
15
+ - 5
16
+ - - 512
17
+ - 3
18
+ - 2
19
+ - - 512
20
+ - 3
21
+ - 2
22
+ - - 512
23
+ - 3
24
+ - 2
25
+ - - 512
26
+ - 3
27
+ - 2
28
+ - - 512
29
+ - 2
30
+ - 2
31
+ - - 512
32
+ - 2
33
+ - 2
34
+ feature_extractor_bias: true
35
+ feature_extractor_layer_norm_convs: true
36
+ feature_grad_scale: 0.1
37
+ num_fbank_channels: 0
38
+ fbank_stride: 0
39
+ sample_fbank_every_k: 0
40
+ pos_encoder_type: conv
41
+ pos_encoder_depth: 1
42
+ pos_conv_kernel_size: 128
43
+ num_pos_conv_groups: 16
44
+ use_conformer: false
45
+ num_encoder_layers: 24
46
+ num_encoder_attn_heads: 16
47
+ ffn_inner_dim: 4096
48
+ dropout_p: 0.0
49
+ attn_dropout_p: 0.0
50
+ ffn_inner_dropout_p: 0.1
51
+ layer_drop_p: 0.1
52
+ norm_order: PRE
53
+ depthwise_conv_kernel_size: 0
54
+ target_vocab_size: 9812
55
+ final_dropout_p: 0.0
56
+ use_masking: false
57
+ temporal_mask_span_len: 10
58
+ max_temporal_mask_prob: 0.0
59
+ min_num_temporal_mask_spans: 2
60
+ spatial_mask_span_len: 64
61
+ max_spatial_mask_prob: 0.0
62
+ min_num_spatial_mask_spans: 2
63
+ llama_config:
64
+ model_dim: 4096
65
+ max_seq_len: 8192
66
+ vocab_size: 9812
67
+ pad_idx: 1
68
+ tied_embeddings: false
69
+ num_layers: 12
70
+ num_attn_heads: 8
71
+ num_key_value_heads: 8
72
+ ffn_inner_dim: 4096
73
+ ffn_inner_dim_scale: 0.6666666666666666
74
+ ffn_inner_dim_multiplier: 1.0
75
+ ffn_inner_dim_multiple_of: 256
76
+ rope_theta: 10000.0
77
+ use_scaled_rope: false
78
+ rope_scale:
79
+ factor: 8.0
80
+ frequency_factors:
81
+ - 1.0
82
+ - 4.0
83
+ original_context_length: 8192
84
+ dropout_p: 0.1
85
+ init_std: null
86
+ init_std_scale: layer
87
+ shard_embed_dim: true
88
+ beam_search_config:
89
+ nbest: 5
90
+ length_norm: false
91
+ compression_window: 100
92
+ compression_threshold: 4.0
93
+ streaming_config:
94
+ is_streaming: false
95
+ segment_secs: 15.0
96
+ sample_rate: 16000
97
+ n_context_segments: 1
98
+ text_tokenizer: ''
99
+ min_audio_ms: 25
100
+ encoder_stacking: 1
101
+ frozen_encoder: 1
102
+ lang_embeddings_p: 0.5
103
+ language_column_name: lang
104
+ context_text_only: false
105
+ n_special_tokens: 1
106
+ unk_idx: 3
107
+ bos_idx: 0
108
+ eos_idx: 2
109
+ pad_idx: 1
110
+ boh_idx: null
111
+ eoh_idx: null
112
+ model_type: LLM_ASR_LID
113
+ n_context_examples: 0