trl-internal-testing/tiny-VoxtralForConditionalGeneration
079
1{2 "architectures": [3 "VoxtralForConditionalGeneration"4 ],5 "audio_config": {6 "activation_dropout": 0.0,7 "activation_function": "gelu",8 "attention_dropout": 0.0,9 "dropout": 0.0,10 "head_dim": 64,11 "hidden_size": 16,12 "initializer_range": 0.02,13 "intermediate_size": 64,14 "layerdrop": 0.0,15 "max_source_positions": 1500,16 "model_type": "voxtral_encoder",17 "num_attention_heads": 4,18 "num_hidden_layers": 2,19 "num_key_value_heads": 20,20 "num_mel_bins": 128,21 "scale_embedding": false,22 "vocab_size": 5186623 },24 "audio_token_id": 24,25 "dtype": "bfloat16",26 "hidden_size": 3072,27 "model_type": "voxtral",28 "projector_hidden_act": "gelu",29 "text_config": {30 "attention_bias": false,31 "attention_dropout": 0.0,32 "head_dim": 4,33 "hidden_act": "silu",34 "hidden_size": 16,35 "initializer_range": 0.02,36 "intermediate_size": 32,37 "layer_types": null,38 "max_position_embeddings": 131072,39 "mlp_bias": false,40 "model_type": "llama",41 "num_attention_heads": 4,42 "num_hidden_layers": 2,43 "num_key_value_heads": 2,44 "pretraining_tp": 1,45 "rms_norm_eps": 1e-05,46 "rope_scaling": null,47 "rope_theta": 100000000.0,48 "sliding_window": null,49 "use_cache": true,50 "vocab_size": 13107251 },52 "transformers_version": "4.56.2",53 "vocab_size": 13107254}55 