Automatic Speech Recognition
Transformers
English
ternary-quant
quantization
ternary
audio
speech-to-text
whisper
Instructions to use AsadIsmail/whisper-small-ternary with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use AsadIsmail/whisper-small-ternary with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("automatic-speech-recognition", model="AsadIsmail/whisper-small-ternary")# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("AsadIsmail/whisper-small-ternary", device_map="auto") - Notebooks
- Google Colab
- Kaggle
| { | |
| "method": "Broad-tritplane3", | |
| "model": "openai/whisper-small", | |
| "model_info": { | |
| "model_name": "openai/whisper-small", | |
| "model_family": "seq2seq_lm", | |
| "model_type": "whisper", | |
| "architectures": [ | |
| "WhisperForConditionalGeneration" | |
| ], | |
| "asset_kind": "tokenizer", | |
| "components": [ | |
| { | |
| "name": "encoder", | |
| "path": "model.encoder.layers", | |
| "kind": "encoder_stack", | |
| "module_type": "ModuleList", | |
| "parameter_count": 85045248, | |
| "linear_like_count": 72, | |
| "sample_linear_like_names": [ | |
| "0.self_attn.k_proj", | |
| "0.self_attn.v_proj", | |
| "0.self_attn.q_proj", | |
| "0.self_attn.out_proj", | |
| "0.fc1", | |
| "0.fc2", | |
| "1.self_attn.k_proj", | |
| "1.self_attn.v_proj" | |
| ] | |
| }, | |
| { | |
| "name": "decoder", | |
| "path": "model.decoder.layers", | |
| "kind": "decoder_stack", | |
| "module_type": "ModuleList", | |
| "parameter_count": 113402880, | |
| "linear_like_count": 120, | |
| "sample_linear_like_names": [ | |
| "0.self_attn.k_proj", | |
| "0.self_attn.v_proj", | |
| "0.self_attn.q_proj", | |
| "0.self_attn.out_proj", | |
| "0.encoder_attn.k_proj", | |
| "0.encoder_attn.v_proj", | |
| "0.encoder_attn.q_proj", | |
| "0.encoder_attn.out_proj" | |
| ] | |
| } | |
| ], | |
| "default_quantization_components": [ | |
| "decoder" | |
| ] | |
| }, | |
| "config": { | |
| "components": [ | |
| "decoder" | |
| ], | |
| "scheme": "tritplane3", | |
| "group_size": 32, | |
| "n_iter": 10, | |
| "salient_fraction": 0.0, | |
| "rescue_fraction": 0.0, | |
| "n_planes": 3, | |
| "allow_all_linear": false, | |
| "target_module_names": [ | |
| "Wqkv", | |
| "att_proj", | |
| "attn.proj", | |
| "attn.qkv", | |
| "c_attn", | |
| "c_fc", | |
| "c_proj", | |
| "dense", | |
| "dense_4h_to_h", | |
| "dense_h_to_4h", | |
| "down_proj", | |
| "fc1", | |
| "fc2", | |
| "ff_proj", | |
| "gate_proj", | |
| "gate_up_proj", | |
| "k", | |
| "k_proj", | |
| "linear", | |
| "o", | |
| "o_proj", | |
| "out_proj", | |
| "per_layer_input_gate", | |
| "per_layer_projection", | |
| "proj", | |
| "q", | |
| "q_proj", | |
| "qkv", | |
| "qkv_proj", | |
| "query_key_value", | |
| "up_proj", | |
| "v", | |
| "v_proj", | |
| "w1", | |
| "w2", | |
| "w3", | |
| "wi", | |
| "wi_0", | |
| "wi_1", | |
| "wo" | |
| ], | |
| "max_length": 160, | |
| "calibration_batch_size": 2, | |
| "calibration_prompts": null, | |
| "vlm_use_demo_image": true | |
| }, | |
| "summary": { | |
| "method_name": "Broad-tritplane3", | |
| "model_family": "seq2seq_lm", | |
| "selected_components": [ | |
| "decoder" | |
| ], | |
| "quantized_modules": 120, | |
| "quantized_params": 113246208, | |
| "quantized_fraction": 0.468472704513612, | |
| "avg_relative_error": 0.14943691343576185, | |
| "avg_effective_bits": 9.0, | |
| "full_model_effective_bits": 12.720691068404715, | |
| "compression_ratio": 1.2577933002193833 | |
| } | |
| } |