whisper-base Magnetron Snapshot
This repository contains a Magnetron snapshot converted from the original Hugging Face model openai/whisper-base.
The snapshot is intended for inference with the Magnetron runtime. All convertible tensors are stored using bf16 where applicable.
Model details
- Source model:
openai/whisper-base - Snapshot file:
whisper-base-bf16.mag - Magnetron dtype mode:
bf16 - Tokenizer: embedded in the snapshot metadata
Snapshot
| Metric | Value |
|---|---|
| Source | openai/whisper-base |
| DType | bfloat16 |
| Tensors | 246 |
| Payload | 138.52 MiB |
| Alignment padding | 399.00 KiB (0.280%) |
| Data section | 138.91 MiB |
| Metadata | 2.72 MiB (incl. 2.37 MiB tokenizer) |
| Container overhead | 2.04 KiB |
| File size | 141.64 MiB |
| Source shards | 276.95 MiB (0.51x) |
| Elapsed | 0.1 s |
| Throughput | 972.93 MiB/s |
Whisper configuration
| Field | Value |
|---|---|
repo_id |
openai/whisper-base |
vocab_size |
51865 |
num_mel_bins |
80 |
d_model |
512 |
encoder_layers |
6 |
encoder_attention_heads |
8 |
encoder_ffn_dim |
2048 |
decoder_layers |
6 |
decoder_attention_heads |
8 |
decoder_ffn_dim |
2048 |
max_source_positions |
1500 |
max_target_positions |
448 |
Tensor manifest
| Name | Shape | DType |
|---|---|---|
decoder.embed_positions.weight |
448x512 |
bf16 |
decoder.embed_tokens.weight |
51865x512 |
bf16 |
decoder.layer_norm.bias |
512 |
bf16 |
decoder.layer_norm.weight |
512 |
bf16 |
decoder.layers.0.encoder_attn.k_proj.weight |
512x512 |
bf16 |
decoder.layers.0.encoder_attn.out_proj.bias |
512 |
bf16 |
decoder.layers.0.encoder_attn.out_proj.weight |
512x512 |
bf16 |
decoder.layers.0.encoder_attn.q_proj.bias |
512 |
bf16 |
decoder.layers.0.encoder_attn.q_proj.weight |
512x512 |
bf16 |
decoder.layers.0.encoder_attn.v_proj.bias |
512 |
bf16 |
decoder.layers.0.encoder_attn.v_proj.weight |
512x512 |
bf16 |
decoder.layers.0.encoder_attn_layer_norm.bias |
512 |
bf16 |
decoder.layers.0.encoder_attn_layer_norm.weight |
512 |
bf16 |
decoder.layers.0.fc1.bias |
2048 |
bf16 |
decoder.layers.0.fc1.weight |
2048x512 |
bf16 |
decoder.layers.0.fc2.bias |
512 |
bf16 |
decoder.layers.0.fc2.weight |
512x2048 |
bf16 |
decoder.layers.0.final_layer_norm.bias |
512 |
bf16 |
decoder.layers.0.final_layer_norm.weight |
512 |
bf16 |
decoder.layers.0.self_attn.k_proj.weight |
512x512 |
bf16 |
decoder.layers.0.self_attn.out_proj.bias |
512 |
bf16 |
decoder.layers.0.self_attn.out_proj.weight |
512x512 |
bf16 |
decoder.layers.0.self_attn.q_proj.bias |
512 |
bf16 |
decoder.layers.0.self_attn.q_proj.weight |
512x512 |
bf16 |
decoder.layers.0.self_attn.v_proj.bias |
512 |
bf16 |
decoder.layers.0.self_attn.v_proj.weight |
512x512 |
bf16 |
decoder.layers.0.self_attn_layer_norm.bias |
512 |
bf16 |
decoder.layers.0.self_attn_layer_norm.weight |
512 |
bf16 |
decoder.layers.1.encoder_attn.k_proj.weight |
512x512 |
bf16 |
decoder.layers.1.encoder_attn.out_proj.bias |
512 |
bf16 |
decoder.layers.1.encoder_attn.out_proj.weight |
512x512 |
bf16 |
decoder.layers.1.encoder_attn.q_proj.bias |
512 |
bf16 |
decoder.layers.1.encoder_attn.q_proj.weight |
512x512 |
bf16 |
decoder.layers.1.encoder_attn.v_proj.bias |
512 |
bf16 |
decoder.layers.1.encoder_attn.v_proj.weight |
512x512 |
bf16 |
decoder.layers.1.encoder_attn_layer_norm.bias |
512 |
bf16 |
decoder.layers.1.encoder_attn_layer_norm.weight |
512 |
bf16 |
decoder.layers.1.fc1.bias |
2048 |
bf16 |
decoder.layers.1.fc1.weight |
2048x512 |
bf16 |
decoder.layers.1.fc2.bias |
512 |
bf16 |
decoder.layers.1.fc2.weight |
512x2048 |
bf16 |
decoder.layers.1.final_layer_norm.bias |
512 |
bf16 |
decoder.layers.1.final_layer_norm.weight |
512 |
bf16 |
decoder.layers.1.self_attn.k_proj.weight |
512x512 |
bf16 |
decoder.layers.1.self_attn.out_proj.bias |
512 |
bf16 |
decoder.layers.1.self_attn.out_proj.weight |
512x512 |
bf16 |
decoder.layers.1.self_attn.q_proj.bias |
512 |
bf16 |
decoder.layers.1.self_attn.q_proj.weight |
512x512 |
bf16 |
decoder.layers.1.self_attn.v_proj.bias |
512 |
bf16 |
decoder.layers.1.self_attn.v_proj.weight |
512x512 |
bf16 |
decoder.layers.1.self_attn_layer_norm.bias |
512 |
bf16 |
decoder.layers.1.self_attn_layer_norm.weight |
512 |
bf16 |
decoder.layers.2.encoder_attn.k_proj.weight |
512x512 |
bf16 |
decoder.layers.2.encoder_attn.out_proj.bias |
512 |
bf16 |
decoder.layers.2.encoder_attn.out_proj.weight |
512x512 |
bf16 |
decoder.layers.2.encoder_attn.q_proj.bias |
512 |
bf16 |
decoder.layers.2.encoder_attn.q_proj.weight |
512x512 |
bf16 |
decoder.layers.2.encoder_attn.v_proj.bias |
512 |
bf16 |
decoder.layers.2.encoder_attn.v_proj.weight |
512x512 |
bf16 |
decoder.layers.2.encoder_attn_layer_norm.bias |
512 |
bf16 |
decoder.layers.2.encoder_attn_layer_norm.weight |
512 |
bf16 |
decoder.layers.2.fc1.bias |
2048 |
bf16 |
decoder.layers.2.fc1.weight |
2048x512 |
bf16 |
decoder.layers.2.fc2.bias |
512 |
bf16 |
decoder.layers.2.fc2.weight |
512x2048 |
bf16 |
decoder.layers.2.final_layer_norm.bias |
512 |
bf16 |
decoder.layers.2.final_layer_norm.weight |
512 |
bf16 |
decoder.layers.2.self_attn.k_proj.weight |
512x512 |
bf16 |
decoder.layers.2.self_attn.out_proj.bias |
512 |
bf16 |
decoder.layers.2.self_attn.out_proj.weight |
512x512 |
bf16 |
decoder.layers.2.self_attn.q_proj.bias |
512 |
bf16 |
decoder.layers.2.self_attn.q_proj.weight |
512x512 |
bf16 |
decoder.layers.2.self_attn.v_proj.bias |
512 |
bf16 |
decoder.layers.2.self_attn.v_proj.weight |
512x512 |
bf16 |
decoder.layers.2.self_attn_layer_norm.bias |
512 |
bf16 |
decoder.layers.2.self_attn_layer_norm.weight |
512 |
bf16 |
decoder.layers.3.encoder_attn.k_proj.weight |
512x512 |
bf16 |
decoder.layers.3.encoder_attn.out_proj.bias |
512 |
bf16 |
decoder.layers.3.encoder_attn.out_proj.weight |
512x512 |
bf16 |
decoder.layers.3.encoder_attn.q_proj.bias |
512 |
bf16 |
decoder.layers.3.encoder_attn.q_proj.weight |
512x512 |
bf16 |
decoder.layers.3.encoder_attn.v_proj.bias |
512 |
bf16 |
decoder.layers.3.encoder_attn.v_proj.weight |
512x512 |
bf16 |
decoder.layers.3.encoder_attn_layer_norm.bias |
512 |
bf16 |
decoder.layers.3.encoder_attn_layer_norm.weight |
512 |
bf16 |
decoder.layers.3.fc1.bias |
2048 |
bf16 |
decoder.layers.3.fc1.weight |
2048x512 |
bf16 |
decoder.layers.3.fc2.bias |
512 |
bf16 |
decoder.layers.3.fc2.weight |
512x2048 |
bf16 |
decoder.layers.3.final_layer_norm.bias |
512 |
bf16 |
decoder.layers.3.final_layer_norm.weight |
512 |
bf16 |
decoder.layers.3.self_attn.k_proj.weight |
512x512 |
bf16 |
decoder.layers.3.self_attn.out_proj.bias |
512 |
bf16 |
decoder.layers.3.self_attn.out_proj.weight |
512x512 |
bf16 |
decoder.layers.3.self_attn.q_proj.bias |
512 |
bf16 |
decoder.layers.3.self_attn.q_proj.weight |
512x512 |
bf16 |
decoder.layers.3.self_attn.v_proj.bias |
512 |
bf16 |
decoder.layers.3.self_attn.v_proj.weight |
512x512 |
bf16 |
decoder.layers.3.self_attn_layer_norm.bias |
512 |
bf16 |
decoder.layers.3.self_attn_layer_norm.weight |
512 |
bf16 |
decoder.layers.4.encoder_attn.k_proj.weight |
512x512 |
bf16 |
decoder.layers.4.encoder_attn.out_proj.bias |
512 |
bf16 |
decoder.layers.4.encoder_attn.out_proj.weight |
512x512 |
bf16 |
decoder.layers.4.encoder_attn.q_proj.bias |
512 |
bf16 |
decoder.layers.4.encoder_attn.q_proj.weight |
512x512 |
bf16 |
decoder.layers.4.encoder_attn.v_proj.bias |
512 |
bf16 |
decoder.layers.4.encoder_attn.v_proj.weight |
512x512 |
bf16 |
decoder.layers.4.encoder_attn_layer_norm.bias |
512 |
bf16 |
decoder.layers.4.encoder_attn_layer_norm.weight |
512 |
bf16 |
decoder.layers.4.fc1.bias |
2048 |
bf16 |
decoder.layers.4.fc1.weight |
2048x512 |
bf16 |
decoder.layers.4.fc2.bias |
512 |
bf16 |
decoder.layers.4.fc2.weight |
512x2048 |
bf16 |
decoder.layers.4.final_layer_norm.bias |
512 |
bf16 |
decoder.layers.4.final_layer_norm.weight |
512 |
bf16 |
decoder.layers.4.self_attn.k_proj.weight |
512x512 |
bf16 |
decoder.layers.4.self_attn.out_proj.bias |
512 |
bf16 |
decoder.layers.4.self_attn.out_proj.weight |
512x512 |
bf16 |
decoder.layers.4.self_attn.q_proj.bias |
512 |
bf16 |
decoder.layers.4.self_attn.q_proj.weight |
512x512 |
bf16 |
decoder.layers.4.self_attn.v_proj.bias |
512 |
bf16 |
decoder.layers.4.self_attn.v_proj.weight |
512x512 |
bf16 |
decoder.layers.4.self_attn_layer_norm.bias |
512 |
bf16 |
decoder.layers.4.self_attn_layer_norm.weight |
512 |
bf16 |
decoder.layers.5.encoder_attn.k_proj.weight |
512x512 |
bf16 |
decoder.layers.5.encoder_attn.out_proj.bias |
512 |
bf16 |
decoder.layers.5.encoder_attn.out_proj.weight |
512x512 |
bf16 |
decoder.layers.5.encoder_attn.q_proj.bias |
512 |
bf16 |
decoder.layers.5.encoder_attn.q_proj.weight |
512x512 |
bf16 |
decoder.layers.5.encoder_attn.v_proj.bias |
512 |
bf16 |
decoder.layers.5.encoder_attn.v_proj.weight |
512x512 |
bf16 |
decoder.layers.5.encoder_attn_layer_norm.bias |
512 |
bf16 |
decoder.layers.5.encoder_attn_layer_norm.weight |
512 |
bf16 |
decoder.layers.5.fc1.bias |
2048 |
bf16 |
decoder.layers.5.fc1.weight |
2048x512 |
bf16 |
decoder.layers.5.fc2.bias |
512 |
bf16 |
decoder.layers.5.fc2.weight |
512x2048 |
bf16 |
decoder.layers.5.final_layer_norm.bias |
512 |
bf16 |
decoder.layers.5.final_layer_norm.weight |
512 |
bf16 |
decoder.layers.5.self_attn.k_proj.weight |
512x512 |
bf16 |
decoder.layers.5.self_attn.out_proj.bias |
512 |
bf16 |
decoder.layers.5.self_attn.out_proj.weight |
512x512 |
bf16 |
decoder.layers.5.self_attn.q_proj.bias |
512 |
bf16 |
decoder.layers.5.self_attn.q_proj.weight |
512x512 |
bf16 |
decoder.layers.5.self_attn.v_proj.bias |
512 |
bf16 |
decoder.layers.5.self_attn.v_proj.weight |
512x512 |
bf16 |
decoder.layers.5.self_attn_layer_norm.bias |
512 |
bf16 |
decoder.layers.5.self_attn_layer_norm.weight |
512 |
bf16 |
encoder.conv1.bias |
512 |
bf16 |
encoder.conv1.weight |
512x80x3 |
bf16 |
encoder.conv2.bias |
512 |
bf16 |
encoder.conv2.weight |
512x512x3 |
bf16 |
encoder.embed_positions.weight |
1500x512 |
bf16 |
encoder.layer_norm.bias |
512 |
bf16 |
encoder.layer_norm.weight |
512 |
bf16 |
encoder.layers.0.fc1.bias |
2048 |
bf16 |
encoder.layers.0.fc1.weight |
2048x512 |
bf16 |
encoder.layers.0.fc2.bias |
512 |
bf16 |
encoder.layers.0.fc2.weight |
512x2048 |
bf16 |
encoder.layers.0.final_layer_norm.bias |
512 |
bf16 |
encoder.layers.0.final_layer_norm.weight |
512 |
bf16 |
encoder.layers.0.self_attn.k_proj.weight |
512x512 |
bf16 |
encoder.layers.0.self_attn.out_proj.bias |
512 |
bf16 |
encoder.layers.0.self_attn.out_proj.weight |
512x512 |
bf16 |
encoder.layers.0.self_attn.q_proj.bias |
512 |
bf16 |
encoder.layers.0.self_attn.q_proj.weight |
512x512 |
bf16 |
encoder.layers.0.self_attn.v_proj.bias |
512 |
bf16 |
encoder.layers.0.self_attn.v_proj.weight |
512x512 |
bf16 |
encoder.layers.0.self_attn_layer_norm.bias |
512 |
bf16 |
encoder.layers.0.self_attn_layer_norm.weight |
512 |
bf16 |
encoder.layers.1.fc1.bias |
2048 |
bf16 |
encoder.layers.1.fc1.weight |
2048x512 |
bf16 |
encoder.layers.1.fc2.bias |
512 |
bf16 |
encoder.layers.1.fc2.weight |
512x2048 |
bf16 |
encoder.layers.1.final_layer_norm.bias |
512 |
bf16 |
encoder.layers.1.final_layer_norm.weight |
512 |
bf16 |
encoder.layers.1.self_attn.k_proj.weight |
512x512 |
bf16 |
encoder.layers.1.self_attn.out_proj.bias |
512 |
bf16 |
encoder.layers.1.self_attn.out_proj.weight |
512x512 |
bf16 |
encoder.layers.1.self_attn.q_proj.bias |
512 |
bf16 |
encoder.layers.1.self_attn.q_proj.weight |
512x512 |
bf16 |
encoder.layers.1.self_attn.v_proj.bias |
512 |
bf16 |
encoder.layers.1.self_attn.v_proj.weight |
512x512 |
bf16 |
encoder.layers.1.self_attn_layer_norm.bias |
512 |
bf16 |
encoder.layers.1.self_attn_layer_norm.weight |
512 |
bf16 |
encoder.layers.2.fc1.bias |
2048 |
bf16 |
encoder.layers.2.fc1.weight |
2048x512 |
bf16 |
encoder.layers.2.fc2.bias |
512 |
bf16 |
encoder.layers.2.fc2.weight |
512x2048 |
bf16 |
encoder.layers.2.final_layer_norm.bias |
512 |
bf16 |
encoder.layers.2.final_layer_norm.weight |
512 |
bf16 |
encoder.layers.2.self_attn.k_proj.weight |
512x512 |
bf16 |
encoder.layers.2.self_attn.out_proj.bias |
512 |
bf16 |
encoder.layers.2.self_attn.out_proj.weight |
512x512 |
bf16 |
encoder.layers.2.self_attn.q_proj.bias |
512 |
bf16 |
encoder.layers.2.self_attn.q_proj.weight |
512x512 |
bf16 |
encoder.layers.2.self_attn.v_proj.bias |
512 |
bf16 |
encoder.layers.2.self_attn.v_proj.weight |
512x512 |
bf16 |
encoder.layers.2.self_attn_layer_norm.bias |
512 |
bf16 |
encoder.layers.2.self_attn_layer_norm.weight |
512 |
bf16 |
encoder.layers.3.fc1.bias |
2048 |
bf16 |
encoder.layers.3.fc1.weight |
2048x512 |
bf16 |
encoder.layers.3.fc2.bias |
512 |
bf16 |
encoder.layers.3.fc2.weight |
512x2048 |
bf16 |
encoder.layers.3.final_layer_norm.bias |
512 |
bf16 |
encoder.layers.3.final_layer_norm.weight |
512 |
bf16 |
encoder.layers.3.self_attn.k_proj.weight |
512x512 |
bf16 |
encoder.layers.3.self_attn.out_proj.bias |
512 |
bf16 |
encoder.layers.3.self_attn.out_proj.weight |
512x512 |
bf16 |
encoder.layers.3.self_attn.q_proj.bias |
512 |
bf16 |
encoder.layers.3.self_attn.q_proj.weight |
512x512 |
bf16 |
encoder.layers.3.self_attn.v_proj.bias |
512 |
bf16 |
encoder.layers.3.self_attn.v_proj.weight |
512x512 |
bf16 |
encoder.layers.3.self_attn_layer_norm.bias |
512 |
bf16 |
encoder.layers.3.self_attn_layer_norm.weight |
512 |
bf16 |
encoder.layers.4.fc1.bias |
2048 |
bf16 |
encoder.layers.4.fc1.weight |
2048x512 |
bf16 |
encoder.layers.4.fc2.bias |
512 |
bf16 |
encoder.layers.4.fc2.weight |
512x2048 |
bf16 |
encoder.layers.4.final_layer_norm.bias |
512 |
bf16 |
encoder.layers.4.final_layer_norm.weight |
512 |
bf16 |
encoder.layers.4.self_attn.k_proj.weight |
512x512 |
bf16 |
encoder.layers.4.self_attn.out_proj.bias |
512 |
bf16 |
encoder.layers.4.self_attn.out_proj.weight |
512x512 |
bf16 |
encoder.layers.4.self_attn.q_proj.bias |
512 |
bf16 |
encoder.layers.4.self_attn.q_proj.weight |
512x512 |
bf16 |
encoder.layers.4.self_attn.v_proj.bias |
512 |
bf16 |
encoder.layers.4.self_attn.v_proj.weight |
512x512 |
bf16 |
encoder.layers.4.self_attn_layer_norm.bias |
512 |
bf16 |
encoder.layers.4.self_attn_layer_norm.weight |
512 |
bf16 |
encoder.layers.5.fc1.bias |
2048 |
bf16 |
encoder.layers.5.fc1.weight |
2048x512 |
bf16 |
encoder.layers.5.fc2.bias |
512 |
bf16 |
encoder.layers.5.fc2.weight |
512x2048 |
bf16 |
encoder.layers.5.final_layer_norm.bias |
512 |
bf16 |
encoder.layers.5.final_layer_norm.weight |
512 |
bf16 |
encoder.layers.5.self_attn.k_proj.weight |
512x512 |
bf16 |
encoder.layers.5.self_attn.out_proj.bias |
512 |
bf16 |
encoder.layers.5.self_attn.out_proj.weight |
512x512 |
bf16 |
encoder.layers.5.self_attn.q_proj.bias |
512 |
bf16 |
encoder.layers.5.self_attn.q_proj.weight |
512x512 |
bf16 |
encoder.layers.5.self_attn.v_proj.bias |
512 |
bf16 |
encoder.layers.5.self_attn.v_proj.weight |
512x512 |
bf16 |
encoder.layers.5.self_attn_layer_norm.bias |
512 |
bf16 |
encoder.layers.5.self_attn_layer_norm.weight |
512 |
bf16 |
mel_filters |
80x201 |
f32 |
Model tree for mario-sieg/magnetron-whisper
Base model
openai/whisper-base