← All Models | GPT - OpenAI's Generative Pre-trained Transformers:

openai/whisper-medium

📊 Model Parameters

Total Parameters 509,751,296
Context Length 2,048
Hidden Size 1024
Layers 24
Attention Heads 16
KV Heads 16

💾 Memory Requirements

FP32 (Full) 1.90 GB
FP16 (Half) 972.3 MB
INT8 (Quantized) 486.1 MB
INT4 (Quantized) 243.1 MB

🔑 KV Cache (Inference)

Per Token (FP16) 98.30 KB
Max Context FP32 384.0 MB
Max Context FP16 192.0 MB
Max Context INT8 96.0 MB

⚙️ Model Configuration

Core Architecture

Vocabulary Size51,865
Number of Layers24

Attention Configuration

Attention Dropout0%
Tied EmbeddingsYes

Activation & Normalization

Activation Functiongelu

Special Tokens

BOS Token ID50,257
Pad Token ID50,257
EOS Token ID50257

Data Type

Model Dtypefloat32
Layer Types:
Attention
MLP/FFN
Normalization
Embedding