← All Models
|
Mistral - French AI startup known for efficient open models (founded 2023):
Mistral-7B-v0.1
Mistral-Nemo-Base-2407
Codestral-22B-v0.1
Mistral-Small-24B-Base-2501
Ministral-8B-Instruct-2410
Mistral-Large-Instruct-2411
Mixtral-8x7B-v0.1
Mixtral-8x22B-v0.1
Mamba-Codestral-7B-v0.1
Pixtral-12B-Base-2409
Voxtral-Mini-3B-2507
Voxtral-Small-24B-2507
Mistral-Small-3.2-24B-Instruct-2506
Ministral-3-3B-Base-2512
Ministral-3-8B-Base-2512
Ministral-3-14B-Base-2512
Mistral-Large-3-675B-Base-2512
Devstral-Small-2-24B-Instruct-2512
Devstral-2-123B-Instruct-2512
mistralai/Codestral-22B-v0.1
📊 Model Parameters
Total Parameters
22,247,282,688
Context Length
32,768
Hidden Size
6144
Layers
56
Attention Heads
48
KV Heads
8
💾 Memory Requirements
FP32 (Full)
82.88 GB
FP16 (Half)
41.44 GB
INT8 (Quantized)
20.72 GB
INT4 (Quantized)
10.36 GB
🔑 KV Cache (Inference)
Per Token (FP16)
229.38 KB
Max Context FP32
14.00 GB
Max Context FP16
7.00 GB
Max Context INT8
3.50 GB
⚙️ Model Configuration
Core Architecture
Vocabulary Size
32,768
Hidden Size
6,144
FFN Intermediate Size
16,384
Number of Layers
56
Attention Heads
48
Head Dimension
128
KV Heads
8
Context & Position
Max Context Length
32,768
Sliding Window Size
Not set
Attention Configuration
Attention Dropout
0%
Tied Embeddings
No
Activation & Normalization
Activation Function
silu
RMSNorm Epsilon
1e-05
Special Tokens
Pad Token ID
Not set
BOS Token ID
1
EOS Token ID
2
Data Type
Model Dtype
bfloat16
Layer Types:
Attention
MLP/FFN
Normalization
Embedding
Attention
MLP
Norm
Embedding
Clear
Expand All
Collapse All