Falcon-40B
Parameters
40B
Context Length
2K
Modality
Text
Architecture
Dense
KV Cache Rate
15 KB/tok
License
Apache 2.0
Release Date
5 Jun 2023
Knowledge Cutoff
Feb 2023
API Pricing (per 1M)
Self-hosted only
Hardware Sizing
Precision
Context
Workload Memory: 21.16 GB
INT4 · 2048 context
Base Weights: 18.63 GB
KV Cache: 0.03 GB
Runtime & Graphs: 2.50 GB
Minimum Required Hardware
1x NVIDIA L4 24GB
21.2 / 21.6 GB usable
2x RTX 4090 24GB
21.2 / 43.2 GB usable · TP 2
Apple M3 Pro 36GB
21.2 / 27.0 GB usable
Determine requirements for your specific workload:
Open in VRAM Calculator
Architecture Diagram
Evaluation Benchmarks
No evaluation benchmarks for Falcon-40B available.
Rankings
Overall Rank
-
Coding Rank
-
About Falcon-40B
Falcon-40B is a foundational 40B parameter causal decoder model developed by TII and trained on 1T tokens from the RefinedWeb dataset. Featuring Multi-Query Attention and FlashAttention, it serves as a robust base for enterprise NLP applications.
Technical Specifications
Attention
Attention Structure
Multi-Query Attention
Attention Heads
64
Key-Value Heads
1
Attention Head Dimension
64
Position Embedding
ROPE
Normalization
Layer Normalization
Dimensions
Hidden Dimension Size
8,192
Number of Layers
60
Tokenizer
Vocabulary Size
65,024
About Falcon
The TII Falcon model family comprises causal decoder-only language models (7B, 40B). Their architecture, adapted from GPT-3, integrates rotary positional embeddings, Multi-Query Attention for inference efficiency, and FlashAttention for accelerated operations. Models are trained on the RefinedWeb dataset.