Large Language Model {LLM} model size, parameter size and parameter count
The model size is the number of parameters in the Large Language Model (LLM).
An LLM has 70B parameters, it means that the model has 70 billion adjustable parameters.
1. Model Size and Memory Usage
The total memory usage of the model:
memory usage (in bytes) = number of parameters
×
\times
× size of each parameter
Number of parameters | 1 B |
---|---|
32-bit precision | 1 × \times × 1,000,000,000 × \times × 4 byte = 4,000,000,000 / 1024 / 1024 / 1024 = 3.7253 GByte |
16-bit precision | 1 × \times × 1,000,000,000 × \times × 2 byte = 2,000,000,000 / 1024 / 1024 / 1024 = 1.862645 GByte |
8-bit precision | 1 × \times × 1,000,000,000 × \times × 1 byte = 1,000,000,000 / 1024 / 1024 / 1024 = 0.93132 GByte |
Number of parameters | 3 B |
---|---|
32-bit precision | 3 × \times × 1,000,000,000 × \times × 4 byte = 12,000,000,000 / 1024 / 1024 / 1024 = 11.17587 GByte |
16-bit precision | 3 × \times × 1,000,000,000 × \times × 2 byte = 6,000,000,000 / 1024 / 1024 / 1024 = 5.58794 GByte |
8-bit precision | 3 × \times × 1,000,000,000 × \times × 1 byte = 3,000,000,000 / 1024 / 1024 / 1024 = 2.7939677 GByte |
Number of parameters | 7 B |
---|---|
32-bit precision | 7 × \times × 1,000,000,000 × \times × 4 byte = 28,000,000,000 / 1024 / 1024 / 1024 = 26.077 GByte |
16-bit precision | 7 × \times × 1,000,000,000 × \times × 2 byte = 14,000,000,000 / 1024 / 1024 / 1024 = 13.038516 GByte |
8-bit precision | 7 × \times × 1,000,000,000 × \times × 1 byte = 7,000,000,000 / 1024 / 1024 / 1024 = 6.519258 GByte |
…
thousand [ˈθaʊznd]
- the number 1,000 (千)
- a large number
million ['mɪljən]
- the number 1,000,000 (百万)
- a large number
billion ['bɪljən]
- the number 1,000,000,000 (十亿)
- a very large number
trillion ['trɪljən]
- the number 1,000,000,000,000 (万亿)
2. Hugging Face
2.1. GPT-Neo 2.7B
https://huggingface.co/EleutherAI/gpt-neo-2.7B/blob/main/README.md
https://huggingface.co/EleutherAI/gpt-neo-1.3B/blob/main/README.md
GPT-Neo refers to the class of models, while 2.7B represents the number of parameters of this particular pre-trained model.
GPT-Neo refers to the class of models, while 1.3B represents the number of parameters of this particular pre-trained model.
2.2. Massive Text Embedding Benchmark (MTEB) Leaderboard
https://huggingface.co/spaces/mteb/leaderboard
Massive Text Embedding Benchmark
https://huggingface.co/mteb
Model Size (million parameters, in number of parameters)
Memory Usage (GB, fp32)
2.3. Open LLM Leaderboard
https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard
the number of parameters (B)
#Params (B)
2.4. LLM-Perf Leaderboard
https://huggingface.co/spaces/optimum/llm-perf-leaderboard
Memory (MB)
Params (B)
2.5. Open Ko-LLM Leaderboard
https://huggingface.co/spaces/upstage/open-ko-llm-leaderboard
#Params (B)
References
[1] Yongqiang Cheng, https://yongqiang.blog.csdn.net/
[2] The Big Benchmarks Collection, https://huggingface.co/collections/open-llm-leaderboard/the-big-benchmarks-collection-64faca6335a7fc7d4ffe974a