Llama 3 70B Instruct, developed by Meta, features a context window of 8000 tokens. The model was released on April 18, 2024, and achieved a score of 82.0 in the MMLU benchmark under a 5-shot scenario.
Overview
Llama 3 70B Instruct
|
|
|---|---|
|
Provider
The entity that provides this model.
|
Meta
|
|
Input Context Window
The number of tokens supported by the input context window.
|
8,000
tokens
|
|
Maximum Output Tokens
The number of tokens that can be generated by the model in a single request.
|
2,048
tokens
|
|
Open Source
Whether the model's code is available for public use.
|
Yes
|
|
Release Date
When the model was first released.
|
2024-04-18
|
|
Knowledge Cut-off Date
When the model's knowledge was last updated.
|
|
|
API Providers
The providers that offer this model. (This is not an exhaustive list.)
|
|
|
Empirical Throughput
The number of tokens the model can generate per second.
|
|
Pricing
|
Llama 3 70B Instruct
|
|
|---|---|
|
Input
Cost of input data provided to the model.
|
Pricing not available.
|
|
Output
Cost of output tokens generated by the model.
|
Pricing not available.
|
Benchmarks
|
Llama 3 70B Instruct
|
|
|---|---|
|
MMLU
Evaluating LLM knowledge acquisition in zero-shot and few-shot settings.
|
82.0
(5-shot)
|
|
MMMU
A wide ranging multi-discipline and multimodal benchmark.
|
Benchmark not available.
|
|
HellaSwag
A challenging sentence completion benchmark.
|
Benchmark not available.
|
|
HumanEval
A benchmark to measure functional correctness for synthesizing programs from docstrings.
|
Benchmark not available.
|
|
MATH
Benchmark performance on Math problems ranging across 5 levels of difficulty and 7 sub-disciplines.
|
51.0
(0-shot)
|
Measure & Improve LLM
Product Performance.
Get Started
© 2023-2024 Context