Falcon-40b-instruct

A large SOTA model of its time which you can safely skip now.

Model

Falcon-40b-instruct

Details

Developer

UAE's Technology Innovation Institute (TII)

License

Apache-2.0

Model parameters

40B

Pretraining tokens

Release date

May 2023

Supported context length

Price for prompt tokens*

$0.75/Million tokens

Price for response tokens*

$0.75/Million tokens

*Note: Data based on 11/14/2023

Model Performance Across Task-Types

Here's how Falcon-40b-instruct performed across all three task types

Metric

ChainPoll Score

QA without RAG

0.59

QA with RAG

0.60

Long form text generation

0.70

Model Info Across Task-Types

Digging deeper, here’s a look how Falcon-40b-instruct performed across specific datasets

Tasks	Insights	Dataset Name	Dataset Performance
QA without RAG	The model does not perform well which show bias and errors in factual knowledge.	Truthful QA	0.46
QA without RAG		Trivia QA	0.72
QA with RAG	The model performs poorly on this which demonstrates weak reasoning and comprehension skills. It scores very less on DROP compared to other dataset which is a sign of bad mathematical skills.	MS Marco	0.78
		Hotpot QA	0.51
		Drop	0.4
		Narrative QA	0.72
Long form text generation	The model performs near satisfactory which shows ability to generate long text without factual errors.	Open Assistant	0.7

💰 Cost insights

The model scores low across all the tasks. It is 4x cheaper compared to GPT3.5 and 2x cheaper compared to Llama 70b variant. We suggest using Zephyr-7b-beta instead of this.

LLMHALLUCINATIONINDEXLLM