Falcon 7B Instruct
Falcon 7B Instruct is an early Falcon chat and instruction model based on Falcon-7B. Its Apache 2.0 weights remain published despite newer Falcon generations.
Training and release background
The base model was trained on 1.5 trillion tokens. Instruction tuning used 250 million tokens drawn from Baize, GPT4All, GPTeacher and a small portion of RefinedWeb. The architecture uses multiquery attention. The card describes mainly English training with limited wider language generalization.Falcon 7B Instruct model card
Use the original model for a specific reason
Practical guidance. This checkpoint is useful when an existing application or experiment depends on it. For a new application, compare current small models using the same prompts. Historical leaderboard claims should not substitute for a present-day task comparison.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | tiiuae/falcon-7b-instruct |
| License | Apache 2.0Falcon 7B Instruct model card |
| Release status | Earlier release; published weight access remains availableFalcon 7B Instruct model card |
Related models
A 7B transformer language model for English, French, Spanish and Portuguese with a 32K context.
Launch date December 2024DownloadableFalcon 40B Instruct is an early 40B chat model adapted from Falcon-40B.
Launch date not confirmedDownloadable