Granite 34B Code Instruct 8K
Granite 34B Code Instruct 8K is a May 2024 code model fine-tuned from the corresponding base. It remains downloadable, and IBM lists the 34B code-instruct model for dedicated watsonx deployment.
Release details
The code-commit instruction data covers 92 programming languages selected from the base model's 116-language pretraining coverage. Training also includes mathematical, code-assistant and language-instruction examples. The repository name omits 8K, but its card and configuration identify that context.Granite 34B Code Instruct 8K model card
Practical use
Practical guidance. Retain this checkpoint for workloads that already rely on its code-specific behavior. Compare both accuracy and serving memory with newer Granite releases before migrating. A large dense code model can be unnecessary for a narrow task that a smaller current model handles reliably.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | |
| Access | API, Downloadable, Provider product |
| Model identifiers | ibm-granite/granite-34b-code-instruct |
| Context window | 8,192 tokens Published checkpoint context; hosted service limits can differ. Granite 34B Code Instruct 8K model card |
| License | Apache 2.0Granite 34B Code Instruct 8K model card |
| Parameters | 34BGranite 34B Code Instruct 8K model card |
| Hosted access | watsonx: deploy on demand.IBM watsonx.ai foundation model library |
Related models
- Granite 8B Code Instruct 128K
IBM ยท Granite Code
A downloadable 8B code-specialist model fine-tuned from Granite's long-context base in July 2024.
Launch date Downloadable