summaryrefslogtreecommitdiff
path: root/gpt2/model.go
Commit message (Collapse)AuthorAgeFilesLines
* Use platform specific defaults for CUDA device IDJonas Knobloch2026-04-231-1/+13
|
* Destroy session on model teardownJonas Knobloch2026-04-101-11/+13
|
* Refactor model optionsJonas Knobloch2026-04-081-11/+7
| | | | * Add options struct
* Refactor model configJonas Knobloch2026-04-081-6/+6
|
* Batch scoringJonas Knobloch2026-04-071-2/+4
|
* Rename log probs outputJonas Knobloch2026-04-071-2/+2
|
* Handle environment initialization on package levelJonas Knobloch2026-04-061-9/+1
|
* Require cache when generating new tokensJonas Knobloch2026-04-061-0/+4
|
* Add parameter to control logits output bindingJonas Knobloch2026-04-061-5/+7
|
* Access outputs by nameJonas Knobloch2026-04-061-9/+3
|
* Bind either logits or log probsJonas Knobloch2026-04-041-1/+1
|
* Add score methodJonas Knobloch2026-04-041-10/+42
|
* Add config structJonas Knobloch2026-04-041-14/+8
|
* Add allocator abstractionJonas Knobloch2026-04-041-177/+32
|
* Remove llm importJonas Knobloch2026-03-171-2/+0
|
* Use vanity import pathJonas Knobloch2026-03-161-2/+2
|
* Update module pathsJonas Knobloch2026-02-281-1/+2
|
* Verify sequence lengthJonas Knobloch2026-02-161-4/+9
|
* Allow controlling IntraOpNumThreadsJonas Knobloch2026-02-111-6/+33
|
* Use seen tokens to warm up cacheJonas Knobloch2025-11-261-47/+77
|
* Use dynamic sessionJonas Knobloch2025-11-241-62/+123
|
* Add perplexity evaluationJonas Knobloch2025-11-181-13/+26
|
* Scaffold byte-pair correctionJonas Knobloch2025-11-171-0/+264